“as
I ‘73-) Iy 5‘55
A n.-

   

 

. in 4.3.4

      
  
  
 
    

'3’} “.

. II. “.3151!“

   
 

x}I'
.I':,'ILI

II.
' 'Iu-z-x .~

_I.I"| . . . I
II‘I‘IIIII . F1I(V‘H‘\III.I‘

II" In“ I:- '.
. II. 'Ilfi'. I???" .‘:III

4'an ’. If‘

  
      
  
     
   
 

     
      
 

          

  

 
    

93:34“ I; 5*...

’ .>

.~r $4..

‘r; $-

. -. “£5?

_ , -

. 3.3“? ”mm" L *3
II

. r 1.
Z “11:
.Q .

 

I 5:
VIII 1.x 2 ”‘fi’

    
   

 
  
   
  
    

V5332. " ""9“ ' "
"I M. 4”
\ II'SLNy‘ $3: 1444‘”;-

    

  
 
  
  
 

  
  

   
 

 

  

... IEIII .
‘ I

1| U 4 ‘K .I "1‘9. 'I‘I".':.-:II 11:: 'II.‘ ’I: .tIau.:‘I{‘n 5.2:” ‘gl 0 'IV
. III..III-I‘-I.:II II IIIJ‘I‘I‘IIIIJIIr-III-I ‘.’(‘:'- \u ‘I. AM VI!:‘ of :fl' . 2’ N‘fic
;.;. ‘-:I§'.‘::-.v1.‘,.:1": 2::p-II:1;‘I'I_;'.'5 su::rfp'...x;w R"
'I I .I ;I m" I 0.1: "IN“. . 'I '.

.' . . \‘I, o ‘5) , \N b

.; ,, I (.4 {4“ 4'1“" "W" My. " 93..

'4 III: ': | “3“: 11:31; I I . ,

   
  
       
 
 
  

 

‘ 2E; ‘ :‘VJII
“(a ii?

“1‘39" “$35? .:

3'59

  

K‘s

....

 
 

  
  
     
 
 

“95.33
'\'

 

    
    
     
  
 
  

I x o' ',I.‘“
‘I.III""I\I\‘..V‘.“‘ 3‘.
. ,I .

I .,
II;\“. I‘h"‘ “,IIIIIIIIII

.Iv‘I'Iu I«\‘
‘ 'I

          

1 '.1;‘I',I...'.;="'.
‘. ,- I.- . L

I: 3;” 4M

II'ILTt'Kr’M

:4

‘.' L'- .. WWI?

I‘ L 1",; '|l‘&."\. - LLI'
".I. I‘ll I.—1 l-I'Wt.

   
 
 

This is to certify that the

thesis entitled

Investigation and Development of Electrothermal

Atomization for Atomic Absorption Spectroscopy

presented by

Eugene Harlan Pals

has been accepted towards fulfillment
of the requirements for

Ph . D . degree in Chemis try

W924

Major professor

 

Date May 26, 1978

 

0-7639

 

 

I .TBRARY

Michigan Star
University

 
 

INVESTIGATION AND DEVELOPMENT OF ELECTROTHERNAL

ATOMIZATION FOR ATOMIC ABSORPTION SPECTROSCOPY

By

Eugene Harlan Pals

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements

for the degree of

DOCTOR OF PHILOSOPHY

Department of Chemistry

1978

ABSTRACT

INVESTIGATION AND DEVELOPMENT OF ELECTROTHERMAL

ATOMIZATIONINHXATOMIC ABSORPTION SPECTROSCOPY

By

Eugene Harlan Pals

A computer-controlled electrothermal atomizer atomic
absorption (EAAA) instrument has been developed which
utilizes sophisticated software and hardware to automate
all of the individual machine functions involved in a
typiCal EAAA analysis. The computerized system contains
a state of the art atomizer heating regulation device,
an automatic sample dispenser, and a computer-controlled
atomization cell positioning device.

The atomizer heating regulation device allows the
operator to select any of four types of feedback regula-
tion of atomizer heating: current, voltage, power, or
radiation. The variable volume automatic sample dis-
penser is capable of rapidly and precisely delivering
1.0 to 5.0 pi sample solution to the atomizer surface,
and this allows total EAAA measurement precisions of l
to 3%. The automatic vapor cell positioner allows the

precise displacement in two dimensions of the atomizer

Eugene Harlan Pals

relative to the optical axis of the instrument over a
one inch range in each dimension for spatial studies of
the atomic vapor plume.

The system software also provides for automatic execu-
tion of the simplex algorithm to optimize a chosen instru-
ment response such as the magnitude of the absorbance
signal. Any 2 to U of the 13 experimental parameters
under computer control may be varied by this algorithm
to achieve the optimum desired response. The computer-
controlled EAAA instrument with its associated software
has been shown to be a powerful and versatile tool for
studies of fundamental EAAA processes. The simplex op-
timization technique is also shown to be a valuable tool

in optimizing the performance of the EAAA instrument.

ACKNOWLEDGMENTS

During the time I was in graduate school many people
have been kind to me when they could have been indifferent,
have helped me when they had more important work of their
own to do, and have somehow made the trials and tribula-
tions of research not only bearable, but at times even
enjoyable. There is no possible way to thank everyone
who deserves credit, but a few people do deserve special
mention.

First, my research preceptor, Professor Stanley R.
Crouch, who managed to guide my research in the proper
direction without stiffling me when I wanted to try some-
thing outside the master plan. I wish to thank Dr. Chris
Enke for serving as my second reader, Dr. Tom Atkinson
for rendering invaluable assistance in the production of
many of the figures presented in this work, and also Dr.
Andrew Timnick because he was such a pleasure to teach
for and a good friend as well.

Also I must thank Dr. Eric Johnson for helpful dis-
cussions in the areas of electronics and computer program-
ming and Marty Joseph for the use of his simplex routine
and his help in applying it to the EAAA instrument.

Jim Gano helped with some of the final stages<xfthis
work and also deserves acknowledgment. Special thanks

go to my coworker, Dr. Dave Baxter, who is one of the

11

most gifted researchers I have ever met and without whom
much of this work would have been impossible.

I would also like to express my gratitude to Coach
Ron Schippers who taught me many things both on and off
the football field and showed an interest in my academic
career before, during and after my playing days at Central
College.

To Gary, Leslie, larty, Betty, Dan, Mel, Roy, Sandy,
Jim, Clay, Rytis, Kathy, and all the other people at MSU
that I have grown to know and love, thank you, my life
would have been something less had I not had the privilege
of knowing you.

Finally, and most important, I want to thank my mother,
my father, my sister Eunice, and especially my sister
Lois. Without them this work might have been possible,
but such a success, unshared with them and unappreciated

by them, would certainly have been an empty victory.

iii

TABLE OF CONTENTS

Chapter Page
LIST OF TABLES. . . . . . . . . . . . . . . . . . viii
LIST OF FIGURES . . . . . . . . . . . . . . . . . x
INTRODUCTION. . . . . . . . . . . . . . . . . . . 1
CHAPTER 1. Historical. . . . . . . . . . . . . . 6
A. Electrothermal Atomization Atomic
Absorption Spectroscopy . . . . . . . . . 6
B. The Simplex Method of Optimiza-
tion. . . . . . . . . . . . . . . . . . . 7
CHAPTER 2. Instrumentation . . . . . . . . . . . 16

A. General Description . . . . . . . . . . . 16

B. Atomization Vapor Cell. . . . . . . . . . 18
C. Optics. . . . . . . . . . . . . . . . . . 21
D Atomizer Power Controller . . . . . . . . 2A
1. Introduction and General
Overview. . . . . . . . . . . . . . . 2A
2. The Control Operation
Amplifier . . . . . . . . . . . . . . 29
3. Current, Voltage, and Power
Regulation Circuitry. . . . . . . . . 31
h. Radiation Regulation Circuitry. . . . 33

Hydrogen Diffusion Flame
Switching Circuit . . . . . . . . . . 35

6. Perspectives. . . . . . . . . . . . . 37
E. Positioner. . . . . . . . . . . . . . . . 39

1. Introduction and General
Overview. . . . . . . . . . . . . . . 39
Physical Structure. . . . . . . . . . 39

Remote and Local Operation. . . . . . “2

iv

Chapter Page

F. General Interface . . . . . . . . . . . . AS

CHAPTER 3. Programming the Computerized
Electrothermal Atomizer Atomic

Absorption Instrument . . . . . . . . “8
A. Introduction. . . . . . . . . . . . . . . M8
B. Choice of a Programming Language. . . . . 51
C. Fortran II Program Capabilities . . . . . 53
D. PDP 8/e Fortran IV System
Software. . . . . . . . . . . . . . . . 55
E. Program Structure for the EAAA
Instrument. . . . . . . . . . . . . . . . 61
F. INPUT Subroutine. . . . . . . . . . . . . 66
G. FLOW Subroutine . . . . . . . . . . . . . 72
H. DATA-CALC and PLOT Subroutines. . . . . . 7A
I. Assembly Language Routines. . . . . . . . 80

CHAPTER A. Construction and Testing of
the Variable Volume Automatic

Sample Dispenser. . . . . . . . . . . 86

A. Introduction. . . . . . . . . . . . . . . 86
B. Construction and Operation of

the ASD o o o o o o o o o o o o o o o o o 89

1. Mechanical Components . . . . . . . . 89

2. Electronic Hardware . . . . . . . . . 96

3. Software. . . . . . . . . . . . . . . 99
C. Evaluation of Performance . . . . . . . . 102
D. Conclusions . . . . . . . . . . . . . . . 109

CHAPTER 5. Simplex Optimization Studies on
the EAAA Instrument . . . . . . . . . 110

A. Introduction. . . . . . . . . . . . . . . lll

Chapter Page

Example . . . . . . . . . . . . . . . 112

Comparison of the Simplex
Method with Other Optimiza-

tion Techniques . . . . . . . . . . . 113
3. The Nelder and Mead Simplex
Algorithm . . . . . . . . 116
B. Simplex Software. . . . . . . . . . . . . 119
l. Initiation of a Simplex
Optimization. . . . . . . . . . . . . 120
2c Run-Time Options 0 o o o o o o o o o o 122

C. Tests of the Simplex Algorithm
on a Known Response Surface . . . . . . . 123

1. The Cadmium Integrated Ab-

sorbance Response Surface . . . . . . 12A
2. Methods of Monitoring the
Progress of the Simplex . . . . . . . 126

3. Comparison of Optimizations
with Different Starting

Simplexes . . . . . . . . . . 132

A. Optimization of the Integrated
Absorbance of Manganese . . . . . . . 136
D. Four Parameter Optimizations. . . . . . . 139

1. Four Parameter Optimization
of the Integrated Absorbance

of Cadmium. . . . . . . . . . . . . . 139
2. Simplex Optimization with a
Complex Response Function . . . . . . 1&6

E. Optimizations of the Measurement
Precision . . . . . . . . . . . . . . . . 151

1. Direct Optimization of the
Measurement Precision . . . . . . . . 151

2. The Determination of the
Optimum Integration Time

Window. . . . . . . . . . . . . . . . ISM
F. Conclusions and Perspectives. . . . . . . 160
REFERENCES. . . . . . . . . . . . . . . . . . . . 16”

vi

Chapter Page

APPENDIX A: Selected Program Listings. . . . . . 169
APPENDIX B: Simple Hardware Control

Approach for Sequencing
Chemical Instruments . . . . . . . . 209

vii

LIST OF TABLES

Table Page

1 Simplex Optimizations of Chemical

Processes and Instrumentation . . . . . . 12
2 Run-time Front Panel Switch

Options . . . . . . . . . . . . . . . . . 85
3 Bit Assignments for the Status

Register and Job Word Register. . . . . . 100
A Measurement Reproducibilities of

Various Delivery Modes. . . . . . . . . . 103
5 Long Term Stability . . . . . . . . . . . 105
6 Experimental Parameters that may

be Optimized with the Simplex

Routine . . . . . . . . . . . . . . . . . 121
7 Results of Optimizations of the

Horizontal and Vertical Positions

for Maximum Cadmium Integrated

Absorbances . . . . . . . . . . . . . . . 133
8 Results of Optimizations of the

Horizontal and Vertical Positions

for Maximum Integrated Absorbance

of Manganese. . . . . . . . . . . . . . . 137
9 Results of Optimizations of Delay

Time, Horizontal Position, and Amount

of Sample for Maximum Cadmium Inte-

grated Absorbance . . . . . . . . . . . . 141

viii

Tables

10

ll

12

Page

Results of Optimizations of the

Gas Flow Rate, Vertical Position,

and Atomization Power for Maximum
Signal-to-Noise Ratio . . . . . . . . . 153
Results of Optimizations of Integra-

tion Start and Stop Times . . . . . . . 157
Evaluation of the Applicability

of the Simplex Determined Integra-

tion Time Window to Other Data

Sets. . . . . . . . . . . . . . . . . . 159

ix

Figure

10

ll

12

LIST OF FIGURES

Block Diagram of the Computer Con-
trolled EAAA Instrument . . . . .
The Atomization Vapor Cell.

Optical Configurations for the

EAAA Instrument . . . . . . . .
Overview of Atomizer Power Regu-
lation Circuitry.

Control 0A Circuitry. . . . . . .
Current, Voltage, and Power Regula-
tion Circuitry. . . . . .

Radiation Regulation Circuitry.
Hydrogen Diffusion Flame Switching
Circuit . . .

Illustration of the Positioner

and the Atomization Vapor Cell.
Diagram of the Overlay Structure

of the GBMDS2 Program . . . . . . . .
Example of a User Dialog with
Subroutine INPUT. . . . . . . . . .
Simplified Flow Chart for Sub-

routine INPUT . . . . . . . . . . .

Page

17
l9

23

27
30

32
3A

36

NO

62

67

71

Figure

13

1A

15

16

17
18

19
2O
21

22

23

2A

25

Page

Flow Diagram for the DATA-CALC
Subroutine. . . . . . . . . . . . . . . . 73
EAAA Instrument with Automatic
Sample Dispenser. . . . . . . . . . . . . 88

Sample Amount Metering and

Delivery System . . . . . . . . . . . . . 90
Transport Mechanism . . . . . . . . . . . 92
Sample Turntable Module . . . . . . . . . 93

Block Diagram of the ASD's

Electronic Hardware . . . . . . . . . . . 97
Long Term Stability . . . . . . . . . . . 106
Volume Delivery Linearity . . . . . . . . 108
A Typical Response Surface. . . . . . . . 11b
The Possible Moves of the Modified

Simplex Algorithm . . . . . . . . . . . . 117
Cadmium Integrated Absorbance

Response Surface with Assorted

Starting Simplexes. . . . . . . . . . . . 125
The Moves of Simplex Optimiza-

tion G. . . . . . . . . . . . . . . . . . 127
Progress Plot of the Cadmium

Integrated Absorbance Response

in a Two Parameter Optimization . . . . . 128

xi

Figure Page

26 Progress Plots of Vertical Position

and Horizontal Position in a Two

Parameter Optimization of Cadmium

Integrated Absorbance . . . . . . . . . . 129
27 Progress Plot of the Cadmium Inte-

grated Absorbance Response in a Four

Parameter Optimization. . . . . . . . . . 1A3
28 Progress Plots of the Vertical

Position and Horizontal Position

in a Four Parameter Optimization

of the Cadmium Integrated Absor-

bance . . . . . . . . . . . . . . . . . . 1AA
29 Progress Plot of the Sample Amount

and the Delay Time in a Four

Parameter Optimization of the

Cadmium Integrated Absorbance . . . . . . 195
30 A Typical Absorbance Peak for

2 pl of 1 ppm Cadmium Solution. . . . . . 156
B1 Basic Flow Chart Units for Trans—

fer Conditions and Transfer Functions.

N: State Number . . . . . . . . . . . . . 211
B2 Sequencer Flow Chart for Bath

Thermostat Example. . . . . . . . . . . . 211
B3 General Schematic Diagram of

Controller. . . . . . . . . . . . . . . . 212
xii

Figure Page

BA Sequencer Flow Chart for Automatic

Fraction Collector. . . . . . . . . . . . 212
B5 Schematic Diagram of Fraction Col-

lector Controller . . . . . . . . . . . . 213
B6 Sequencer Flow Chart for Filament

Positioner. . . . . . . . . . . . . . . . 21“
B7 Schematic Diagram of Filament

Positioning Controller. . . . . . . . . . 215

xiii

INTRODUCTION

Since all fields of science are interrelated, develop-
ments in one area often profoundly effect other areas of
science. Nowhere is this more clearly demonstrated than
in the relationship of electronics to analytical chemistry.
The electronic developments of the vacuum tube, photo-
multiplier tube, transistors, and integrated circuits have,
in each case, permitted the manufacture of a new generation
of analytical instrumentation and opened up new research
areas. Most recently, large scale integration and micro-
processor developments have been revolutionizing analytical
instrumentation by integrating into the instrument computa-
tional, sequencing, and decision making tasks previously
relegated to the operator.

Although the impact of computerization and automation
on analytical chemistry has been profound, the present
developments are Just the tip of the iceberg. Already
many research laboratories are equipped with computers
which are much more powerful than any computer in exist-
ence 30 years ago. If the present pace of micro-minaturi-
zation of electronic circuitry continues it is not only
conceivable, but indeed probable, that the research sci-

entist in therufiztoo distant future will have at his

disposal a laboratory computer as powerful as any of the
largest computers now Operating. The potential uses of
such a computer system are mind-boggeling. And, the
possibility that systems will be designed that are so
gargantuan, intricate, and complex that it will be im-
possible for anyone but an expert to use them is very
great. How could this new computer power be best used?
Or, to rephrase the question in a more leading form -
what should be the objectives of present research in this
area?

One way to answer that question is to consider the
past and present roles of the computer in the context of
the classical scientific method (observation, hypothesis,
experiment, theory or new hypothesis) and then extrapo-
late along the present path to the most reasonable and
logical conclusion. Computers were first used in the
experimental step to calculate results from raw data.
Later, on-line computers were used in the actual execu-
tion and data gathering, and now it is quite common
for a computer to control both the data acquisition
and data analysis.

Computer curve-fitting programs are allowing the
computer to enter the hypothesis and theory steps in a
limited manner, but the most promising software develop-
ments in this respect are hueristic, "artifical intelli-

gence" software and closed loop optimization software,

which can play a very active role in the theory and new
hypothesis steps. Unfortunately, the recent trend in
computer usage has been directed more toward the wide-
spread application of existing techniques to previously
non-automated experiments, rather than toward developing
software which moves the on-line computer into areas of
research other than Just the experimental step. Because
of the increased capabilities of the new generation of
computers for routine on-line instrument interfacing, it
is now possible to consider automating the experimental
research process rather than Just automating an instru-
ment. This distinction bears further elucidation. In-
strument automation entails development of the hardware
and software necessary to allow an instrument to gather
data without any operator intervention except to start
the experiment. Automation of research involves the

use of the computer not only to assist in the execution
of an experiment and the analysis of the results, but
also to assist in the process of either establishing a
theory from the results or generating a new and reason-
able hypothesis to test. Total automation of research
may never be possible since it is difficult to conceive
how a computer could be programed to have the intuition
and serendipity of a good researcher, but the computer
could be a very valuable aid in many areas where it is

not now being used.

Electrothermal atomization has become a very popular
technique in environmental and clinical laboratories
in particular because of the very small sample sizes re-
quired and the extreme sensitivity of the technique for
many metallic elements important in these fields. Al-
though many electrothermal atomizer atomic absorption
instruments have been interfaced to a dedicated computer
or microcomputer, few have used the computer to its full—
est capabilities. Many have not automated the sample
introduction step so that the operator must constantly
attend the instrument even during routine analysis.

The purpose of this work was to design and construct
a highly automated electrothermal atomic absorption I
instrument that takes full advantage of the power of
the system's computer. The automation of this technique
involved the construction of several pieces of special
purpose equipment. A positioning device was constructed
to translate in two dimensions (with respect to the
optical axis of the instrument) an atomic vapor cell con-
taining the atomizer. A computer-controlled automatic
sample dispenser, capable of accurately and reproducibly
depositing microliter samples onto the atomizer, was
constructed and has been extensively characterized in
this work. The computer software necessary to drive the
instrument is described. The computer-controlled instru-

ment was also programmed to execute the simplex algorithm,

which is a sophisticated, time-efficient optimization
technique. The results of some of these optimizations

as well as the characteristics of the algorithm as applied
to electrothermal atomization atomic spectroscopy are

also discussed. The maJor significance of this work is
that it demonstrates that once an instrument has been
automated, an additional effort to expand the role of

the system past the normal data acquisition and calculation
assignments into sophisticated self—optimization tasks
greatly enhances the utility of the instrument not only

as a research tool, but also as a tool for practical

analyses.

CHAPTER 1

HISTORICAL

A. ELECTROTHERMAL ATOMIZATION ATOMIC ABSORPTION SPEC—

TROSCOPY

Atomic Absorption (AA) spectroscopy was introduced
as an analytical tool by Walsh (1,2) in 1953 and Alkemade
and Milatz (3,A) in 1955. The absorption phenomena had
been observed in 1802 by Walston (5), but the only appli-
cation of this technique outside astronomy before Walsh
was the determination of mecury vapor in air (6). The
near coincidental development by independent researchers
of AA spectroscopy as an analytical measurement tech-
nique a century and a half after the discovery of the
phenomena was due primarily to the development of good
photoelectric detectors and the development of hollow
cathode discharge tubes at about this time.

Most early work in AA spectroscpy utilized a flame
as the atomic vapor cell. In 1959, L'vov (7) developed
a resistively heated graphite furnace which could be used
as an atomic reservoir. This work attracted little atten—
tion until the late sixties when Woodriff (8-1“) and Mass-

mann (15,16) introduced their own versions of graphite

furnaces. The technique rapidly became popular after

1970 when Perkin-Elmer Corporation produced the first

commercial AA instrument equipped with a graphite fur-
nace atomizer (17-19).

Since that time, many other companies have followed
suit and introduced their own versions of EAAA instru-
ments. One very significant variation on the design of
electrothermal atomizers was the introduction of fila-
ment atomizers by West (20-23). Many types of base
materials and physical structures have been used as
filament atomizers. These include the graphite braid
atomizer, which was introduced by Crouch and Montaser
(2A-27), and characterized by Baxter (28). This atomizer
was used exclusively in this work. The application of
EAAA and other non-flame techniques such as chemical
vaporization and sample boat techniques to real analyses
has been exhaustively reviewed elsewhere (28-30) and will

not be reiterated here.

B. THE SIMPLEX METHOD OF OPTIMIZATION

 

In analytical chemistry, as well as daily life, one
is constantly confronted with the task of optimizing
the performance of some system, process, or device.
Normally, this optimization is attempted by iteratively
optimizing the response of each individual variable. The

failure of this single factor at a time optimization

process was first pointed out by Box (31). Box and Wil-
son (32) presented the first organized discussion of re-
sponse surfaces and hill-climbing theory in 1955. Later
Box introduced the EVolutionary OPerations (EVPO) tech-
nique as a method of finding the response surface optimum.
The occasional failure of this approach to find the true
optimum was pointed out by Lowe (3A).

In 1962 Spendley, Hext, and Hinsworth (35) introduced
the simplex method of optimization. The simplex is a
geometric figure of n+1 dimensions, where n is the number
of factors (variables) to be optimized. This figure is
moved across the response surface toward the optimum by
the simplex paradigm. The response is first evaluated
at each of the vertices and the worst (lowest response)
vertex is reflected through the centroid of the opposite
face of the simplex. The response at this new vertex is
then evaluated, the new worst vertex determined and re-
flected to create a new simplex, and so on.

This paradigm has been modified a number of times,
most notably by Nelder and Mead (36) and Denton (37).
The Nelder and Mead paradigm is called the modified
simplex. It allows the worst vertex to be displaced to
any of four points on the line defined by the worst ver-
tex. This allows the simplex to be expanded or contracted
so that it can approach and converge on the response sur-

face optimum rapidly. A more detailed description of the

modified simplex and optimization terminology is given
in Chapter 5.

The Denton modification was named the super modified
simplex. This modification allows the new simplex vertex
to be located anywhere along the line defined by the worst
vertex and the centroid of the opposite face of the sim-
plex. The displacement distance is determined by an
evaluation of the shape of the response surface in the
region of the simplex. Three points are used in this
evaluation, the response at the old worst vertex, at the
centroid of the opposite face, and at the point obtained
by reflecting the worst vertex through the centroid of
tHKBOPPOSiteface. The concave or convex nature of the
response surface, as perceived by this three point evalua-
tion, indicates whether the next move should be an expan-
sion or a contraction. The distance of the move is
determined by the overall slope of the response surface
again as determined from these three points. This ap-
proach theoretically allows the simplex to move more
rapidly than the modified simplex through plateau regions
and also to converge on the optimum without seriously
overshooting it.

The modified simplex algorithm has been used in chem-
istry for regression analysis in enzyme competitive in-
hibition (38), analysis of multicomponent spectra (39),

dead-time estimation in chromatography (A0),

10

chromatographic absorption isotherms (Al), and pattern
recognition (A2,A3). The considerations involved in

the applicationCfiithe simplex algorithm to numerical anal-
ysis are somewhat different from those involved when the
technique is applied to the optimization of dynamic sys—
tems. Thus, the numerical uses of the algorithm need not
be discussed in detail here.

Lowe (3U), Long (AA), and Deming (AS-A7) have all
discussed the difficulties encountered when the simplex
optimization technique is applied to dynamic chemical
systems. Long (AA) gives a particularly good discussion
on how to choose the factors to be optimized, how to
select the quantity to be optimized, and how to select
the size and location of the starting simplex. Long also
points out that noise on the response surface can cause
moves of the simplex away from the true optimum, and that
the simplex paradigm should be self-correcting for these
false moves unless the noise is excessive and causes
premature contraction of the simplex.

King (A8) castigated some of the early users of the
simplex optimization technique for misuses. One common
error is the attempt to optimize coupled variables such
as reagent concentration and reagent volume. These
factors contribute multiplicatively to the more funda-
mental factor, which is the amount of reagent in the

final solution. When two coupled variables are chosen

11

for optimization, the response surface contains a station—
ary ridge which defines the optimum combinations of these
factors. Two other common errors are the incorrect defi-
nition of the response and the poor choice of factors to
be optimized. In the example of the optimization of a
spectrophotometric determination, the response is some-
times claimed to be sensitivity when in actuality the
response optimized is the absorbance. The two may be
assumed related with some validity only if one of the fac-
tors chosen for the optimization is not in some way relat-
ed to the sample amount. If sample amount is chosen as
one of the factors then either the response should be
defined as the absorbance, or if the response is defined
as sensitivity then the absorbance must be normalized

for the amount of sample present. Several researchers
have made this type of error (49-51).

Some applications of the simplex technique to chem-
istry are listed in Table 1. Several of these are of
particular significance to this work and warrant more
detailed discussion. Rippetoe, Johnson, and Vickers
(57) performedtflueOptimization of the signal-to-noise
ratio of calcium emission with respect to arc current,
slit width, slit height, and are optical path on a DC
plasma arc instrument. This appears to be the only direct
application of the simplex technique to optimizing the

precision of a measurement. Unfortunately no particulars

12

 

 

 

Table l. Simplex Optimizations of Chemical Processes
and Instrumentation.
System Response Optimized Ref.
p-rosanaline
detn. of S02 absorbance AA
AA detn. of Ca sensitivity 52
continuous flow precision,minimum carry 53
spectrometry over interaction,enzyme
activity
organic synthesis reaction yield 5A
Lieberman-Buchard sensitivity A7
detn. of cholestrol product stability
drug design biological effectiveness 55
drug design biological effectiveness 56
DC plasma arc Ca
emission signal-to-noise ratio 57
NMR spectrometer field homogeneity 58
pulsed hollow cathodes peak intensity 59
continuous flow
spectrometry sensitivity 60
anodic stripping
voltametry minimum analysis time 61
AE detn. of Ca signal intensity 62

 

 

13

are given by these authors concerning the actual method
of the determination of the signal-to-noise ratio or the
behavior of the simplex paradigm in this type of optimiza—
tion. Johnson, Mann, and Vickers (59) later reported

the optimizationcfl‘the peak intensity of a pulsed hollow
cathode lamp, since the direct optimization of the signal-
to-noise ratio was impossible due to the imprecision in-
volved in the determination of the noise level (63).
Parker, Morgan, and Deming (52) optimized the absorbance
of a calcium solution with respect to the fuel flow rate,
lamp current, burner height, and the volume of water in

a graduated cylinder located at the end of the lab bench
some distance from the spectrometer. The simplex con-
verged on values for the water volume and lamp current
even though they were later shown to have no effect on
the magnitude of calcium absorbance. These workers sug-
gest that the significance of factors should be tested
after the simplex optimization by initiating a multi-
parameter factorial design study in the region of the
optimum. Note that these workers optimized the calcium
absorbance for a fixed amount of analyte and assumed this
was related to the sensitivity and probably the precision
of the measurement. Other statistical search methods
have also been used to optimize the absorbance response
in AA analysis (6U-67) with results not dissimilar to

those obtained in the simplex optimization.

1“

One study listed in Table 1, that is of particular
interest because of its unique approach, is the optimiza-
tion of analysis time in anodic stripping voltammetry by
Thomas, Kryger, and Perone. The controlled factors for
this study were the plating time and scan rate. The op-
timization was carried out within the limits imposed by
a "performance criterion" which was a user-defined minimum
acceptable value for either the signal-to-noise ratio or
the signal-to-background ratio. Also, the computer was
used to calculate some preliminary constants from theo-
retical equations. These constants were necessary for the
evaluation of whether the simplex might be moving into
a region where the selected "performance criteria" would
not be satisfied. Unfortunately, although the work is
exemplary in a number of areas, the authors chose to use
their own terminology rather than the standardized ver-
nacular of response surfaces and optimization techniques.
The response optimized is actually a combination of the
analysis time and the chosen "performance criteria".

The correct term for these "performance criteria" is
constraint or boundary, since a violation of the con-
straint is signalled to the simplex algorithm as a penalty
to the delay time response. This work also involves the
unusual feature in that one of the parameters to be
optimized has only 11 discrete levels. The computer

was used to round off the simplex suggested value of this

15

parameter to the nearest of the discrete levels, and the
authors maintain that this caused no significant problems
if the initial simplex was large in this dimension.

In summary, the simplex technique has been widely used
and misused. Most optimizations have only contained two,
three, or occasionally four significant factors so that
the number of iterations required were within reason.

The response chosen to be optimized is often some response,
such as absorbance, which is then assumed to be related
to the response that would really be the most desirable
to optimize, such as the measurement sensitivity or pre-

cision.

CHAPTER 2

INSTRUMENTATION

A. GENERAL DESCRIPTION

 

A block diagram of the computer controlled Electro-
thermal Atomization Atomic Absorption (EAAA) instrument
is given in Figure l. The computer is a PDP 8/e equipped
with 16k of mainframe memory and an extended arithmetic
element. The computer peripherals used in this work
include: a Sykes 7000 dual floppy disc, a DEC RK05 cart-
ridge disc, a DEC LA—30 Decwriter used as a line printer,
and an ADDS Consul 980 CRT used as the system terminal.

Non-computer commercial equipment included: a GCA
McPherson EU-703—70 hollow cathode discharge tube (HCDT)
power supply, assorted Jarrel-Ash and Westinghouse hollow
cathode discharge tubes, a GCA McPherson EU—700/E mono-
chromator, a GCA McPherson model EU-701-30 photomultiplier
module with an RCA 1P28A or Hamamatsu R166 solar blind
photomultiplier (PMT), a model U27 Keithley current ampli-
fier, and a Tektronix model 50u3 oscilloscope.

The remainder of the instrument is composed of modules
built by my coworker, Dr. Dave Baxter,and myself. The

modules discussed in this chapter include the atomization

l6

l7

.PZm—zamkmg <<<m
BJJOKFZOU KMFDQEOU NI... no 23545 x0015; mmeE

 

 

 

- , - - >uuaam
5 «022850202 5522.4 Biaipowga Bo:—

-r r """"""""""""""""" J

 

 

 

 

 

 

    

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

mmE 5&8 $.38 mwmzuama
w_l >ow>w.o.u Tamas“: 3&2» 5.29:8“;
WcoomofioanT 1784 _2 o3
- BEES.
4452mm
Adz—2mm... «whats—cu «pry—Ea
Eu - m2... -

 

 

 

 

aura”

 

18

vapor cell, the atomizer power controller, the atomizer
vapor cell positioner, the general interface, and the
instrument optics. The automatic sample dispenser is

discussed in Chapter A.

B. ATOMIZATION VAPOR CELL

 

A simplified scale drawing of the atomizer vapor
cell is shown in Figure 2. The graphite braid atomizer
(Union Carbide Corp., Parma, OH) is held in the Jaws of
a brass holder at each end. When a screw in the upper
Jaw of the holder is loosened, the Jaw may be slid away
from the atomizer so the atomizer can be removed and re-
placed. The power connections are made to the bottom of
the lower Jaw near the outer end of the holder. It was
necessary to mount the atomizer holder on brass heat
sinks since prolonged heating of the atomizer would heat
both the holder and the material it was mounted on enough
to deform plexiglass. The brass heat sinks are elec-
trically isolated from the steel support structure by a
layer of plexiglass. The channel formed by the brass
heat sinks on two sides, and a plexiglass plate on the
other two sides, contains the sheath gas flow apparatus.
This apparatus was fashioned from close packed glass
capillaries glued together and was supported on a ridge
along the inside walls of the plexiglass box. The

laminarity of gas flow (28) from this type of apparatus

ATOMIZ ER
HOLDER

19

 

a QUARTZ
; ‘5‘”I’CELL

WALLS

GRAPHITE
BRAND
IHTMMZER

 

 

 

F$:52:'?:4:1$Wfl£‘?_{

. ATOMIZER
'~. ggl """““" HOLDER

      

 

 

GLASS
CAPlLLARY
ATOMIZER

._
'4

.->'+13<‘.~‘=;§E

lull“:

 
  

 

 

 

 

 

:;:;:;:;:;' 33:5: 6—

i Me AR PURGE GAS

,-
.....

 

lllll
------
-----
IIIIII
lllll
IIIIII
u-l

<—-HEIGHT ADJUST

 

‘ "‘.':‘:':':":::" -:<N:-:~ +- .........
+'-:::-:=:-:-:;.-‘-i‘~._«. ¢:.:._.- 3°" -" :35:- r'.,-: -:-:

Micah f':=:i:-:??3:35§:§:?:i:::3'-'c ........... 35:33:31 :73: ......

FIGURE 2: THE ATOMIZER VAPOR CELL

20

far exceeds that of any previous designs which had holes
drilled in a metal block. The position of this sheath
gas flow apparatus relative to the braid could be changed
by rotating a thumbwheel mounted at the lower end of the
support structure. The axle on which this thumbwheel is
mounted and the base plate of the sheath gas apparatus are
threaded so that rotation causes movement of the sheath
gas apparatus relative to the atomizer. To change the
orientation of the atomizer relative to the optical axis
of the instrument, the braid holders and associated parts
may be lifted from the support structure, the sheath gas
apparatus may then be removed by rotating the elevation
thumbwheel until the sheath gas apparatus is unfastened.
The brass base plate of the sheath gas apparatus is then
removed and mounted in an alternate position. The whole
system may then be reassembled with the atomizer orienta-
tion 90 degrees different from the original position.

Along with the novel glass capillary gas flow shaping
apparatus, a unique feature of this design is the capa-
bility to configure the system so that the atomizer may
be mounted parallel to the optical axis with the colli-
mated light beam directed at grazing incidence Just above
it. The advantages of parallel mounting include a reduc-
tion in the amount of atomizer continuum background radia-
tion viewed by the monochromator, a reduction in the

critical nature of the location of the sample on the

2l

atomizer, and a sensitivity enhancement due to the in-
crease in the absorption path length. The effects of
sample size should also be reduced when the atomizer is
mounted parallel to the optical axis, since sample spread-
ing will not cause any sample to be vaporized from a region
of the atomizer that is not under observation. To achieve
this parallel configuration it was necessary to design the
vapor cell walls from optical flat quartz plates (ESCO
Optics, Oak Ridge, NJ) and to design the braid holders
so that no part of them would occlude any source radiation.
An additional restriction on the design of the holders
was that as little as possible of their structure should
protrude from the walls of the vapor cell into the gas
flow. This is critical in maintaining the good laminar
gas flow generated by the glass capillary array.

A phototransistor (now shown in Figure 2), which is
used to measure the blackbody radiation of the atomizer,
is mounted on top of one of the plexiglass side walls at
about a “5 degree angle to the atomizer, and views the
atomizer through one of the quartz walls of the vapor

cell.

C. OPTICS

Atomic vapor mapping experiments with the EAAA instru-
ment require that the source optics yield a high intensity,

collimated, small diameter beam of radiant energy, whereas

22

the optimal optics for other experiments, such as detec-
tion limit studies, may be somewhat different. For this
reason the optics were designed so that they could be
arranged in several different configurations. The base
of the instrument is a GCA McPherson EU 703 optical rail,
and all components are fastened to it in some way. A
diagram of the EAAA instrument (simplified by omitting
the positioner) is shown in Figure 1M, Chapter A.

Three different lenses can be used. Two are 2"
diameter, 2" focal length quartz lenses with precision
lap finishes, the third is a l/2" diameter, 1/2" focal
length quartz lens. Each lens is mounted in a Teflon
ring and is fastened to a 1/2" aluminum shaft by a gimbal
mount so that height and tilt adjustments may be easily
made. One fixed position receptacle for the lens mount
is attached to the atomization cell positioner. Also on
this positioner is a receptacle mounted in a channelled
track on the positioner, so that this lens may be moved
over a 3" range along the optical axis. The third recep-
tacle is a free standing mount and may be placed anywhere
along the optical axis. The atomization cell positioner
itself is designed so that it can be easily moved to any
position along the optical axis.

The two most useful configurations are shown in Figure
3. Configuration A has a better collection efficiency

and light throughput than configuration B and is thus

23

.Pzwzamkm; <<<m MI... mom mZOEBGDDEZOU 4<U_._.n_0"m mmeE

 

 

 

 

 

 

 

 

 

 

wmahdmmn:
wuz<mhzw _\.._ _\...
«Ohgmluozoz KUNEOP< mzmq thdba mzmq th<30 #00:
r 1
.a V '7
C i
TR; X a. MA .8. n % ..m u.
m ZOE.<¢DOEZOU
.EJm
moz<mhzm _\n. _\.._
IOF<ZOGIUOZOZ mzw._ thdjo «BEIGE» mzm: thdao kooI

.V:‘.'v

Will .3 V7? er.

 

8V /—\ =V /_

\7 \_
< zocqgorizoo

9?

2U

preferred for detection limit studies. Also, since the
observed area above the atomizer is smaller than configura-
tion B, this arrangement is useful when volatilizing an-
alytes that have a short half—life as free atoms in the
atomic vapor. Note, however, that this configuration
precludes mounting the atomizer parallel to the optical
axis of the instrument so that the advantages of parallel
mounting described in the previous section may not be
realized.

When mapping experiments are required, the optics
must be set up in configuration B, but the atomizer may
be mounted either parallel to or perpendicular to the
optical axis. Also note that although this configuration
produces a beam several mm in diameter and with a small
divergence, the area of the vapor cell observed is actually
defined in the horizontal plane by the monochromator slit
width and in the vertical plane by the diameter of the
aperature (typically 3 mm) at the entrance slit of the

monochromator.

D. ATOMIZER POWER CONTROLLER

1. Introduction and General Overview

One of the critical parameters in EAAA is the tempera-
ture of the atomizer during the desolvation, ashing, and

atomization heating steps. In the atomization heating

25

step, the rate of temperature increase, as well as the
final temperature,can have a profound effect on the size
and shape of the absorbance signal. The purpose of the
atomizer power controller is to provide a good, stable
temperature related regulation of the atomizer power sup-
ply and to allow easy comparison of this regulation method
to the more conventional regulation methods.

The temperature of the atomizer is related to the
power input, power loses, and the mass of the atomizer.
Power input is easily defined as the product of the current
through the atomizer and the voltage drop across the
atomizer (27,29). As the atomizer ages its mass and
resistance change due to sublimation of the graphite dur-
ing heating. If a conventional power supply is used to
regulate either the current through or the voltage drop
across the atomizer, the final temperature for a requested
current or voltage will change as the atomizer ages (27,
29). Regulating the power dissipated (power regulation)
during a heating step should offset the effect of changes
in resistance of the atomizer due to aging, but will not
counteract the effect of atomizer mass changes. It would
be even more desirable to choose a feedback signal that
was directly related to the atomizer temperature. When
moderate to high heating levels are desired this type of
regulation is possible (27,29) by monitoring the black-

body emission of the graphite atomizer (radiation

26

regulation). The circuitry to handle these functions

will be described in this section along with the circuitry
which provides the hydrogen diffusion flame. Character-
ization of the different methods of regulation as they
relate to overall EAAA measurements were carried out by
Dr. David Baxter and are described elsewhere (28).

A simplified diagram of the power controller and its
relationship to the general interface and the atomizer
power supply is shown in Figure A. The raw power supply
and water cooled power transistor bank were built by the
MSU Chemistry Department electronics shop for Dr. Akbar
Montaser (27,29). Regulation is achieved through a stan-
dard Darlington transistor configuration and a control
OA. One unfortunate characteristic of this design is
the amount of power wasted in the ballast and sensing
resistors. Since the atomizer resistance is about one
ohm, and the sum of the resistances of the ballast and
sensing resistors is about .8 ohms it is obvious that
much of the power supplied by the raw power supply is
dissipated in these resistors and is not available for
heating the atomizer.

All of the feedback signals necessary for current,
voltage, power, or radiation regulation are brought into
the controller box. In the first section of circuitry
they are modified by amplification, level shifting, etc.,

and then the prOper signal is selected to be passed to

27

>mt30w=o 20:415me Gut/On. EMS—20.5 m0 3m_>mm>0 " v mmDGE

 

 

 

 

 

w>J<>

 

 

90238
96

 

 

 

 

"llIllllIIlllJ

$39.58 $30.". $520.2.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.
n _
_ llamll: _
_ \\\\\4l _ IX)
_ >
u 38 Swim _
6/5 _ 548%.“. _
_ . «3.205
:mtaomo 29.22282 _ ._>>
_ 2.10:3 ._<zo_m _ a
Illa .5: _ > WV
7 v _
_.. I I I 11 lllllllllllll L ..|._
2.... .V
hm 292.50% .
04.: .3... > 29.255 . +
E3 5&3
34mm»; 43.38 $.30“.

 

 

 

 

 

 

 

 

 

34m

 

 

¥z<m

"5.5.9de
1950a

28

the control operational amplifier where it is compared
with the reference voltage. This OA then sends an approp-
riate signal to the power transistor bank to correct the
current through the atomizer.

Signals from the general interface circuitry provide
the heating level reference voltage, the signal to switch
to radiation regulation, and the signal to switch on the
gas for the hydrogen diffusion flame (HDF). Front panel
switches include the following: power on-off, kill
atomizer power, hydrogen diffusion flame override switch,
and two regulation mode selector switches for the heating
steps. The last two multi-position switches allow a dif-
ferent regulation mode to be selected for the ash and atom-
ize, or Just the atomize heating step or steps. Thus
it is possible to select current, voltage, or power regula—
tion for the desolvating step and any of these or radiation
regulation for the ashing and atomizing, or Just the
atomizing step. A precision potentiometer with a locking
mechanism allows the adjustment of the sensitivity of the
radiation feedback circuitry relative to the other regula-
tion methods and may be used to counteract differences in
phototransistor sensitivity if the phototransistor has
to be replaced or mounted in a new position. Back panel
banana Jack outlets provide monitors of the current,
voltage, power, and radiation levels regardless of which

regulation method is actually being used. Also on the

29

back panel are the ports for receiving the interface
signals, the power outlets to the gas valve switch, and
the connectors necessary to communicate with the power
transistor bank, phototransistor, and the current sensing

resistor.

2. The Control Operational Amplifier

The control OA section of the power controller circuitry
is shown in Figure 5. The first transistor in the Darling—
ton cascade is actually in the power controller box and
not in the power transistor bank. The base of this tran-
sistor is connected to the output of the control OA (0A1)
and regulates the current from the raw power supply into
the base of the power transistors in the Darlington cas-
cade. This OA operates on a current null comparison and
its inputs are protected from damage by overvoltage
spikes by two diodes. Also the resistor, capacitor
circuit serves to the limit frequency response and thus
power supply oscillation. OA2 is configured as a differ-
ence amplifier and as such provides the dual purpose of
buffering the input reference voltage and providing some
protection from ground loops. The grounds of the interface
circuitry and the power controller circuitry are connected,
but a small potential difference might exist and cause
not only imprecision but also oscillation in the power

supply. The 1 k9 potentiometer allows some adjustment of

30

nmomzm

.>m._._30m_0 <0 Jomkzoonm mmDOE

 

 

 

 

 

 

 

 

 

31

the atomizer heating level relative to the voltage refer-
ence input. The feedback signal is fed from the front
panel regulation mode selector switch to 0A3 where the

polarity is reversed but the magnitude is unchanged.

3. Current, Voltage, and Power Regulation Circuitry

The current regulation is quite straight forward and
is shown in Figure 6 along with the voltage and power
regulation circuitry. OAl buffers the voltage signal
from the sensing resistor that is located in the power
transistor assembly bank. The output of OAl is sent to
the front panel mode select circuitry and the back panel
current monitor output.

In previous designs the voltage drop across the atom-
izer was monitored with an instrumentation amplifier (27,
29). However, a difference amplifier (OAl) is quite suf-
ficient since the input voltages are not large, and load-
ing of the voltage source is not possible in this case.
The potentiometer on the non-inverting input was adjusted
so that the CMRR of the difference amplifier was maximized.
Again the output voltage for the voltage regulation cir-
cuitry is sent to the front and back panels.

The power regulation circuitry utilizes an integrated
circuit (IC) four-quadrant multiplier to compute the
product of the current through and the voltage drop across

the atomizer. The analog multiplier contains two on—chip

32

Embaom.
0 2073..
wam mm
30.... oz .
4 w0<._.:_
O> szm
~50 “m m
~50.“—

 

 

 

 

 

 

 

x. v.0.

33

difference amplifiers. The multiplier's transfer function
is the product of the differences of the voltages appear-
ing at the difference amplifiers inputs divided by ten.
The atomizer current is represented by the voltage drop
across the sensing resistor, and this voltage is pre-
sented to one input of one of the difference amplifiers.
The series resistor and the head-to-head Zener diodes
protect the analog multiplier chip from the overvoltage
damage. The inputs to the second difference amplifier

are similarly protected and are used to monitor the voltage
drop across the atomizer. At these inputs the two vol-
tage divider networks scale the signal voltages to the
appropriate levels and again a trimpot is provided to
maximize the CMRR. The 20 k9 potentiometer is used as a
fine adjust on the output voltage level. The output of
the analog multiplier is buffered by a gain-adjusting

OA (OA2) before being sent to the front and back panels.

A. Radiation Regulation Circuitry

The radiation regulation circuitry is shown in Figure
7. The current output of the phototransistor, which is
mounted to observe the atomizer blackbody emission, is
converted to a proportional voltage by 0A1, and the gain
adjusted and the polarity reversed by OA2. Since this
regulation mode is not applicable at low temperatures

because the atomizer emission level is very low, a

311

.>m._._30m_u 20:44:09“ 29.5204”. "h meoE

>>
a
>6 9. >
._<zo_m

02.2452...
5339. 92.25.83 20.554: .232.

 

 

 

 

 

 

 

 

 

 

 

35

switching circuit is necessary to introduce another con-
trol mode for these temperatures. This switching circuit
allows the desolvation, or desolvation and ashing, heat-
ing steps to be regulated in a different manner from the
following heating step or steps. The switching is accom-
plished by an IC voltage switch controlled by a signal
from the logic circuitry. One input of the voltage switch
is from the regulation mode select for the initial heating
steps and the other from the front panel regulation mode
selector for the regulation mode for the final heating
steps. The grandiose logic circuitry consists of one 7402
quad NOR state, which is used to provide a switch signal
to each of the analog gates on the voltage switching IC.
Since this is the only logic chip in the entire circuit,

a cheap, but effective, DC to DC power supply was con-
structed from OA3. Again, the radiation feedback signal
is fed to both the front panel mode select switch and a

back panel monitor outlet.

5. Hydrogen Diffusion Flame Switching Circuit

The circuit necessary to switch the line voltage
solenoid gas valve for the hydrogen diffusion flame is
shown in Figure 8. The transformer, rectifier bridge,
and filter capacitor form an unregulated 10 V DC power
supply for the triac trigger circuit. When the hydrogen

diffusion flame (HDF) is requested, the interface sends

36

EDUEU 02.10:; 9244“. ZOGDnEE zmwomo>1 ”m manner...

U>J<> m<0
QOZNJOm

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

37

a TTL signal to the optical coupler and causes the photo-
diode to emit. The phototransistor receiver is turned on
when the light is received, and the current from it is
boosted by transistor Q1 to a level that will turn on

triac T1. The optical coupler is a safety device that
serves to isolate the digital circuitry in the inter—

face from the line voltage necessary to operate the sol-
enoid gas valve. When the triac is triggered, the solenoid
gas valve (Cat. No. 2-l2, General Valve Corp., E. Hanover,
N.J.) opens and allows the hydrogen gas to mix with the
argon flowing into the atomization cell. The H2 OVERRIDE
switch is wired in parallel to the triac switching cir-
cuit and thus may be used to actuate the gas valve in-
dependent of the remote switching circuit. This is useful
when the rotameter that controls the H2 gas flow rate needs
to be adjusted. The sudden switching of hydrogen gas
causes a surge of gas through the atomization cell. This
effect can be mediated somewhat if some glass wool is
packed into the H2 gas lines to provide some backpressure.
The hydrogen reacts with the traces of oxygen in the argon
sheath gas only when the atomizer reaches a dull red heating
level (lOOO—l200°C). Thus it is necessary to exercise
caution when the H2 flame is used. If the flame con-
sistently does not light, the laboratory may experience

a sudden violent metamorphosis as this accumulated, unspent

H2 reacts rapidly to form water. Under normal conditions,

38

the amount of H2 used is less than 1 Amin and the H2 is
switched on only during atomization. The diffusion flame
produced occupies the entire height of the quartz walls

of the atomization chamber. Fortunately, the flame is
relatively cool and does not cause extensive heating of
the cell walls. Also, it is nearly totally optically
transparent. In fact, often the only real visual evidence
that the flame has been ignited is the observation of the

Schlerien effects it produces.

6. Prospectives

Much of this circuitry was constructed using archaic
techniques and components and, because of this, suffers
from contact resistance and other problems. Still, it
has performed quite well. The major drawback of this
design is that the useful range of the temperatures under
radiation regulation is limited by the dynamic range of the
phototransistor. This means that to have acceptable regu-
lation at high temperatures, a neutral density filter must
be inserted in front of the phototransistor to decrease
the amount of light it observes. Ideally, radiation
regulation would have at least two separate photosensi-
tive elements; one for the low temperature range and
another for the high temperature range. A phototransistor

is perhaps the best for the low temperature range, since

39

it has good sensitivity. But, in the high temperature
range a photodiode should be used because it has a faster

response time and a wider dynamic range.

E. POSITIONER

 

1. Introduction and General Overview

The positioner is the device which displaces the atom-
ization cell in two dimensions with respect to the optical
axis of the instrument and thus allows the observation of
the atomic concentration in any locale within the atomiza-
tion cell. This device consists of two physically sepa-
rate modules. One module contains the interface, general
logic circuitry, stepper motor power supplies, and front
panel controls. The second module (shown in Figure 9)
consists of an aluminum superstructure on which is
mounted the translational stepper motors, an x and a y
translational stage, and the atomization cell with its
associated optics. The positioner can operate in either
a local mode or under computer control. Details concern-
ing the operation of the logic circuitry can be found

elsewhere (28,68) and will not be discussed here.

2. Physical Structure

Two translational stages (Cat. No. “20-51, Newport

Research Corp., Fountain Valley, CA) and two stepper

HO

.m wizard

 

motors (Cat. No. 23D—6102A, Computer Devices, Santa Fe
Springs, CA) accomplish the horizontal and vertical dis-
placement of the atomization cell. In each case the step-
per motor shaft is directly coupled to the screw drive of
the translational stage so that no imprecision is intro-
duced by a gear system. The direct coupling was accom-
plished by drilling a hole the size of the stepper motor
shaft in the head of the translational stage lead screw
and locking the shaft and the screw together with two
allen screws.

Both translational stages are perpendicular to the
optical rail and are mounted side by side. The side of
the base structure of the atomization cell is fastened to
the mobile section of the vertical translational stage.
The vertical stepper motor is fastened to the immobile
section of the vertical translational stage and, as the
stepper motor is pulsed, the lead screw which drives the
translational stage rotates inside a nut attached to the
mobile part of the translational stage. Thus, as the
stepper motor is pulsed, the mobile section of the trans-
lational stage moves relative to the position of the im-
mobile section. The entire assembly is mounted on the
mobile section of the horizontal translational stage.

The horizontal motor and the immobile section of the
horizontal translational stage are connected to the

positioner superstructure which is, in turn, fastened to

U2

the optical rail. As in the vertical section, the hori-
zontal stepper motor is secured to the immobile section

of the horizontal translational stage. As the motor is
pulsed, it turns the lead screw which causes the mobile
section of the translational stage to move relative to

the immobile section. Optical interrupters and occluders
(mounted on the atomization cell, translational stages,
and the positioner superstructure) are used to define the
limits of travel of the translational stages and to pro-
vide a reliable reference position. The positioner super-
structure is fastened to the optical rail in such a manner
that it is possible to move it anywhere along the optical
axis of the instrument. This structure also contains one
fixed position and one mobile lens mount in-line with the

entrance slit of the monochromator.

3. Remote and Local Operation

When the front panel mode switch is set to LOCAL, the
other front panel switches are activated. The RESET
switch causes the positioner to drive both the horizontal
and vertical translational stages to the reference point
defined by the optical interrupters described previously.
This resets the counters in the logic circuitry to zero
and, as further movements are made, these counters are

used to point to the present position of the atomization

”3

cell.

To move the atomization cell away from the reset
position, numbers, which correspond to the desired hori-
zontal and vertical destinations, are entered into the
front panel thumbwheel switches, and the GO switch ac—
tuated. This causes the motors to begin concurrently
to seek the desired location. If the direction of the
movement is positive, the motors will move slightly past
the desired location and then return to it. This means
that the final direction of movement is always negative
and, whatever slop there may be in the drive mechanism,
is thus corrected.

The thumbwheel switches are U digits wide, but the
logic circuitry accepts only 12 binary bits (a concession
to the 12-bit computer used in the remote operation mode),
so that legal decimal locations are from O to A095 in both
the horizontal and vertical directions. If a destination
outside the legal limits is requested, a fault condition
occurs, which lights the front panel FAULT LED and aborts
the movement. To correct this fault condition, a reset
must be executed. The position horizontal O and vertical
0 is the reset position, and the position horizontal
20H8 and vertical 0 locates the atomizer just beneath the
path of the absorbing light beam. This latter location
is also used as the place for sample deposition when auto—

matic sample deposition is desired. The actual physical

UH

diSplacements that correspond to these user units can be
easily calculated since each user unit corresponds to 2
stepper motor pulses, the stepper motors require 200 steps
per revolution, and the lead screws are machined to have
no threads to the inch. Thus, the end—to-end travel in
both the horizontal and vertical directions is limited to
8192 pulses (1.02“ in).

If the front panel mode switch is set to the LINE
position, the other front panel switches are deactivated.
Commands are then accepted only from the computer. The
possible computer commands are GO, RESET, skip on flag,
clear flag,enable interrupt, disable interrupt, load hori—
zontal destination register, and load vertical destination
register. The CO and RESET commands function exactly the
same in the remote mode as they do in the local mode, and
the load location register commands serve the same function
as loading the front panel thumbwheel switches. Now,
however, it is impossible to load an illegal destination
since the locations are transferred in l2-bit binary.

The skip-on—flag and clear flag commands are necessary

so that a new command will not be accepted before the
previous command has been completed. The positioner
interface was constructed so that it could be interrupt
driven. However, because of the timing and position con-
straints imposed on the positioner by the automatic sample

dispenser and the data acquisition system, no practical

“5

time savings would be realized by operating the device
under interrupts. Thus, the added software complexity

of an interrupt system could not be justified. A more
detailed discussion of design concepts of the positioner
is given in Appendix B and a very detailed description of
the positioner's electronic circuitry is also available

(28).

F. GENERAL INTERFACE

 

All communications to and from the computer with
the EAAA instrument occur through the general interface,
except for the positioner which contains its own inter-
face. The interface functions may be conveniently divided
into five categories: automatic sample dispenser, analog-
to-digital conversion, digital-to-analog conversion,
general purpose flags, and input switch register. An
overview of the automatic sample dispenser interface is
given in Chapter A and is discussed in great detail else-
where (28).

Voltages representative of the photomultiplier photo-
currents are quantized by an analog-to-digital converter
(Cat. No. ADC-HY12BC, Datel Systems, Canton, Mass.). This
l2-bit, hybrid, successive approximation converter has a
conversion time of about 8 us and a maximum non-linearity
of il/2 LSB. In practice the quantization error due to

the l2—bit resolution of the converter was observed only

A6

in the measurement of dark current and was small enough

so that it can be discounted as a significant source of
error. The maximum conversion rate was about one order

of magnitude faster than the software could handle the
data. Even so, the software-limited conversion rate was
still more than an order of magnitude greater than the
slowest rate acceptable for an accurate integration of
even the fastest absorbance signals. Although a converter
of this speed is unnecessary for normal integration of
absorbance transients, kinetic studies and studies of the
rate of atomizer temperature increase might require a fast
converter such as the one used.

The digital—to—analog converter (Cat. No. DAC—HY12BC,
Datel Systems, Canton, Mass.) is used to supply the
reference voltage for the atomizer power supply control-
ler. This l2—bit hybrid converter has a conversion time
of 3 us and a maximum non-linearity of il/Z LSB. Again
these specifications are much better than required by this
particular application. For example, in radiation regula—
tion, the resolution of this converter (2.5 mV) corresponds
to a temperature resolution of l.0° to 3.0°C (the actual
resolution depends on the particular temperature region).
Atomizer temperatures determined by optical pyrometry have
a precision of at best i25°C. Furthermore, because of
the physical nature of the atomizer, the temperature in

some regions on the atomizer may be more than 25°C different

“7

from the nominal temperature. It is important that the
atomization temperature be stable and reproducible, but
for most elements the absolute value of this temperature
may be any value within a several hundred degree range
and still not affect the precision and accuracy of the
measurement. The converter's slew rate of 20V/us and
settling time of 3 us far exceed the response time of the
atomizer heating circuitry and in no way limit the time
it takes for the atomizer to reach its final tempera-
ture. The hardware associated with the three general
purpose flags is composed of three flip-flops and associ-
ated circuitry. It is so routine that it does not war-
rant discussion here. The user input switch resister

is an array of six single-pole, single-throw switches in
a dual-in-line package on the front panel of the general
interface box. On computer command, the status of these
switches is read into the 6 most significant bits of the
accumulator. Again the interface circuitry is of a com-
mon garden variety and does not warrant detailed discus-
sion here, since it has already been discussed in detail

by Dr. Baxter (28).

CHAPTER 3

PROGRAMMING THE COMPUTERIZED ELECTRO-THERMAL

ATOMIZER ATOMIC ABSORPTION INSTRUMENT

A. INTRODUCTION

 

The electronic revolution of the 1960's and 1970's
caused a significant increase in the availability of
computers to chemists. At first chemists used computers
only in the passive role of data reduction. Typically
experimental data was manually transformed into a computer
compatible form such as punched cards and physically trans-
ported to the site of a large central computer for data
analysis. This large computer generally had a high level
language such as FORTRAN available if the chemist had
chosen to do his own programming, or sometimes he might
even be able to use a routine supplied by the computer
vendor. When the computer price structure changed so that
it was reasonable to dedicate a computer to one instrument
the scope of the Chemist's software problem changed sig-
nificantly. Now the computer could play an active role
in the experiment including sequencing events and control-
ling data acquisition. More often than not computers in

this price range had no laboratory oriented high level

A8

A9

language and even if one was available special control
functions and data acquisition functions required routines
which had to be written in assembly language. Now when-
ever a chemist found it necessary or desirable to modify
an existing computerized instrument, computerize an exist-
ing instrument, or to build a totally new computerized
device, he needed to deal with programming on a much more
complex level. The problem was attacked in one of three
ways: hire a professional programmer, use a special in-
house developed high level laboratory oriented language
(70—72,7A) or the chemist would learn how to do his own
programming. The basic problems associated with hiring
professional programmers include communication difficulties
between the chemist and the programmer concerning the exact
requirements of the software and also the basic inflex—
ibility of this approach in dealing with changing require—
ments of the laboratory. Special in-house laboratory
oriented high-level languages never have achieved wide-
spread popularity since the initial effort of developing
them for each particular type of computer and interface
requires too much time investment. Worse still some of
these languages wasted some of the computer's power by

not making some functions of the computer accessible to
the user (71,72), and also lacked good error diagnostic
messages.

A number of developments have made the task of

50

educating the chemist to be his own programmer easier

and more attractive. First, the languages available on
small computers have developed far past the original
assembly languages so that most general purpose computers
will support FORTRAN or BASIC and often times both of
these languages. Also the price of peripherals has de-
clined so that mass-storage based operating systems are
often available to the chemist. These systems make the
computer easier to use and decrease programming time
significantly.

Another consequence of the declining cost of computer
mainframes has been the increased proportion of the cost
of an experiment in terms of both time and money that was
invested in software. In fact, in some cases software
costs may comprise 80% of the total development costs (7A).
The availability of good software support is now one of
the major factors in the choice of which particular
computer a chemist will purchase. Even if the choice of
the computer is already made, there is still the choice
of what language to use. Writing in a high level language
may use 50 to 300% more core memory (75) and slower execu-
tion time, but still may be the method of choice because
it takes less time to learn the language and less time
to write and debug a program. Poor programming techniques
and an inadequate programming philosophy in either assembly

language or any high level language can severely limit

51

the capabilities of the interfaced instrument, limit
future modifications and expansion, expand software main-
tenance costs, and needlessly limit the instrument's
throughput. The remainder of this chapter describes how
these considerations affected the design of the software
of an electro-thermal atomizer AA instrument. The result-
ing package of hardware described in Chapter 2 and this
software yielded a versatile, powerful, and easy to use

analytical research instrument.

B. CHOICE OF A PROGRAMMING LANGUAGE

 

To make an intelligent choice of a programming lan-
guage it is first necessary to define clearly the tasks
the program will be required to handle. The minimum a
reasonably useful program would be expected to do for our
experiments included sequencing the operations of the
autosampler, positioner, braid heating apparatus, and data
acquisition system. Also, it should compute the integrat-
ed absorbance and have provisions for automatically scan-
ning the horizontal and vertical displacements for a one
or two dimension profile of the atomic cloud above the
braid. Further, a hardcopy of the integrated absorbance
and experimental parameters should be available and a
plot of absorbance versus time should be presented on

the system terminal. In summary, the program is required

52

to do significantly complex calculations both for determin-
ing integrated absorbance and scanning parameters, and

also control a significant amount of I/O with the com-
puter's peripherals and the instrument's interface.

The languages available on the 8/e which might be
used to write this program included: PALD, PAL8, MACRO8,
SABR, RALF, FOCAL, BASIC, FORTRAN II, and FORTRAN IV.

Of these PALD, MACRO8, and FOCAL are not supported by

the 08/8 operating system and since the operating system
is a powerful tool used in creating, debugging, and execut-
ing programs, these three languages were immediately
rejected as unsuitable. The assembly languages remain-
ing, PAL8, SABR, and RALF, were rejected as possible
alternatives for two reasons. First, the program required
significant device dependent I/O, computation, and complex
sequencing, all of which are much more easily dealt with
in a high level programming language. Second, the mini-
mum required program, even when written in a high level
language, should fit in the available core memory and not
significantly degrade execution time performance.

The remaining high level languages are FORTRAN II,
FORTRAN IV and BASIC. FORTRAN IV and BASIC are more
powerful than FORTRAN II with respect to the capabilities
designed into the compilers. However, each time machine
level code needs to be executed it must be referenced

as a subroutine call. Writing this code requires a good

53

deal of familiarity with the subroutine calling conven-
tions of these compilers and also is memory inefficient.
0n the other hand FORTRAN II allows in line insertion of
SABR code which allows the programmer to write efficient
code while remaining fairly ignorant of much of the inner
workings of the FORTRAN II compiler. Further, FORTRAN II
does have all the necessary I/O handling and calculation
routines. Thus FORTRAN II was initially chosen as the
programming language for the instrument, primarily because
of ease of use and programmer familiarity with the lan-

guage.

C. FORTRAN II PROGRAM CAPABILITIES

The program which was written using FORTRAN II as the
source language meets all the minimum requirements neces-
sary for a useful program as described in the previous
section. It is written in a modular fashion so that
whenever possible each major function is allotted its
own particular subroutine. Several of these routines were
written by Dr. Eric Johnson including the routines to
drive the autosampler (SAMP), positioner (POS), the routine
to oversee one sample run (GBRUN), and the plotting pack—
age (ADPLOT).

The main program, GBEXEC, sequences calls to sub-
routines to initialize the positioner, initialize the

autosampler, and take the dark current reading. Then it

5A

interacts with the user so that he may choose such things
as: the type of experiment (autosampling; zero, one, or
two dimensional scan; etc.), and program options (hard
copy of results, plot of absorbance versus time on system
terminal, etc.). When a requested series of sample runs
has been completed the program allows the user several
options concerned with changing parameter values, changing
program options, or exiting the routine.

For each individual sample run GBEXEC sets up the
proper variable values and passes them to a subroutine
called GBRUN. This routine interprets the variable values
so that the proper subroutines are called in the proper
sequence for one complete sample run. This will include
some or all of the following functions: sample deposition
(SAMP), movement of the braid to the sampling position
(POS), heating the braid (TKDATA), and data acquisition
(TKDATA), calculation of integrated absorbance, hard copy
of results, and plot of results (ADPLOT library).

The TKDATA subroutine sets the digital to analog con-
verter for the proper heating levels for the desolvate,
ash, and atomize powers. The routine uses the real time
clock (CLOCK) routines to execute the programmed timed
delays in the delay, desolvation, and ash steps. The
atomize time delay is provided internally to the data
acquisition section of the TKDATA routine. This section

takes data at 500 Hz and stores the individual analog

55

to digital conversions in the FORTRAN II defined arrays
IDATA and IBACK. The data are routed to IDATA for a
sample run and IBACK for a background run.

In addition to the above routines, several utility
routines were developed including one named BELL. This
routine was used to signal the operator that his inter-
action was required to do some mechanical function (such
as open the PMT shutter) or to alert him that the present
task had been completed. The program caused the terminal
bell to be rung periodically until the user typed a spec-
ial character which caused either resumption of program
execution or an exit to keyboard monitor (the 08/8 operat-
ing system).

The program as described occupied all of the avail-
able four fields of core memory. It used the fixed volume
sampler designed by Akbar Montaser (29), and the pro-
gram was used routinely in the process of checking the
reproducibility and long-term drift characteristics of
this sampler when it was coupled with the graphite braid

atomizer.

D. THE PDP 8/e FORTRAN IV SOFTWARE

The FORTRAN II source program described in the pre-
vious section was adequate for initial experimentation,
but proved somewhat inflexible in meeting the changing

demands of our research program. First, the original

56

autosampler needed to be replaced by one that was more
versatile and reproducible. This second sampler required
much more complicated machine level coding, which would
have further cramped the available memory. Furthermore,
any significant departure from the alternatives designed
into the program required wrestling with modifying the
already tortuous flow diagram of GBEXEC. Writing an
equivalent but more powerful program using FORTRAN IV be-
came a necessity. Some of the advantages gained by using
FORTRAN IV as a source language were the following: in-
creased effective program size through the use of FORTRAN
IV overlaying capabilities; improved calculation times
since FORTRAN IV uses the extended arithmetic element -
FORTRAN II does not; improved program execution time due
to the efficient interrupt driven structure of the FORTRAN
IV run time system; DEC's FORTRAN IV also has full ANSI
compatibility so that programs from outside sources can
be more easily assimilated into the experimental routines;
and FORTRAN IV has a more powerful set of possible program
statements and instructions, which make programming and
debugging easier.

Before attempting to explain the structure and capa-
bilities of the FORTRAN IV GBA program, it is necessary to
explain briefly DEC's FORTRAN IV software package. The
package contains four major components: a compiler,

assembler, loader, and run time system. The FORTRAN IV

57

compiler produces source files in RALF assembly language
in three passes with an optional fourth pass for listing
files. The RALF assembler accepts the compiler output
and creates relocatable binary modules for execution on
machines that do not have a floating point processor (if
the FFP is available a FLAP assembler is used). The third
component is a LOADER which combines all of the inde-
pendently compiled and assembled relocatable binary files
of the main program, subroutines, and library routines into
a load module. The fourth component is the Fortran Run
Time System (FRTS). This program supervises loading the
load module, swapping overlays as necessary, and handles
the I/O requested by the program.

The capabilities of the FORTRAN IV compiler not avail-
able in FORTRAN II that are of particular interest for
our uses include the availability of logic variables,
logical IF statements, and Hollerith formats, which
allowed easy implementation of the monitor type program
structure described later. In line insertion of error
statements in listing files greatly simplifies debugging
operations. Most important, however, was the possibility
of more versatile device independent I/O, which allowed
hard copy results to be generated on the line printer and
data to be stored in two or more data files open for input
concurrently. FORTRAN IV also has available an extensive

library of utility routines. One utility routine of

58

particular note is ONQI which adds the user written inter—
rupt handler to the run time system's polled interrupt

skip chain. Another utility routine, ONQB, adds a user
written routine to the background task list of the run

time system. These two routines allow the programmer to
structure execution of routines such as plotting and

device drivers (e.g., autosampler handler) in such a

manner that instrument throughput is enhanced. Compiler
options A and F which allow the generation of RALF assembly
language source files, and fully annoted listing files,
respectively, are of great utility. Thus, when machine
level functions are required, a skeleton FORTRAN sub-
routine can be compiled under the A option and can then

be edited so that the desired machine level code is added
at the proper position. This greatly simplifies the
machine level subroutine writing, since much of the calling
convention and argument passing is already taken care of

by the compiler. This modified routine can then be loaded
with FORTRAN routines and called in the same manner as

any other FORTRAN subroutine call. To modify a RALF source
file, the programmer need only understand a few of the RALF
instructions and argument passing conventions. The RALF
assembler operates in two modes: the ralf mode and the
eight mode. The ralf mode has one, two, or three word
instructions that deal with a software simulated three

word floating point accumulator and seven, one word

59

index registers. This format allows 15 bit direct
addressing, base page relative addressing, and register
indirect addressing. The eight—mode RALF code uses the
8/e machine level instruction set of PAL8 except for a
slightly different indirect addressing convention and a
different method to indicate the memory locations occu-
pied by the routine. The main task of the machine level
programmer is to conform to the RALF conventions with
respect to argument passing and transferring execution
back and forth from the ralf-mode and the eight-mode code.
The FORTRAN IV loader differs from the FORTRAN II
loader in that it allows overlaying. Up to eight levels
of overlays are allowed with up to 16 overlays per level,
except for level 0 which is always core resident and may
never be overlayed. Level 0 must contain the main program,
library routines, and any modules containing eight-mode
RALF code. There are a few important restrictions to
keep in mind when designing an overlay structure. Func-
tionally, the loader regards all overlays in a given level
as having the same length as the longest overlay. All
overlays are multiples of two blocks long. Also, as a
general rule, no program may execute a call which would
cause itself or any program that called it to be overlayed.
The loader provides an optional loader map listing which
designates overlay levels, program locations, etc. In

the loader map there are two error messages which do not

60

appear in the DEC documentation. An : beside a subroutine
name designates that there is a possible violation of the
overlay calling rule mentioned above. A :EE designates
that the loader did not locate any entry point correspond-
ing to this subroutine name, even though it had been re-
quested to be loaded.

The load module created by the loader is executed
under the auspices of the FORTRAN Run Time System (FRTS).
It is this program that allows devices to be defined for
device independent I/O, sets up appropriate buffers, and
provides the interrupt driven handlers to allow fore-
ground I/O with background computation. Another nice
feature is error trace back at run time. This function
detects fatal run time errors, gives terse error state-
ments, and fingers the culprit both as to line number and
subroutine name. Typing a CONTROL B at run time provides
this same trace back function.

The PDP 8/e FORTRAN IV package as described here
possesses many attributes not found in the FORTRAN II
package and was used to generate a more flexible, power-
ful, and easily modifiable program than the program which

was written using FORTRAN II as a source language.

61

E. PROGRAM STRUCTURE FOR THE EAAA INSTRUMENT

 

Because of the limitations of the FORTRAN II program
and the extended possibilities of programming using the
DEC FORTRAN IV package, it was advantageous to expand the
tasks required of the GBA software. Rewriting and restruc-
turing the software using FORTRAN IV as the source lan—
guage resulted in improvements in the following areas:
decreased execution time, raw and processed data avail-
able in output files, more error checking functions, in—
creased modularity so that software was more easily modi-
fied, and added features which expanded the abilities of
the program to handle situations and experiments not
originally designed into the software.

The overlay structure of the resulting program
(GBMDS2) is shown in Figure 10. In the figure, routines
which occupy the same level but different overlays are
separated by a colon. These routines occupy the same
portion of core but are swapped into memory at different
times during execution. If the routines occupy the same
level and overlay, their names are separated by a comma.
These routines are always co-resident in core since the
run time system always swaps into core entire overlays
not just portions of the requested overlay. The lines
connecting programs denote that the program in the upper
level may call the lower level routine. Note that routines

in level 0 may be accessed by routines in any level, but

62

Emgmcga «magma on» we ogauuzgam Ampgo>o mgu we Eagmm.o .ofi «Lama;

 

 

 

 

xa<¢o . mom n
mamxxm . pi e
~w4_uuz:c.ozamucno<mxlpaa.4m<4um_x<.hzamnozpm:m"wzomam,»_x_4.a<mnz~c<anxc<oumzap.oa<m m
hoaao< N
wwmo

o<ux-auoumzlhxuudx_mumzpm:m . goud.ua<o-<g<o ” zouu ” padz~ _

cxu>z< .xuoau .Jaum .zu»_zm ///////
xoo< .u<o .m<4u .<»<o xuoao .‘~<x o

x<mwoma Ju>w4

63

to avoid mass confusion in the diagram these connecting
lines were left out. The routines residing in level 0 are
the main routine and the RALF subroutines containing
eight-mode code. Since the level 0 routines may not be
overlayed, great effort was made to keep these routines

as short as possible. Also in level 0 is the special
function routine BLOCK DATA, which contains no executable
code and is used to initialize the values of common vari-
ables.

Subroutine INPUT is called immediately upon starting
execution of the main program. Almost all user inter-
action with the experiment is through this routine. Under
user commands the default values of experimental param-
eters set in the BLOCK DATA subroutine may be changed.
Also, initialization and demonstration routines for the
positioner, autosampler, and data acquisition system may
be executed. When the user completes this setup, he gives
a special command to start the execution of the experiment.
The control is then passed back to MAIN which normally
calls the subroutine FLOW, unless in the simplex mode in
which case the SIMPLEX subroutine is called. In the FLOW
subroutine, all experimental parameters are checked for
illegal values. Error messages are printed for each
illegal value detected, and the control is then directed
back to subroutine INPUT. If no illegal values were
detected, new arguments which will later be passed to

the sampler and positioner subroutines are calculated,

6A

the sampler and positioner subroutines are calculated,
and control is passed to the DATA-CALC subroutine. The
DATA-CALC routine interacts with the sampler, positioner,
digital to analog converter (DAC), and analog to digital
converter (ADC) subroutines to acquire, calculate, and out-
put the results of a single sample run to the requested
output devices. If a plot of absorbance versus time for
a sample run was requested, control is passed to the PLOT
subroutine. If no plot was requested, control is passed
back through the main program to the FLOW subroutine.
After the PLOT subroutine has completed a plot of absor—
bance versus time, program control is passed back through
MAIN to the FLOW subroutine, unless the wait mode has
been asserted. In the wait mode, the user has the option
of requesting a plot of background levels or exiting this
routine back to MAIN, which will respond by recalling
subroutine FLOW. In overlay five of level one, the sim-
plex package is loaded. Discussion of the functions of
these routines has been reserved for Chapter 5.

The complete program GBMDS2 may be easily created
if the relocatable sources of the programs in Figure 10
and the FORTRAN IV library are available. This is the
first point of easy program modification since any of
the individual modules may be modified and used in place
of the original version at this point. Further the routines

SUBONE, SUBTWO, SUBTHR, and SMXSUB are optional routines

65

which may, but need not, be loaded concurrently with the
original routines. If loaded, they may be executed at
specific times during program execution if the user re-
quests them at run time. These routines allow for such
things as non-standard calculation schemes or file out-
put formats.

When the user has created one of these new routines or
modified one of the system routines, he can create the new
GBMDS2 program and start execution using the following
dialog. This dialog utilizes the batch stream program

GBMDS2.BI to create the new program.

.SUBMIT SYS:GBMDSZ.BI(UTE)
.R FRTS

*GBMDS2

*FILE1.DA</5
*SYS:FILEZ.DA</6
*$BATCH.TX/A

*$

The first line of the dialog starts the batch stream
and thus creates the load module. The next line starts
the FORTRAN run time system. The first argument to the
run time system is the name of the program's load module
that has just been created by the batch file. The remain-
ing arguments deal with file structured I/O and will be

explained later in the chapter. A more detailed

66

description of some of the most important subroutines is

given in the remainder of this chapter.

F. THE INPUT SUBROUTINE

Most user interaction with the program occurs in the
subroutine or subroutines called by INPUT. The structure
and capabilities of this routine are probably the most
important factors in making the FORTRAN IV program more
usable than the FORTRAN II program. An example, which
illustrates the default parameter values and a sample
dialog is shown in Figure 11. INPUT first gives the pres—
ent version number which indicates to the user that this
is the standard software package or that it contains some
special purpose modifications. The program prompts the
user to enter a command by outputing a colon and ringing
the terminal bell once. At this point the user may enter
any one of approximately 80 legal commands. These commands
are from two to six alphanumeric characters long. Some
of these commands additionally request a numeric argument.
A complete list of the commands along with a brief des—
cription of the function of each one is given in the docu-
mentation section of the INPUT subroutine listing in Ap-
pendix A. In the example given in Figure 11, the first
command selected by the user was HL. The program responded
by giving a listing of the default values of all pertinent

variables. The next command reflected the user's desire

GBA UER 3.01
:HL
EXP: N0 TITLE

FLT F LPT T
531 F 932 F

DELT= 10.00
DESP= 10.00
UER= 2.0000
:NRAU

:DELT

=20

:HL

EXP: EXAMPLE
FLT F LPT T
831 F 882 F
DELT= 20.00
DESP= 10.00
UER= 2.0000
:60

Figure 11.

RAE! T PM”: F
8B3 F FL1 T
BEST: 10.00
A8HP= 0.100
HOR= 0.0000

0F INPUT COHHAND

RAD F HDF F
5B3 F FL1 T
BEST= 10.00
ASHP= 0.100
HOR= 0.0000

67

WAIT F SCAN F
FL2 F SINT F
ASHT= 5.00

ATMP= 0.500
AMNT= 2.00

USAGE

WAIT F SCAN F
FL2 F SINT F
ASHT= 5.00

ATMP= 0.500
AMNT= 2.00

6/21/77
SAMP F
PINT F DARK F
ATMT= 0.35
FREG= 1000.0
NHRUN= 1
6/21/77
SAMP F
FINT F DARK F
ATMT= 0.35
FREG= 1000.0
NMRUN= 1

Example of a User Dialog with Subroutine INPUT

68

to use power programming rather than radiation programming.
With the DELT command and the associated numeric argument,
the user has changed the time that the instrument is re-
quested to wait between sample runs. Finally, the user
checked the status with another HL command and after a
few other intervening commands started the experiment with
the GO command. In most cases the command to change an
argument is the same as mnemonic given in the table of
variable status output on an HL command. Some additional
commands are worthy of special note. For example, several
commands are used to open the data files specified as
arguments to the FRTS in the previous section (FILEl.DA
and FILE2.DA) so that these files contain the present
date, program version number, user comments, etc.. An-
other set of commands initializes the positioner (PSIN)
and the sampler (SPIN). The commands POSDEM and SAMPDM
allow the user to test the individual functions of the
positioner and sampler. The DARK command takes 100 dark
current readings and returns the average. Other commands
allow the user to activate calls to the special subroutines
he may have written and added to the program.

A special set of commands is necessary to allow the
powerful technique of batch streaming command strings
from an 08/8 file to the INPUT subroutine. This batch
file is created using either EDITOR or TECO to save a

list of commands that the user creates just as if he

69

were entering them directly to the subroutine INPUT from
the system terminal. This file may then be specified to
FRTS as device four (file BATCH.TX in the example given
in the previous section). When the INPUT subroutine is
executed, it requests commands from the system terminal
until it receives the command BATCH. Then commands are
requested from the previously designated batch file and
echoed at the system terminal until either an erroneous
command or the command LOCAL is detected in the batch
file. At this point commands are again requested at the
system terminal. Subsequent BATCH commands will start
command execution from the batch file at the command im-
mediately following the last one executed from the file
in the last access. A few functions accessible by INPUT
commands require user interaction. For example, to take
the dark current, the user must be prompted to close the
PMT shutter, and program execution must wait until this
function has been completed. Two special batch-mode
commands help handle this situation. The MSG command
causes the next line of code in the batch file to be echoed
to the system terminal and not interpreted as a command.
In this case the message might be "MASTER PLEASE CLOSE
THE PMT SHUTTER SO THAT WE CAN CONTINUE THE EXPERIMENT".
The BELL command is then used to ring the terminal bell
periodically until the user completes the task and exits

the bell routine by typing a character at the system

70

terminal. All programs that require some run time user
interaction similar to the previous example always re—
quest some interaction at the terminal, even when in the
batch mode, so that these functions are never accidently
bypassed.

Uses of this batch mode of operation are numerous.
The user could easily create a library of batch files
under each element's name and each of these would con—
tain the proper experimental parameters for that par-
ticular element, along with any auxiliary messages to
the user concerning any necessary special procedures.
In more complex applications, the user might use the
batch stream to execute a series of tasks not specifically
provided for in the basic software. For example, the user
might wish to scan through various atomization powers;
a task which is not easily done with the normal routine.
However, with the appropriate batch stream this non-
standard experiment could be executed without the user
having to interact with the program personally each time
he wished to change that parameter. An adequately con-
structed batch file should allow even a moderately com-
petent technician to operate the instrument with good
results.

A simplified flow diagram for the INPUT routine is
given in Figure 12. Note that this type of free-flowing

structure makes it quite a simple task to insert or delete

 

 

 

 

 

 

 

T PROGRAM
[0N NUHBEH

I":
,—
02

 

 

 

 

 

 

ISSUE
PROHPT

 

 

 

 

 

READ
COMMAND
STRING

 

 

 

      
     

 

 

 

 

 

 

 
 

COMMAND OE! HDF
18 HDF? ’ TO TRUE
COMMAND 0E1 HOP

15 NHOF?

 

’ TO FALSE

 

 

 

 

Execurt
FUNCTION 0,

 

 

 

 

 

 

 

ISSUE ILLEGAL
COHIAND HESAOE

 

 

 

 

Figure 12. Simplified Flow Chart for Subroutine INPUT

72

commands, since the position at which the command check
is inserted is quite arbitrary (it only needs to be in-
serted somewhere amongst the existing command checks).

In summary, the main attributes of the subroutine
INPUT include its ability to use batch-mode command input
to make the instrument simple, direct, flexible and highly
automated. Also, the loosely structured program construc-
tion allows easy modification such as deleting or adding
commands as needed. Demonstration routines for the
sampler and positioner are useful in equipment testing and
maintenance operations. The file structure and optional
user subroutines add to the versatility of the program
and allow it to meet the changing demands of a research

project.

G. THE FLOW SUBROUTINE

 

As stated previously, the FLOW subroutine handles

the generation of the proper variable values for a sample
run when the experiment involves either position scanning
or automatic sample deposition. Since some of the modes

of operation are quite complex, the flow chart for this
subroutine is not included here. For example, a two dimen-
sional scan proceeds by starting at the requested position
and scanning vertically at this position and then repeat-
ing this vertical scan at horizontal increments first at

a positive displacement and then at an equal magnitude

73

negative displacement, then at a larger positive displace-
ment, etc. Superimposed on this action may be requests
for a series of different samples and sample sizes.
Further options allow a number of sample runs to be
taken at each scan point and more than one scan to be
taken. This routine also has the assignment of detect-
ing the end of the assigned task and passing control
through MAIN back to the INPUT subroutine, rather than
the normal mode where control is passed to the DATA—CALC
routine. At each call this routine scans a hardware
switch register which may be used for several functions.
An end of task can be simulated by setting one of these
switches so that control is immediately passed back to
INPUT. Several other switches are used to output the
status of run counters, scan variables, and sampler con-
tents. Another switch forces a call to PLOT, which
produces an absorbance versus time plot of the previous
sample run. These switch register options are very
important in program debugging and checking the status
of a long term experiment, but since these tasks are
implemented only on request system throughput is not
degraded. The program listing for FLOW is given in

Appendix A.

7A

H. THE DATA-CALC AND PLOT SUBROUTINES

 

The DATA-CALC subroutine is the routine that inter-
acts with the instrument's hardware in a typical sample
run. The sequence involved in a sample run is diagrammed
in Figure 13. On entry this routine first checks to see
if the positioner is presently busy, and if not busy,
it is sent to the sample pick up position. If the required
time has elapsed since the last sample run, the auto-
sampler is engaged, and the requested sample is deposited
on the braid. The positioner is sent to the requested
test location, and the braid is heated at the desolvate
power for the desolvate time. Now the 100% T reading is
taken. The reasons for waiting until this time to take
the 100% T reading are: first, if the sample has not been
desolvated the sample droplet may occlude some of the light;
second, the 100% T reading varies at different sample loca-
tions due to changes in the optical properties of the
quartz cell walls; finally, to minimize the effects of
HCDT drift, the reading is taken as close to the atomiza—
tion time as possible. Following this the ash step is
executed, and then the radiation and hydrogen diffusion
flame flag is set if these were requested for the atomiza-
tion step. During the sample run the transmittance data
are stored in the array XVAL. Following this an abbre-
viated sequence of delay, ash, and atomize steps is carried

out to acquire and store data related to the background

75

 

 

 

[

 

 

 

 

 

 

SAagZTNB? DEPOSIT SAMPLE ON BRAID

   

 

    

T

NO r-
[ncva POSITIONER TC 1557 LOCATION 1

 

 

 

L 551 BAG :0 DESOLVATION POWER ]

 

 

  
  

   

 

 

DES. TIME
ELAPSEO°

 

 

 

 

 

 

 

L SET BRAID PONER 10 zsno ]

I
[ AVERAGE 10c READINGS 0F erzr ]

F o
L 557 0a: to ASH POKER ]

 

T

 

 

 

SET RADIATION PROS. FLAG

 

l

 

 

SET HYDROGEN DIFFUSION FLAME FLAG

 

I

Figure 13. Flow Diagram for the DATA-CALC Subroutine.

'76

A
[EET DAC To ATDNIZE FOKEP]

11 T
L ACCUIRE DATA |

 

 

 

     

    

    
 
 

BACKGROUND TYPE

SAMPLE
OF RUN?

 

 

 

 

 

[STORE DATA IN AAAAT BACK] EADAE DATA TA ARRAY XVAL]
I T

 

 

 

L ELEM ALL FLAGS j

T

Ltg SET DAC To ZERO J

L

    
   
   
    

SAMPLE
DELAY 3 SECGADG T___>(:)

BACKGROUND
[ START POSITIONER TDAAAD DAIGIA ]

  

SUBONE
TRUE?

 

CALL SUBDNE ]

    
  

YES

SKIP
CALCULATIONS
REQUESTED?

   
   

 

 

 

[g CALCULATE INTEGRATED ABSORBANCE
T

 

 

7E5 CALL SUBTND

F"

 

 

NO

 

YES

 

WRITE ABSORBANCE VALUES INTO FILEZ

 

 

ND
T

RETURN

Figure 13. Flow Diagram of the DATA-CALC Subroutine (continued).

 

77

emission of the braid. The positioner is then started
back toward the sample deposition position, and the calcu-
lation of the integrated absorbance is begun.

The calculation of the integrated absorbance for a
sample run requires four measurements: 100% T, sample
absorbance, background braid emission, and dark current.

The photocurrent components of these are given below.

I1 = IOL + ID

I2 = IAL + ID + IBE + IME

I3 = ID + IBE + I0L

In = ID
where IOL is the photocurrent due to the unattenuated
line source, ID is the dark current, IBE is photocurrent

due to continuum braid emission at the analysis wave-
length, IME is the photocurrent due to the thermally
excited atomic emission of the analyte, and IAL is the
photocurrent due to the line source when it has been
attenuated by the analyte and any background absorbance
of the matrix at the analysis wavelength. If the con—
tributions to the photocurrent by analyte emission and
continuum background absorbance are negligible then the
absorbance at any given time is

I0L Il-IA

A = log ——— = log ,

 

78

and the integrated absorbance is the sum of these absor-
bances over the specified time interval. Note that this
is not a true integration and that it depends on sampling
frequency and peak shape (28).

In the actual coding, the mathematical identity that
states that the sum of a series of logarithms is equal to
the log of the multiplicative product of the series was
used. This improves execution speed since multiplication
is a much faster computer operation than computing
logarithms.

If a continuum source is used instead of the line

source, the pertinent measurements are

In = ID
I5 ’ IOE + ID

I7 ID + IBE + IOE

The new photocurrent components are 100’ which is the
photocurrent from the unattenuated continuum source, and
IH, which is the attenuated continuum source. Because
of the bandpass of the monochromator, the attenuation of
the continuum source by the analyte can be disregarded
when compared to any broad band absorbance by matrix
species. Again if the analyte emission is neglected

Ioc IS'IA

A = log ——— = log
C IC I6'I7+IS'IA

 

79

If the background level is constant from run to run,
the true sample absorbance (AS) is AL = AS - AC. For most
of our work Ad was assumed to be very small so that AL
was taken to approximate AS.

The PLOT subroutine uses a plotting package written
by Dr. E. R. Johnson to present an absorbance versus time
plot. In this plot, the origin of the time axis is the
time at which atomization power was first applied to the
braid. When the plot has been completed, and if not in
the wait mode, the program control is directed back to
the subroutine FLOW. If in the wait mode the user may
request a plot of background transmittance versus time.
Normally, however, the wait mode is used to allow manual
sample deposition, and program execution may be restarted
on command when this manual deposition has been completed.

Listings of the DATA-CALC and PLOT subroutines are
in Appendix A, and more information concerning the exact
mode of operation can be extracted from the extensive docu-
mentation at the beginning of each listing. The important
features to note in the DATA-CALC routine are the switch
register options, which aided debugging and progress
monitoring of long running experiments. Also, note the
importance of point by point braid emission background
correction of absorbance values since especially in the
beginning of the atomization period this background level

changes rapidly.

80

I. ASSEMBLY LANGUAGE SUBROUTINES

 

Because of the number of hardware devices that needed
to be controlled in the GBA instrument, a good deal of
assembly language programming was necessary. In some
cases the skeletons of these RALF routines were derived
from compiling FORTRAN IV subroutines as suggested early
in this chapter. In other cases where no argument passing,
etc., was necessary the FORTRAN IV compiler was overly
verbose so that even the ralf-mode RALF code was written
without aid from the compiler. The program ADCR is an
example of this type and its listing is given in Appendix
A. Also in Appendix A is the listing of AD8M which is
the eight-mode complement of ADCR. These two routines
function together to acquire data, convert the data to
FORTRAN IV compatibility, and store the data in a floating
point array in the calling program. To use this routine,
three programs in the ADCR package must be called. First,
FREQS is called with an argument denoting the requested
frequency of conversion. The requested frequency is
rounded off to a multiple of 1., 2., or 5. between 1. x
10"3 and 5. x 10”. Then the proper arguments are calcu-
lated, passed to the GBA hardware clock, and the clock
started.

The user then calls SET to pass the number of con-
versions requested and the name of the data array. Fin-

ally, when the atomization step is to begin, the user

81

calls GO to initiate the actual data taking. The ralf-
mode programs FREQS, SET, and GO interface to the respec—
tive complementary sections of eight mode code in AD8M.
The eight-mode code and ralf-mode code were kept separate
so that the ralf-mode routines could be overlayed if
necessary.

The complex portions of the task are handled mostly
by the eight-mode code. First it must pick up the address
of the data array in the calling program and determine
the data field in which it starts. As data are acquired,
they are converted to FORTRAN IV floating variable form
using the Extended Arithematic Element and stored in the
requested data array. Another level of complexity is
added by the fact that FORTRAN IV arrays may be loaded
across field boundaries so that the code must also detect
when a data field change is necessary. The routine will
execute without missing a data point at data rates in
excess of one kilohertz. If, however, the requested con—
version frequency is excessive and causes a data point to
be missed, an error halt with l in the accumulator occurs.

A single FORTRAN IV callable RALF subroutine handles

all the functions of the positioner:
CALL POS (HPOS, VPOS, COM)

where HPOS is the horizontal destination; VPOS is the

vertical destination; and COM is the command type. There

82

are three possible command types: RESET, WAIT, and G0.
The RESET command causes the positioner to seek the
reference position defined by the vertical and horizon-
tal optical interrupters. The WAIT command causes the
computer to wait for the positioner to set its done flag
before returning from this subroutine to the calling
routine.

The remote mode GO command functions much the same
as the local mode GO command with a few minor distinc-
tions. The horizontal and vertical destinations are the
HPOS and VPOS subroutine arguments, respectively. The
software converts the arguments to the machine coordinate
system (which is l2-bit binary with the 0,0 position at
the optically defined reference point) from the user co-
ordinate system, which uses mm as units and has the origin
defined as the position where the atomizer is just below
the HCDT light path (i.e., midway horizontal travel, ver-
tical as high as possible without blocking the light beam).
Thus legal destinations are from —l2.5 to 12.5 mm hori-
zontal and from 0. to 25.0 mm vertical.

A FORTRAN IV callable subroutine written in RALF is

used to drive the DAC. The calling form is:
CALL DAC (ARG)

where ARC is a real number which represents the desired

output voltage. The unique feature of this subroutine

83

is that it may be compiled under several different com—
piler options, which can cause the argument to be inter-
preted differently (69). If no compiler option is speci-
fied at compilation time, the real argument is converted
to its binary equivalent and passed directly to the DAC
input latch. With compiler option 1, the DAC is assumed
to be a unipolar 0. to 10. V DAC, and the argument will
be interpreted as a voltage. This voltage will appear
at the DAC output. Compiler option 5 produces a similar
interpretation. But in this case, the DAC is assumed to
be a bipolar -5 to +5 V DAC.
Subroutine FLAG is also written in RALF assembly

language and is called in the following manner:

CALL FLAG (RAD, HD, SP)

where RAD, HD, and SP are real arguments with the values
of -l, 0, or +1. An argument with a value of +1 causes
the flag to be set. A -1 argument causes the flag to be
cleared, and 0 does not affect the state of the flag.

The RAD argument is used to set or clear the flag, which
causes a changover to radiation regulation of the atomizer
heating. The HD argument is used to set or clear the

flag which triggers the gas valve for the hydrogen dif-
fusion flame. The SP flag operates similarly to the RAD
and HD flags, but it has not yet been assigned a specific

purpose.

8A

The subroutine which reads the user switch register
into the accumulator is actually a function type sub-
routine and an example of a typical call to this routine

is given below.
IF (SWITCH (5.) .E0. 1.) CALL PLOT

The argument to the function subroutine indicates which
switch is being interrogated, and, if that switch is

set, the function returns a value of 1.. If it were not
set, a value of 0. is returned. In this example, the
SWITCH function subroutine is used in a logical IF state-
ment so that the PLOT subroutine is called only if switch
5 is set. The functions actuated when the other switches
are set are given in Table 2.

Although this switch register may at first seem to
serve a rather frivolous function, in actual practice it
is almost indispensible. In fact, the necessity for this
type of interaction was not recognized in the original
design. However, it became so necessary to have this
type of run-time control, this switch register was added
after the construction and testing of the rest of the
circuitry was completed.

Also used by GMBDS2 are the clock routines written
by Dr. E. R. Johnson. These routines use the 8/e's real
time clock to provide accurate time delays. The soft—
ware to drive the automatic sample dispenser is discussed

in some detail in Chapter A and will not be discussed here.

85

 

 

 

Table 2. Run-time Front Panel Switch Options.
Switch Function
0 Interrupt task execution,go to GBA monitor
1 Print run counter, position, and logic variables
on system terminal
2 Print scan variables on system terminal
3 Print sample dispenser variables on system ter-
minal
A Plot Absorbance vs. Time on system terminal
5 Unassigned

 

 

CHAPTER A

CONSTRUCTION AND TESTING OF THE VARIABLE

VOLUME AUTOMATIC SAMPLE DISPENSER

A. INTRODUCTION

.————

 

Automatic sample delivery in electrothermal atomiza-
tion atomic absorption (EAAA) measurements can improve
measurement precision and result in a substantial time
savings for the experimenter. Mechanically assisted
(76L semi—automatic (77,78),and automatic (68,79,80),
discontinuous sample introduction systems have shown that
these types of systems can improve both the precision of
volume metering and the precision of positioning the sample
over manual sample delivery. In many cases however, the
gain in precision is much less important than the pos-
sible time savings. A great number of the AA instruments
equipped with electrothermal atomizers have automated
all instrument functions except sample introduction, so
that while the instrument handles all the rigorous tasks
such as data acquisition and analysis the experimenter
is left with the mundane task of operating a hypodermic
syringe.

In our laboratory we wished to perform experiments to

86

87

map the atomic vapor in two dimensions above the atomizer.
Each of these maps would require at least one hundred
sample depositions and such maps would be required for
several elements, matrices, and atomization temperatures.
With a task of this magnitude, automation of the sample
introduction technique was imperative. Choice of the
particular design of the sample introduction system was
affected by previous experience with automatic sample
delivery systems (79,80) and the physical constraint
that the system had to fit on an optical rail already
crowded by the presence of the automatic positioner for
the atomizer cell (28). Further, to yield maximum flex-
ibility the system was designed so that the sample volume
and the sample type could be changed under computer control.
The relationship of the Automatic Sample Dispenser
(ASD) to the other components of the computer—controlled
EAAA instrument is shown in Figure 1A. The diagram has
been simplified for clarity by omitting the automatic
atomizer vapor cell positioning device. The ASD consists
of a sample turntable module, a transport and sample
delivery module, and the electronics associated with
controlling these modules. The sample turntable module
contains sample cups which allow the selection of any of
six different samples. The transport mechanism provides
vertical movement to lower the delivery system either into

the selected cup or onto the filament atomizer. The rotor

102035 0323 033.32 5:; 205335 <<<m .3 0.53.1

 

«Basra—DEORVTE
oz<
30b<20¢ 100202

«F.»
p0

 

 

 

 

 

 

 

 

 

 

3.592
32:23» VIJlfla—aumv
”:23;

 

 

L

FL

30mm: >52..qu

 

 

 

Suhm>m 02.555
EDOE< wati<mlv

 

D'f‘

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

14 _ ,.I|L HI ,IL
A;
1 . mmmzwa
025.300"—
umN_EO»(
n
TU
n
1....
0 522.282
.1“ Tlll§0mm2<¢h
m E
r r
LE.
»
<52 39.08

 

J TIL

 

 

 

 

mu 30m
~10..—
mac-=5
30.30:

89

arm rotates the delivery system between the position
where the delivery needle is directly above the sample
turntable and the position where the needle is directly

above the center of the atomizer.

B. CONSTRUCTION AND OPERATION OF THE ASD

 

1. Mechanical Components

 

The components of the automatic sample dispenser (ASD)
are shown in Figures l5, l6 and 17. The delivery system
is shown in detail in Figure 15 and is based on the
adaptation of a 200 pt manual syringe (SllOO, Gilmont
Instruments, Inc., Great Neck, N.Y.) for automatic opera-
tion. When stepper motor A is pulsed, the coupling adap-
tor C and outer syringe sleeve D rotate with the motor
shaft since the outer syringe sleeve and the syringe body
are threaded, motor rotation causes the syringe sleeve to
advance or retreat on the syringe body. The teflon syringe
plunger F is mechanically coupled to the outer syringe
sleeve D and so advances or retreats inside the fixed
position glass barrel G whenever the motor is pulsed.

An air tight seal is maintained by a teflon plate and
Vitron O-ring inside nut E. Thus, motor movements cause
the displacement of whatever fluid is contained in the
glass barrel. Teflon high pressure liquid chromatography

components H couple the .3 mm i.d. teflon tubing to the

90

Figure 15. Sample Amount Metering and Delivery system

 

 

 

/
[LP
:3

 

 

 

 

 

 

 

 

 

 

Q-nrn

 

:L'

 

 

(A) stepper motor, (8) limit microswitches, (C) coupling adaptor,
(D) outer syringe sleeve, (E) plastic nut, (F) Teflon syringe
plunger, (G) precision bore glass barrel, (H) HPLC components,
(1) glass tubing, (J) Teflon tubing

Figure 16.

91

Transport Mechanism.

(A) stepper motor, (B) counter weight, (C) 2-1/2"
gear, (D) Teflon spacer (E&F) aluminum transport
plates, (G) occluder arm for interrupters, (H)
1/2" gear, (I) 1/2", 2-1/2" double gear, (J)
optical interrupters, (K) 11" long geared rack,
(L) channeled aluminum struts, (M) base plate,

(N) delivery system and amount metering system,
(0) dovetail slide system for x, y, z positioning,
(P) vertical motor, (Q) rotation motor, (R) 1/2"

gear, (T) optical occluders, (U) optical inter-
rupters.

92

   
    
 

V

rm
3:;

           

r—_-L..___T

TOP VIEW

 

Figure 16.

agauuzcumgmazm E=:PE=.~ 51V
.waau mPQEnm mpnm>oems agave we use Acv .opaauccau mpasmm Amy .ccma =x ~ Amy
.meo =x any .LOpaacgoucP ocm~ gag Auv .mpxm opaaucgsu Amy .Lcuos sacaoum A<V

2:82 02353 £95 .2 2%:

 

 

«a.
C/

 

 

 

 

 

gw. /- ///

 

 

III‘.)‘:ILLI:]:
L_.

 

 

u-O

 

 

 

 

 

 

 

 

 

 

 

 

9“

glass barrel G via a ground glass luer joint fitting.

The glass tubing I is epoxied to the teflon tubing J so
that the delivery needle is physically rigid. Micro-
switches B detect the limits of the fill or deliver opera-
tions when the outer rim of the syringe sleeve D actuates
them.

A second teflon delivery needle system with a .5 mm

i.d. was constructed from HPLC components (Anspec, Ann
Arbor, MI). Also tested in this work was a stainless
steel needle (.5 mm i.d.) which had a luer Joint fitting
and was easily interchangeable with the teflon needles.
A second stainless steel delivery system was constructed
by epoxying the final l/2 inch of a 10 Hi syringe (Hamil-
ton Co., Reno, NV) to the end of a stainless steel needle
identical to the needle described above.

The assembly which transports the syringe between
the delivery position (above the atomizer) and the sample
loading position (at the sample turntable - Figure 17)
is shown in Figure 16. The ASD is secured to the optical
rail by an aluminum base plate M and two angle brackets
fastened to the base plate. Also fastened to the base
plate are two aluminum struts L which are channeled so
that they form a track in which the transport base plate
F can travel. The 1/2 inch gear J is fastened to the
shaft of the stepper motor 0 which is mounted on, but

behind, the transport base plate F. This geardrives

95

against the larger member of a double gear system I whose
smaller member rides against the geared rack L. This
geared rack L is fastened to one of the supporting aluminum
struts so that the rotation of the motor shaft is trans-
duced into a linear vertical motion with a five to one
mechanical advantage. These aluminum struts are 16 inches
long and spaced 3 inches apart. The arrival of the trans-
port base plate F at the top of its travel, the level of
the sample turntable, or the atomizer is signaled when

the appropriate optical interrupter J is activated by the
interrupter arm G.

When the delivery head is at the top of its travel it
can be rotated between the delivery position (above the
sample atomizer) and the load position (above the sample
turntable) by pulsing stepper motor N (see top view -
Figure 16). Gear R on the stepper motor shaft intermeshes
with gear C and drives the delivery head between the two
positions which are defined by optical interrupters P
and interrupter occluders Q. Also shown in the top view
in Figure 16 is an alternate view of the channeled struts
and the gear system that drives the vertical movements.

The sample turntable operation is quite straight-
forward (Figure l7). An aluminum superstructure supports
both the stepper motor A and the axle on which the sample
turntable is fixed. Rotation of the stepper motor shaft

is coupled to the rotation of the sample turntable by a

96

l/2 inch gear D and a 2-1/2 inch gear E. Cup position
zero is defined when a small aluminum tab (not shown)
attached to the sample turntable breaks the beam of the
optical interrupter F. Each of the eight sample turn-
table positions may contain a five ml sample cup. In
normal operation cup zero is used to store wastes and

cup one contains water for rinsing the syringe.

All of the gears, axles, ball bearings, and the geared

rack were purchased from PIC Design (Ridgefield, CT).

2. Electronic Hardware

The ASD is controlled by a PDP 8/e computer which has
16K of random access memory, a real time clock, and an
extended arithmetic element. Peripherals include a DEC
RKOS cartridge disk, a Sykes 7000 dual floppy disc, an
ADDS Consul 980 graphics terminal, an LA—3O decwriter and
a Heath EU 801 Computer Interface Buffer. This computer
system also controls an EAAA instrument (81) in addition
to the ASD. The ASD interface was constructed on a gen-
eral purpose wire wrap board using standard LSI and MSI
integrated circuits and communicated to the computer
through the Heath interface buffer.

A block diagram of the electronic hardware specifi-
cally involved in controlling the ASD is shown in Figure
18. In a typical sequence the computer latches out a

command into the interface, the interface decodes the

386:: 35.385 To? 9: mo 5235 v.88 .2 953...

 

 

 

 

..-...7o hm¢¢=»l*||||||4
.pcnmmnach .-.--;w_o o_<¢ow|IIIIJ

 

 

 

 

inns-.. ~ormo» .

 

 

x .-....xm= ppaxu
>a -....-_xm: 44:; Till

2°
nh’
cc:

 

 

 

 

 

97

 

 

capo: ------ so hum;

 

 

capo: ooonuc ~o pro—c

 

 

capo: cannon
buccah no ocwn

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Dumnu .
m\’n~uo ,
”WWW“ AHmufimmsz $5.28

(W

 

 

 

 

 

 

 

 

 

 

 

98

command and pulses the appropriate stepper motor drive
circuit. This drive circuit channels power from the
stepper motor power supply down the stepper motor power
lines in the proper sequence causing the motor to step.
The motor is stepped in this manner until the desired
movement is completed. The interface recognizes a task
as complete when the motor movement causes a change of
state at the appropriate optical interrupter (OI) or micro-
switch (uSW) (this type of interaction is denoted by a
dashed line in Figure 17). The syringe motor may also
be stopped when a specified number of steps have been
executed. The interface also contains error checking
logic so that if the execution of a requested command would
cause a movement that is self-damaging, then that move-
ment can not be executed. An example of this type of
command would be a request to rotate the sample turntable
when the syringe needle was in one of the sample cups.
Either the completion of a legal task or the request of
an illegal movement causes the interface to signal the
computer that the task has been acted on. The computer
may then interrogate a status register which contains
flags indicating the legality of the last requested move-
ment and the status of each of the position indicating
optical interrupters and microswitches.

The syringe and turret stepper motors are models

202215D200—Fl.6 and lO-20l3DAO-F75, respectively, Sigma

99

Instruments, Braintree, MA. The rotor and transport step-
per motors are models 23D6102A and 23D6306A, respectively,
Computer Devices, Santa Fe, CA. The stepper motor drive
circuits were constructed as recommended by the Sigma
Stepper Motor Handbook (82). A filtered, but regulated
+25V power supply was built in-house to provide the raw

power for the stepper motors.

3. Software

 

Three Fortran IV callable assembly language subroutines
control all the actions of the ASD. The first, INIT,
safely moves the ASD from whatever position it may be at
power up to the state in which the syringe is empty and
positioned above the sample turntable sample cup zero.

The second, TUNE, moves the sampler so that the delivery
needle is at atomizer level and then waits for the operator
to fine tune the delivery position on the atomizer. When
this task has been completed the ASD returns to the initi—
alized position. The third routine, SAMP, is used to re-
quest the delivery of a specified volume of sample solution
onto the atomizer. SAMP, INIT, and TUNE all contain a

list of Job words in sequential memory locations, and each
of these Jobs words corresponds to one or more motor move-
ments. Table 3 gives the bit assignments for both the

Job word and the status registers. For example, the job

word list for INIT contains a Job word for each of the

 

100

Table 3. Bit Assignments for the Status Register and
Job Word Register.

 

 

Accumulator Bit

Status Register

Enable Register

 

OKOCONQU'I

Illegal Vertical
Illegal Rotor
Illegal Syringe
Illegal Turret
Syringe Empty
Syringe Full
Rotor Left

Rotor Right
Vertical Top

Vertical Bottom

Turntable at Cup 0

Interrupt Enable
Top/Purge
Up/Down

Vertical Enable
Left/Right

Rotor Enable
Turret MSB
Turret Bit 2
Turret Bit 3
Turret LSB

Syringe Special
Instruction Flag

 

 

 

101

following movements: ascend to the top (7A00), rotate
the syringe over the sample turntable (0336), send the
sample turntable to cup one (0302), descend to the sample
turntable (0400), empty the syringe (0100), and finally
ascend to the top position (7A00).

The Job word list for SAMP is much lengthier and has
three separate entry points which correspond to the purge,
load, and deliver functions. To decide which functions
are necessary, the software keeps track of the sample
number in the syringe and the volume of the solution
contained in the syringe. The purge function, which
involves the rinsing of the syringe twice with distilled
water, is executed only when the sample requested is dif-
ferent from the sample presently in the syringe. The load
operation, which entails refilling the sample syringe,
is executed only if the sample in the syringe is the same
as the last one requested but the amount is insufficient,
or when the last function executed was a purge operation.
If the request involves no sample changing and the amount
in the syringe is sufficient, then only the delivery func-
tion is executed.

When the proper entry point has been determined, this
routine picks up individual Job words, sends them out to
the interface, and waits for the interface to set the
done flag. When the done flag is set, the routine reads

the status register and halts if any of the illegal flags

102

are set. If the command was not illegal, then the move-
ment is assumed complete, and the next Job word is fetched
and acted on. This process continues until the requested
delivery is completed. The lone exception to this type
of operation is the special case of syringe instructions.
These instructions require an additional information word
which indicates the number and the direction (i.e., load
or deliver) of syringe motor pulses to be executed.

A request for a sample delivery from a Fortran routine
is in the form of a subroutine call with two arguments
one which indicates the sample number, the other which
indicates the amount to deliver in p2. The only restric-
tion is that before the first delivery is requested the

INIT and TUNE routines must have been properly performed.

C. EVALUATION OF PERFORMANCE

 

The important criteria to be considered in the evalua-
tion of the ASD are the delivery accuracy, precision,
and inertness toward the sample solution. The accuracy
and precision of the delivery and thus the total AA mea-
surement precision is quite dependent on the mode of opera-
tion and the particular delivery needle used. The most
obvious conclusion to be drawn from the study of precision
given in Table A is that an entrapped air pocket is quite
detrimental, especially for needles with small diameters.

This entrapped air pocket is formed in the precision glass

103

Table A. Measurement Reproducibilities of Various

Delivery Modes.

 

 

 

 

 

Needle Material i.d. stol TRSD2 %RSD3 %RSD”
St. Steel .5 “.3 12.A 2.8 120
St. Steel <.2 2.2 * 2.2 *
Teflon .5 1.8 100 1.5 65
Teflon .3 3.“ 95 3.3 23

RSDl Atomizer not heated during delivery, no entrapped

air space in syringe.

Atomizer not heated during delivery, entrapped air
in syringe.

Atomizer heated during delivery, no entrapped air

space in syringe.

Atomizer heated during delivery and entrapped air

space in syringe.

Deliver totally ineffectual.

10“

barrel between the syringe plunger head and the surface

of the sample solution whenever the ASD initially fills
itself. The only way to prevent this air bubble is to
manually charge the syringe barrel with sample solution
before the fill function is executed. When the ASD is
operated with an entrapped air pocket the inner diameter
of the delivery needle significantly affects the delivery
precision. This is explicable if one considers that at

the time of delivery the sample solution is subjected to
capillary forces of the braided graphite atomizer and the
syringe needle. Also the surface tension of the solution
and the hydrophilic or hydrophobic nature of the syringe
needle material affect the reproducibility of the delivery.
Heating the atomizer to desolvation temperature during the
delivery enhances the ability of the atomizer to compete
for sample solution and when an entrapped air pocket is
present this enhancement is very detrimental to delivery
precision. In this part of the study no scavenging of

the analyte by either the stainless steel or teflon needles
was observed.

Table 5 shows the results of long term tests on the
precision of sample delivery for three individual atomizers.
In Figure 19 the graphical representation of one of these
experiments demonstrates that no perceptable scavenging
of the 1 ppm cadmium by the syringe needle or precision

glass barrel occurred during the duration of this

105

 

 

m.m
N.m
m.H
o.m
:.m
m.m
m.m
w.m
m.m

om.mH
wH.mH
:H.NH
No.mH
mo.mH
mm.HH
ow.mH
mH.mH
mm.mH

m.H
m.H
>.H
~.H
m.H
m.H
~.H
n.m
:.m

:m.ma
oo.mH
wa.mH
H:.NH
H:.NH
om.NH
Hm.mH
wH.mH
om.mH

m.m
>.m
m.m
m.m
m.m
:.m
~.m
m.m
=.m

mm.NH
:H.NH
mH.mH
:o.ma
:w.aa
mm.HH
mm.HH
mN.HH
mo.ma

owmlamm
ommlawm
owmlflzm
onmlaom
oomIHmH
owHIHmH
omHIHm
owIH:
Gala

 

0mm

.mn< vopmpwoucH

m

Qmm .mm< UmpmhmmpCH

m

0mm

.mp< Uopmpwoch

H

gonadz cam
.02 noNHEOp<

 

 

.spfififieapm some wees

.m mHQmB

106

.-
..
,—

-Ir-
,-

a).

,—
..
b
-r-
,—
—

JJ'LLl'lll'lJ
lll'IIj

0.000

I.H%
0

'lll'lil'lll
'ITIIIIII

J'JII'LII

[IT‘IIIITFIIIUI

 

 

1.500”-

‘I’

7. 3x
moz<mmomm< ome<momeza

0.000

RUN NUMBER

-2

X10

Long Term Stability

Figure 19.

107

experiment (approximately 5 hours). What small drift
there is in this data may be attributed to aging of the
atomizer which changes the sensitivity of the measure—
ment. These results were acquired using the small diameter
stainless steel needle, no entrapped air space, and de-
solvation heating during delivery. Results with the
other delivery needles under similar conditions were quite
similar to those shown.

The ASD was also tested for the linearity of volume
delivery with the smallest diameter steel needle and 1
ppm Cd. The results for sets of six requested deliveries
of volumes between 1.0 and 2.0 pi are shown graphically

in Figure 20. The data obey the following equation:

y = (.60i.02) x + (.ou:.o3)

where y is the integrated absorbance in arbitrary units
and x is the delivery volume requested.

When automatic sample changes are required it is
necessary to operate the ASD with an entrapped air pocket
to prevent possible solution mixing effects. In the multi—
ple sample mode the ASD showed no memory effects when it
was programmed to rinse twice with distilled water and
twice with the new sample solution between samples, how-
ever, the precision was degraded to typical values of 10

to 15%.

108

    

1.500m‘
LU ..
L) -_
I: -.
< -r
m m.
01' J-
O «-

00
m ..
< __
C) ‘F
LL. -—
l— '1-
< a...
O: _....
(D' ..
UU __
F‘ --
Zi .-

H
1

 

0'000LH+HH+%HH+HH%+HH+HHH+HHHflHHHHI
0

.000 2.500
VOLUME (UL)

Figure 20. Volume Delivery Linearity

 

109

D. CONCLUSIONS

 

The automatic sample delivery system can precisely
and reliably deliver pl sized samples onto an electro—
thermal atomizer, and is suitable for long term computer-
controlled experimentation. Although the ASD was only
tested for 1 ppm Cd, with proper treatment of the surfaces
in contact with the sample solution (1,10), it is expected
to perform equally well for solutions of other elements

and more dilute solutions of cadmium.

 

——_-E—t'—————— _.-_.-_—.——.~ -____.i_,, _-

 

CHAPTER 5

SIMPLEX OPTIMIZATION STUDIES

ON THE EAAA INSTRUMENT

The attractiveness of on-line optimization of experi-
mental parameters was pointed out in the introduction to
this thesis, and the small amount of work which deals with
tfluaapplicaticwiof simplex optimization techniques to chemi-
cal problems was reviewed in Chapter 1. The work in this
chapter demonstrates the ease with which the simplex
optimization technique was adapted to the optimization
of experimental parameters in the EAAA instrument and also
the value of this approach as a tool to help solve general
EAAA analysis problems.

This chapter contains a description of the Nelder and
Mead simplex paradigm and the software used to implement
this optimization technique on the EAAA instrument. The
cadmium integrated absorbance as a function of horizontal
and vertical displacements of the observation window away
from the atomizer was recorded and the on—line simplex
optimization technique was tested and characterized on
this known response surface. The cadmium results were

compared to results of optimizations of the manganese

110

111

integrated absorbance in similar experiments where the
shape of the response surface was not totally character-
ized. A four parameter, on—line optimization of the
cadmium integrated absorbance as a function of the sample
amount, time delay between sample runs, and the horizon—
tal and vertical displacements of the observation window
away from the atomizer, was also carried out. An on-line
optimization with a complex response function, which was
to minimize analysis time and maximize the integrated
absorbance signal of cadmium, was also attempted. An
attempted optimization of the measurement precision in a
semi-automated, multicomputer system is also described.
The optimization of the time limits of the integration

of the cadmium signal was carried out off-line on pre-
recorded data. Finally, some general observations about
on-line simplex optimizations and some suggestions for
improving the performance of the simplex algorithm in the

presence of noise are given.

A. INTRODUCTION

 

A brief introduction to simplex nomenclature and the
simplex algorithm is necessary before a detailed discus-
sion of the experimental work may be attempted. A system
is any process or device that operates on an input or
inputs to give an output or outputs. The response is a

system output, or combination of system outputs, that

112

is chosen as the entity to be optimized. Factors are a
selected subset of the system inputs chosen by the experi-
menter to be varied in order to maximize the chosen res-

ponse.

1. Example

 

In order to define factor, system, and response in a
more familiar environment, a simple chemical reaction
A + B + C is used here as an example. There are several
experimental variables which, if they affect the chosen
system response, may be termed factors. Possible factors
in this system include: the mole fraction of A, the mole
fraction of B, the temperature, the pressure, the reaction
time, the presence or absence of catalysts, inhibitors,
light, etc.. The particular system output chosen as the
response is entirely at the experimenter's discretion.
Some reasonable choices for this example include the fol-
lowing: percent yield, time of reaction, amount of useful
by-products, input energy costs, ease of cleanup, etc..
In real chemical systems the response chosen is often a
weighted combination of several of these responses.

A plot of the response value as a function of the
values of the selected factors yields a response surface.

The response surface is conveniently described as the system

 

response output under every conceivable combination of

input values of all the chosen factors, when all other

113

variables are held constant. In our example, if we chose
temperature and pressure as factors, and the percent
yield as the response, then the response surface might

be shaped like the response surface represented in Figure
21. Figure 21a is a pseudo-three-dimensional plot of the
response surface. In it the optimum yield is located

at the "top of the hill". This same response surface is
also shown in a two-dimensional contour plot in Figure 21b.
In each plot iso—yield lines connect the combinations

of temperature and pressure that give identical yields.
Note that the "top of the hill" in Figure 21a is located
inside the smallest of the ellipses in Figure 21b. The
simplex algorithm is an optimization technique that uses
a set of rules to move a geometric figure located on the
response surface toward the response optimum. The geo-
metric figure is called a simplex and has n+1 vertices

where n is the number of factors.

2. Comparison of the Simplex Method with Other

Optimization Techniques

In order to establish the value of the simplex opti-
mization technique, it should be compared to the other
commonly used optimization methods. The most widely
used optimization technique is the iterative method.

In this method, all variables except one are held constant,

11A

 

 

TEMPERATURE

BBDSSBHd

 

 

 

VI

TEMPERATURE

Figure 21. A Typical Response Surface

013M

115

and the chosen variable is adjusted until the response
optimum is found. Then, a second variable is chosen,
and its value adjusted until the response is again
optimized. This process is then iterated until a change
in the value of any of the variables results in a decrease
in the system response. The failure of this technique
to find the true optimum is demonstrated by the example
response surface shown in Figure 21b. When point A is
reached, the adjustment of either the temperature or
pressure yields a decrease in the yield. To visualize
the failure of the iterative optimization method to find
the true optimum imagine a mountain climber who must
climb under the following restrictions. First, every
move must result in a net gain in altitude, and second,
he may move only directly north, south, east, or west.
The particular mountain that this climber must ascend
has a ridge running in a northwesterly direction from
the base to the peak. When the climber finds himself on
this ridge, movement south or east will be downhill and
movement north or west will be down the side of the ridge
and will result in a net loss in altitude. Under these
restrictions, which are the same as those imposed by the
iterative optimization technique, the climber will never
reach the mountain top. This type of ridge is known to
occur frequently in chemical systems (A7).

To find the true maximum yield, in the chemical

116

example, both parameters must be adjusted at the same
time. Since the simplex has three vertices in a two dimen-
sional optimization it can change the direction of its
movement to any of the points of the compass and so it
can move diagonally toward the true optimum.

The most rigorous optimization techniques involve a
patterned testing of the entire response surface. This
is impractical for complicated systems that have more than
two or three factors since these techniques require an
inordinately large amount of experimental data to be

collected.

3. The Nelder and Mead Simplex Algorithm

 

The moves of the Nelder and Mead simplex algorithm
(36), which was used exclusively in this work, can be ex-
plained with the aid of Figure 22. For two factors the
simplex has three initial vertices, which are tested and
ranked according to the response at each vertex. The
initial vertex is the triangle BNW, where B is the best
vertex, N the next best, and W the worst vertex. Vertex
W will be discarded by the simplex algorithm, and a new
vertex for the simplex will be generated somewhere along
the line defined by the old vertex W and the centroid of
the opposite face of the simplex. The first test is of
the response at the point R, which is an equal distance

from the centroid as point W but on the opposite side of

 

 

117

«Lamvu

 

 

 

sepraeep< xmpaewm eeecseez mes co me>ez m_eemea we» .NN
H epeemea>
z a
so
u
6
m m

>l§L°r-.Dr-0 N

 

118

the line BN. If the response at R is greater than the
response at B, then an expansion to point E, which is
twice the distance from the centroid as vertex R, is at-
tempted. If this response is better than R, then the new
simplex will be the triangle NBE.

If the response at R is between the responses of B
and N, but greater than the response at E, the new simplex
will be RBN. If the response at R is less than at N but
greater than at W, a contraction to the BNCr simplex is
executed. If the response at R is worse than the respon-
ses at N and W then the response at point Cw will be
evaluated and the new simplex will be BNCw. The last
two types of moves (contractions) are executed only if
the response at this contraction is better than at W.

In the original Nelder and Mean algorithm, if all of
these attempted moves failed to achieve a better response
than W, a more massive contraction is executed. In this
work, the next worst vertex N of the original simplex is
reflected throughout the centroid of the simplex face
opposite it. A second modification of the Nelder and Mead
algorithm was necessary to deal with instrumental errors.
If a vertex had been retained by the simplex for more
than n+1 moves, it was re-evaluated and the old response
was replaced by the new response. In the EAAA instrument
occasionally a drop of solution would cling to the exter-

ior of the delivery needle during the refill operation

119

and would be delivered to the atomizer along with the
requested sample, and this would result in an erroneously
high evaluation of the response at that point. If this
response was not re-evaluated, the simplex would rotate
about this point in a manner not unlike a dog chasing its
tail. Note that most researchers have averaged the old

and new values of the response to get the response value

to be used in the simplex algorithm (A7,52,59). This action
is best suited for averaging out random error; the replace-
ment operation is best suited for reducing systematic
instrumental errors, since with averaging it might take

an unreasonable amount of testing to overcome the effects

of the one incorrect response evaluation.

B. SIMPLEX SOFTWARE

 

The fundamental FORTRAN IV routine used to implement
the simplex algorithm on the EAAA instrument was adapted
from a routine written by Mr. lartin Joseph and Dr. Eric
Johnson in our laboratories (85). Some extensive modi-
fications were necessary to allow this routine to be em-
bedded into the EAAA routines and to allow the desired
flexibility in the selection of the variables to be opti-

mized and the calculation of the response.

120

1. Initiation of a Simplex Optimization

The user may enter the Simplex mode of operation by
specifying the :SMP command, to the GBA monitor (subrou-
tine INPUT). Upon receiving this command the program
requests the user to select the number and the names of
the variables to be optimized. At this point the user
may chose 2 to A of the 13 possible experimental param-
eters listed in Table 6 as factors for the simplex optimiza-
tion. The selection of these variables must be influenced
by certain practical considerations for the simplex to
operate properly. The software, however, will accept any
combination of these variables. The rules for making
acceptable choices of factors in this particular applica-
tion are discussed later in this chapter. The user is
then prompted to enter the upper limit, lower limit, and
the desired convergence precision for each variable.

For example, the lower limit for the variable DEST (desol-
vation time) might be set at 0. s, since negative times
are meaningless. The upper limit might be set at 60. s

to prevent the simplex algorithm from making a response
test at a desolvation time that is so long that it is
experimentally impractical. The value entered for the
desired precision relates to the detection of the con—
vergence of the simplex on the optimum. When all the
vertices of the simplex have values for each factor

which are separated by less than the specified precision

121

Table 6. Experimental Parameters that May be Optimized
With the Simplex Routine.

 

 

 

Variable Description
DEST Desolvation time
ASHT Ashing time
ATMT Atomization time
DELT Delay time between sample runs
DESP Desolvation power
ASHP Ashing power
ATMP Atomization power
AMNT Amount of sample delivered
HPOS Horizontal location of observation window
VPOS Vertical location of observation window
FREQ Data acquisition rate
INTST Number of the data point at which the
integration is begun
INTSP Number of the data point at which the

integration is stopped

 

 

122

for that factor, the simplex is halted. The user is

then requested to enter the factor values for each of the
initial simplex vertices. There is also a choice of the
optimization mode, and the options include: limited inte—
gration mode (this mode must be used if INTST and INTSP

are chosen as factors to be optimized), whole integration
mode (normal mode), or peak absorbance. The user may

also specify more than one sample run to be taken per
response evaluation. Finally, one can enter an itera-

tion limit, which will cause the simplex to halt after

this specified number of moves even if it has not converged
on an optimum. This last option is quite useful if it is
desired to perform several optimizations under batch stream

control as described in Chapter 3.

2. Run—Time Options

Some of the option variables function in the simplex
mode of operation similar to the manner they function in
the normal mode. If the LPT option is set, an output of
the initial simplex parameters, as well as a running
account of the progress of the simplex, is output to the
lineprinter. The PLT option causes an absorbance versus
time plot to be output to the system terminal. This has
the same appearance as the plot given in the normal mode,
but uses a separate set of subroutines because of overlaying

requirements. Switch 0 on the front panel of the general

123

interference box still serves the function of aborting
the task and exiting to the GBA monitor. Switch A causes
an absorbance versus time plot even if PLT was not re-
quested. The file option variables in the simplex mode
pertain to file output formats entirely different from
those in the normal mode. One file format outputs the
iteration number, the factor values, and the response at
the vertex just tested. This file is useful for the crea-
tion of plots of the movement of the simplex and plots

of the progress of each individual variable. The other
file output keeps a record of all of the response values

and is useful for archival storage since it is short.

C. TESTS OF THE SIMPLEX ALGORITHM ON A KNOWN RESPONSE

 

SURFACE

On-1ine optimizations of experimental parameters
for a particular instrument must deal with noisy response
surfaces and, if parameters other than those parameters
being optimized change, the algorithm must also deal with
shifting response surfaces. The success of simplex
optimizations is dependent on the characteristics of the
system being optimized. Thus it is desirable to test the
simplex on a "standard" response surface, where the global

optimum is known.

12A

1. The Cadmium Integrated Absorbance Response Surface

Two-dimensional mapping experiments of cadmium absor-
bances (83) yielded a known response surface for one
particular optimization that might be attempted on the
EAAA instrument. The optimum integrated absorbance in
the response surface was assigned an arbitrary value of
100, and the iso-absorbance lines were drawn to define
the regions where the integrated absorbance decreased by
10% increments from this optimum value.

Figure 23 shows the response surface and several
starting Simplexes used in the optimization of the cad-
mium integrated absorbance signal. The global optimum
is located near 0. mm vertical displacement and 0. mm
horizontal displacement. The response drops off rapidly
in all directions (the slope of the response surface is
inversely proportional to the distance between the iso-
absorbance lines). The slope for vertical displacements
is less severe than the slope for horizontal displacements.
Also the pointed intersection between the iso—absorbance
lines near zero horizontal displacement are indicative of
a sharp ridge in the response surface. The irregularities
in the iso-absorbance lines above 12 mm vertical displace—
ment suggest a non—smooth surface. Also, since source
flicker noise (hollow cathode lamp flicker) is a signi-
ficant source of imprecision, the signal-to—noise ratio

should be poorer for the regions bounded by iso-absorbance

125

 

 

 

 

 

 

24
2| '-
l8-

V
E
R I5-
T
I
(3 I2? ‘
A
L 9..

(mm)

6..
3..

 

 

 

 

 

 

 

HORIZONTAL (mm)

Figure 23. Cadmium Integrated Absorbance Response Surface with
Assorted Starting Simplexes.

126

lines which indicate low cadmium concentrations. An addi-
tional source of noise in the upper regions of the vapor
cell is caused by the breakup of flow laminarity as the
sheath gas begins to encounter the surrounding atmosphere
near the top of the atomization cell. The slight asym-
metry of the surface in the upper regions is probably
indicative of the prevailing wind currents in the labora—

tory.

2. Methods of Monitoring the Progress of the Simplex

 

The movements of the simplex which used triangle G in
Figure 23 as the starting simplex are shown in Figure 2A.
Note that it initially moves rapidly toward the horizon-
tal center of the vapor cell and down toward the atomizer
with two expansions. Then two contractions on the only
vertex remaining from the original simplex cause a rapid
approach toward the horizontal optimum. After one more
expansion in the general direction of the global optimum,
the algorithm causes a series of reflections and contrac-
tions until an optimum very near the true global optimum

is found. This type of plot is instructive, but somewhat

confusing, and applicable only to two-factor optimizations.

Single parameter progress plots such as Figures 25, 26a,
and 26b, are more easily analyzed. These plots show both
the successful and nonsuccessful moves of the algorithm

for a single factor at a time, and thus may be used to

7.0"

5.0--

amndofl10<

3.0--

:o-o-c-t-umov

:3:

2.0--

 

0.0
-4.0

Figure 24.

127

 

 

 

I l J
-3‘.o -230 -1.0 0.0 1.0
Horizontal Position (m)
The Moves of Simplex Optimization 6.

RESPONSE

 

128

 

 

 

 

TE’
32
3E
5.‘j:'
2C
4-'::'
j:
3.-EE—
3E
20-—
' IE
Li}
SE 1
00"!!! IIIIIJIJIJIIIJIIIIIIJ
- ITWIIIITIITIWWDIIIIIIIIT—l
1 5 10 15 20
SIMPLEX N0.
Figure 25. Progress Plot of the Cadmium Integrated Absorbance

Response in a TWO Parameter Optimization

129

mucnncoma< voumgmmacm savanna mo :ovumu?ewaao Lmumsmgca
03H m c? co*u*moa PopcoNALo: use co+uwmoa Puuvugm> yo muopa mmmcuoca

.02 medsz .02 qudsz
on a“ a. m a e~ a” a. m

Give.

0'-
on.
I .N.

O”-

(NH) NOIlISOd WVINOZIHOH

.eN seemed

.o—

.(HN) NOIlISOd TVDILHBA

130

monitor the progress of an optimization of any dimension-
ality. These plots were created by slightly modifying

the optional simplex mode output file so that it could

be used as an input file to the program MULPLT on the
CEMCOMGRAF facility. The progress curves are generated

by first plotting the value of the parameter of interest
at each vertex of the starting simplex and then connecting
the points with lines. A line is then drawn from the last
plotted vertex value to the value of that parameter at
each attempted move. The line drawing process is then
repeated using the value of the parameter at the last
successful simplex move as the new origin for further
lines to values at the next attempted move, etc..

Figures 25, 26a, and 26b indicate the movements of
simplex optimization A in Figure 23. Some interesting
conclusions about the optimization process may be deduced
from a careful consideration of these figures. The move-
ment of the response (integrated absorbance) is shown in
Figure 25. For the initial moves, the improvement in the
response is rapid, because the simplex is large and lo-
cated on a steep gradient of the response surface. How-
ever, when the simplex is smaller and on a more level
portion of the response surface, the changes in the response
are quite small. It seems quite incongruous that the
response seems to degrade from its peak value at simplex

7 to a somewhat smaller value when the simplex is halted.

131

This could be due to several factors. It is possible that
this vertex was actually retained in the final simplex
and only the other two vertices moved. However, inspec-
tion of the data in the data file that recorded the simplex
moves revealed that this had not occurred. It is also
possible that there might have been a change in some
parameter not being tested. Atomizer aging, which can
cause a loss in measurement sensitivity, could cause a
shift in the response surface and this process might
yield a progress plot like the one obtained. Indeed this
may be part of the cause of the degradation in response.
But, as will be shown in the analysis of the vertical
parameter progress plot, this type of sawtooth plot is
also indicative of the approach of the simplex to an
optimum that lies on one of the parameter bounderies.

If the progress of the vertical variable, Figure 26a,
and the progress of the horizontal variable, Figure 26b
are compared, it can be seen that the simplex converged
on the horizontal optimum much faster than it did on the
vertical optimum. This cannot be attributed solely to
the fact that the response surface has a larger slope in
the horizontal variable than in the vertical variable.
First, convergence for a single parameter is dependent on
the size and orientation of the initial simplex. Second,
because the size of the simplex moves are limited and any

attempted moves outside the designated boundaries are

132

assigned very unfavorable responses, the simplex tech-

nique has difficulty converging on an optimum that is

located on a boundary.

3. Comparison of Optimizations with Different

 

Starting,Simplexes

 

The results of starting the optimization at each of
the starting simplexes shown in Figure 23 are given in
Table 7. Optimizations A1 through A6 all shared the common
starting simplex A. On a noise free surface the algorithm
would make exactly the same movements each trial, and
the optima found would be identical. Experimental noise
causes six different optima to be located, but all of
these are in the immediate vicinity of the true global
optimum. Although the least efficient optimization (A5)
took nearly 60% more moves than did the most efficient
(AA), both processes took considerably fewer experimental
observations than any mapping experiment that could be
expected to find an optimum as near the true global op-
timum.

The D, E, and F initial simplexes all failed to con-
verge on the true global optimum. For optimizations E
and F, the reason is that the initial simplex was located
in a region of the response surface that has an unfavor-

able signal-to-noise ratio, and this caused the simplex

133

Table 7. Results of Optimizations of the Horizontal and
Vertical Positions for Maximum Cadmium Integrated
Absorbance.

 

 

 

HPOS VPOS
Optimum Optimum Iterations
Optimization (mm) (mm) Required
A1 - .29 .21 1A
A2 .09 .A9 1A
A3 .19 .28 1A
AA .0 .27 13
A5 - .13 .5A 23
A6 .13 .27 18
B .03 .67 16
C .22 .67 16
D - .19 1.A 8*
E -1.8 1A.O 8*
F - .80 13.5 12*
G .01 .5A 11
A6' .20 .21 25
G' .1A .29 1A

 

 

*Failed optimization due to noise on the response surface.

13A

to contract prematurely. Noise causes more contractions
than expansions, since contractions are statistically
favored by the simplex algorithm. It is possible, but
highly unlikely, that there are indeed some small local
optima which the simplex found in this region. These
optima would have to be quite small or they would have
been visible in Figure 23. The reasons for the failure
of optimization D to converge on the true global optimum
are more subtle. First, the initial simplex was created
so that its size, location, and orientation made it dif-
ficult for the simplex algorithm to move it toward the
true optimum. Because it was very close to the vertical
boundary and oriented so that its first moves were con-
tractions, the simplex fell prey to instrumental noise
and contracted before it could re-orient its movement in
the direction of the optimum and make significant progress.
In fact, all the eight moves of this optimization were
contractions.

Optimizations B, C, and G all converged in the vicinity
of the true global optimum in a reasonable number of move-
ments. Along with optimizations Al through A6, these
optimizations demonstrate that a judicial choice of a
starting simplex involves the placement of the simplex
in a region of the response surface that has a significant
slope, low noise, and is not near any variable boundaries.

Another form of failure for the simplex, which was

135

not observed in these tests, is the loss of dimension-
ality of the simplex on a response surface ridge. If the
simplex were to somehow locate two of its vertices on the
response ridge (i.e., both vertices have a horizontal
value near 0. mm), the algorithm would cause a series of
reflections and contractions on the one vertex not on the
ridge until the simplex resembled a straight line. When
this happens, the movement of the simplex is restricted
to movements in the direction defined by this line, and
the true optimum may never be located. None of the op-
timizations in this study showed this failure mode, be—
cause although there is a ridge in the response surface,
there is still a consistent slope in the variable which
parallels the ridge. (In this case the change in response
for vertical displacements along the vertical ridge is
consistent and large enough to prevent the loss of dimen-
sionality.)

The software caused the simplex to be halted when the
range of values for each parameter at all the vertices
was within a specified precision. In these optimizations
the precision specified for both the horizontal and verti-
cal variable was 0.5 mm. If this restriction is lifted,
the optimum located is closer to the true global optimum.
Optimizations A6' and G', which are continuations of
optimizations A6 and G, demonstrate this point. In these

optimizations, the simplex was allowed to continue until

136

the operator could no longer detect any significant move-
ment of the simplex. The optima located are significantly
closer to the true global optimum, and the range of values
for the horizontal factor under these conditions reaches
0.03 mm and 0.32 mm for optimizations A6' and 0', respec-
tively. Similarly, the range of values for the vertical

factor is 0.13 mm and 0.1A mm.

A. Optimization of the Integrated Absorbance of

Manganese

 

The results of studies on the optimization of the
horizontal and vertical position for maximum Mn integrated
absorbance are shown in Table 8. The vertical concentra-
tion profiles for manganese (8A) and cadmium (28) show
that the atomic population of manganese decreases with
the increasing distance above the filament more rapidly
than the atomic population of cadmium. For this reason,
the initial simplex was located rather near the atomizer
surface. (It was actually the same as simplex A in Figure
23.) Because of the short half-life of atomic manganese
in the sheath gas, its response surface probably has very
similar slopes for horizontal and vertical displacements.
For optimizations A and B, in Table 8, the simplex converged
on an optimum a bit higher and farther in the negative
horizontal direction than the cadmium optimizations, but

convergence was achieved in a comparable number of

137

 

 

 

m Ama.v m. A:~.V m. ma ms. mm. n a
H Amo.v m. Ao:.v m. mm mm. em. 0
H Amm.v m. Aoz.v m. MH H.H m.HI m
H Aom.v m. Amm.v m. mH am. am. I <
ASEV AEEV
mom> momm
coapmsam>m AmoNHHmopv AEEV AEEV coapmn
omQOQmom oopmmsvom oogfisvmm Essfipdo Esefipdo IHEHuqo
\mCOfipmppr :0fimfioogm wCOHpmpmuH mom> momm

 

 

.omocmmcmz pom mocwnLOmn< coumpwmQCH Ezefixmz

pow mCOHuHmom HwOHupo> ocm Hopcomfipom mo mCOApmNHEHqu on» no mpazmmm

.m magma

138

iterations. Perusal of the progress of optimization C
indicates that it contracted too much initially, and noise
problems made it difficult for it to find the true optimum.
In optimization D, three sample determinations were aver-
aged for each response evaluation. This should decrease
the susceptibility of the simplex algorithm to random
noise. After 15 iterations the atomizer suffered catas-
trophic failure from aging, and this occurred before the
true optimum was located. From the standpoint of achiev-
ing the most efficient optimization, increasing the number
of runs per response evaluation does not help. The simplex
algorithm is self-correcting for random errors if they

are not so overwhelming that they cause premature contrac-
tion (A9).

In the manganese optimizations, the algorithm con-
verged on the vertical optimum more rapidly than it con-
verged on the horizontal optimum in three of four optimiza-
tions. For cadmium, the convergence on the horizontal
optimum was always rapid. This difference is probably
due to the greater vertical slope in the manganese res-
ponse surface than in the cadmium response surface. The
overall performance of the simplex algorithm in finding
an optimum close to the global optimum was poorer for
manganese than for cadmium, which seems to indicate that
there is more noise on the manganese response surface

than on the cadmium response surface. Since the

139

atomization temperature for manganese is about 1000°C
higher than the atomization temperature for cadmium, the
disturbance of the sheath gas flow laminarity near the
atomizer is greater for manganese, and undoubtedly this
is at least part of the cause for the noisier response

surface.

D. FOUR PARAMETER OPTIMIZATIONS

The labor-saving characteristics of the simplex
algorithm as compared to grid—search optimizations become
more significant as the dimensionality of the optimiza-
tion is increased. To collect data to evaluate the
entire response surface for a four factor optimization
is too time consuming even for a totally automated system,
such as the EAAA instrument. Since the operation of the
algorithm in a four parameter optimization was untested,
it was necessary to test the technique on four parameters
whose effects on the selected response were either known

or could be easily deduced.

1. Four Parameter Optimization of the Integrated

Absorbance of Cadmium

The four factors chosen were the horizontal position,
the vertical position, the amount of sample, and the

length of the time delay between sample deliveries. The

1A0

chosen response was again the cadmium integrated absor-
bance. There is not expected to be any interaction be—
tween the horizontal and vertical position variables and
the two new factors, so the horizontal and vertical optima
should be the same as in the two parameter optimization.
Obviously, the optimum value for the amount of sample
solution delivered should be the maximum amount allowed

by the upper boundary, and the length of the time delay
between sample determinations should have no effect on

the response.

In the five test cases shown in Table 9, the starting
simplexes were located randomly in factor space according
to the guidelines established in the previous section.
Note that the number of iterations required for conver-
gence was significantly greater than the number required
in a two parameter optimization. Even though the delay
time has no effect on the response chosen (integrated
absorbance), in each case the simplex algorithm converges
on an "optimum" value for this factor. The inconsequential
nature of this factor is signalled by the totally random
values on which the simplex converges for this parameter.
This does demonstrate that the technique can find a true
optimum even if one of the chosen factors has no effect
on the response. The horizontal positions designated at
the optimum are not unlike those found in the two parameter

optimization, except that optimizations B and C yielded

1A1

Table 9. Results of Optimizations of Delay Time, Hori-
zontal Position, Vertical Position, and Amount
of Sample for Maximum Integrated Absorbance for
Cadmium.

 

 

 

2:21:21:- as xx ‘82:? All?
A 9.3 0.0 .6A 3.93 AA
B 18.0 -.63 3.6 3.97 31
C 5.A -.51 .87 3.9A 39
D 15.3 .28 .38 3.89 28
E 9.0 .11 1.1 3.92 30

 

 

1A2

a result a little farther from the true global optimum
than any of the optima found in the two parameter optimiza-
tions. This effect is even more pronounced in the opti-
mization of the vertical parameter where one of the
"optimum values" found was over 3 mm from the true optimum
value. The approach of the sample amount to the upper
boundary level is consistent throughout the test as was
initially expected. Since this parameter has a very
strong effect on the response, the convergence precision
for this parameter is very good.

For optimization E, the Figures 27, 28A, 28B, 29A and
29B are progress plots of the response, vertical position,
horizontal position, sample amount, and delay time, respec-
tively. Comparison with parameter progress plots for the
two dimensional optimization show that the improvement
in response is more gradual, and the movement toward the
horizontal and vertical optima is also more gradual.

Also, the approach to the vertical and sample amount

optima, which are located at the parameter boundaries,

is rather slow and tentative. The difficulty that this

simplex algorithm has in locating an optimum located at

a boundary seems to be characteristic of this algorithm.

The super modified simplex is thought to approach optimum

located at a boundary value much more readily (37).
Unfortunately the progress plot of the delay time

parameter contains no distinctive characteristics which

RESPONSE

0.0

Figure 27..

1A3

 

 

 

 

 

 

 

 

2E

3;

2:

SE

1:

2:

'1':

::

‘LIIIJIJI ll! llJllllJlJlllllllllllJ
IIIIIII III IIIITTIIITIIIIIIIIIIV

1 s 10 1s 20 25

SIMPLEX N0.

Progress Plot of the Cadmium Integrated Absorbance
Response in a Four Parameter Optimization.

1AA

.mucaacomn< umuogmmuca savanna on» mo co¢ue~.s.uao cmpmsmgma used a e? .
co*u.moa Poucourgoz we» ecu covurmoa Fourucm> as» ea muopa mmmcmoca .mm «camva

 

 

 

 

 

 

 

.oz xmnmzmm .02 anasz
n~ on a. e_ a . mu e~ m. o. m L
0.0 E._________~—L_______E_EP___—~___ ad
1 .2 __. _-_:J_u_____ 4fi_______..
#
. .T
. on An 2
0 H A
. ”U .. 3
at. fi.~8
Z .. l
on 0 u.- o I
N Junxu
i am V
. .r .l
. w” nu ;
a d
o d 1.. o 0
a nu Aw a e.
S U. I
. l .. i
o I nun-61
m .N.
oh luv... oh
\I .1 \I
. H H
. w n... w
( 1. I\
+
& mm&
.2 mm.

 

.S

1A5

.mucwngoma< vmuacmmucn savanna ecu mo covuanpsvuno

gmumsagme hack a cw we?» Ampmo on» ten acaos< upasmm mzu we «ope mmmgmoga .am mesmvu

   

 

 

.oz xmndzmm. .oz xmnmsz
mu ca 3 S m a 3 on 3 2 m a
n 4
1.- A”
An.” w
+ u
H i
A. .. ..
.. n. H
1r
.. a. n
fiJw agn. ..
”n v. H
.. ..A :I
.u. .2 ..
.. .1 u
H I ..
.n. .2 H u
H a. H
_ luv: .2 \I 1.”:
“n as n
i -. . 3 H
\ nu .uqs :
.. r. n
n. .2 A.
H H
_ .n .2 n
H m
LLfi oh" hr

 

 

o...
.u S
V
N
d
1
3

.~
V
N
O
n
N
.n i
\I
.H
1
I\

J

1A6

would point out that this parameter has no effect on the
response chosen. It appears that the only way to show
definitively the insignificance of this factor is to com-
pare the optimum values for this parameter in several

optimizations.

2. Simplex Optimization with a Complex Response

Function

The second four parameter optimization attempted was
of a more practical nature. It is generally desirable in
EAAA determinations, as with many other analytical deter-
minations, to maximize the accuracy and precision of the
measurement process and to minimize the analysis time.
This means that the value used as feedback to the simplex
algorithm must be some type of weighted sum of individual
responses, which is representative of the analysis time
and the precision and accuracy of the measurement. Within
this context, the parameters chosen as factors for the four
parameter optimization were horizontal position, vertical
position, desolvation time, and desolvation power. Since
a major portion of the analysis time under normal operat-
ing conditions is spent in desolvating the sample, minimiza-
tion of this variable could represent a significant in-
crease in instrument throughput. The desolvation power
was chosen as a factor because it interacts with the

desolvation time. That is, the time necessary to

1A7

desolvate the sample is not independent of the desolvation
heating temperature.

The response chosen to measure the analysis time
is straightforward. However, an experimentally practical
measure of the measurement precision and accuracy is more
difficult. In normal practice the standard deviation is
used as a measure of the precision and an indication of
the accuracy. The number of experiments necessary to
give a reasonably reliable estimate of the standard devia-
tion would cause the optimization to take a prohibitively
long time. If one assumes that the noise in the region of
the global optimum is homoscedastic and that the EAAA
measurement is well-behaved with respect to the relation-
ship of the integrated absorbance to the cadmium concen-
tration, then the assumption that the magnitude of the
integrated absorbance is representative of the precision
and accuracy of the measurement is justified.

Rules for designing the function which calculates the
combinational response from two individual responses have
been established (3A,A7). However, each particular system
requires some prior knowledge and careful consideration
before a satisfactory response function can be established.
First, it was necessary to write an equation for each
individual component of the response function that would
normalize each individual response to be near unity at

its optimum value. For the signal response contribution,

1A8

Ra, the integrated absorbance was divided by the estimated
maximum value of the integrated absorbance. Under these

conditions,

ABSINT (1)

Ra = 3.

The time response contribution, Rt, was normalized with
respect to the upper and lower boundary constraints as

follows.

Rt = 1. - (DEST - 5.) (2)

A0. - 5.

 

The combinational response, Rc, was then calculated as

a weighted sum of these normalized responses by Equation

3,

Rc = .75Ra + .25Rt . (3)

A subroutine was written to execute the calculation
of this response function. The routine was assimulated
into the load module for GBMDS2 program, and activated in
the simplex mode of operation by setting the variable
which indicates an alternate response evaluation will
be required. After A0 iterations, the simplex algorithm
had converged to the expected precision for the horizontal

position (-0.A5 mm), the vertical position (0.A6 mm), and

1A9

the desolvation power (.A6 volts at the DAC which controls
the braid temperature). At this point the average desolva-
tion time for the simplex was l6.A s. But, the range of
values for the vertices in the final simplex was 13.9 to
20.3 s.

The optimization was not totally successful. The
shape of the response surface under this weighting scheme
allowed a proper convergence for the horizontal and verti-
cal parameters, but not for the other two factors. When
a combinational response is used, the response surface
may be thought of as the convolution of the individual
response surfaces. This process will make it likely that
the combinational response surface will be less regular
than most individual response surfaces. Thus, the prob-
ability of a successful optimization is diminished. In
this particular case, it appears that the weighting of
the response toward the signal magnitude initially favored
finding the location of the horizontal and vertical optima.
Recall that it has already been pointed out that the simplex
will converge on an "optimum" for a factor even if this
factor is insignificant. In this weighting scheme, the
effect of changes in the delay time were effectively
screened by the large effect the position variables had
on the calculated response. When the position variables
had converged, the size of the simplex had diminished so

much that it was entirely located on a plateau in the

150

response surface. This caused experimental noise to

mask the impact that changes in the desolvation parameters
had on the response. Changing the weighting factors to
favor the desolvation parameters over the position param-
eters might allow the algorithm to find the true optimum.
From this study, it appears that the parameters that are
least significant and cause plateau regions in the response
surface should be weighted most heavily so that they
control the initial movements of the simplex. Thus their
optima will be located before the simplex contracts sig-
nificantly. This is not a panacea, because the portion

of the response surface traversed by the simplex is de-
pendent on both the starting simplex and the random ef-
fects of instrumental noise. For this reason, weighting
schemes that might work well in one trial may fail miser-
ably if the optimization is restarted with a different or
even the same initial simplex. The super modified simplex
algorithm might achieve more consistent results in this
type of optimization since it has been suggested that it
is immune to noise (37).

It has been shown in this section that four parameter
optimizations are possible on the EAAA instrument if the
response surface is well-behaved. It was also shown that
combination response functions add a degree of complexity
that increases the probability of failure of the simplex

algorithm.

151

E. OPTIMIZATIONS OF THE MEASUREMENT PRECISION

 

1. Direct Optimization of the Measurement Precision

 

As stated in the previous section it would be desir-
able to optimize the precision of the EAAA measurement
directly. However, this would require a large number of
experimental observations. For this reason, only a three
dimensional Optimization was attempted. Three parameters
were chosen which were thought to have a significant ef-
fect on the measurement precision: the vertical height,
the sheath gas flow rate, and the atomization power.

The vertical position should affect the precision by
several mechanisms. The higher the observation window,
the smaller the expected signal and the less laminar the
sheath gas flow. Both contribute adversely to the pre-
cision. On the other hand, low observation heights in-
crease the noise by increasing the-amount of shot noise
from the observation of background radiation from the atom-
izer. At very low positions, the atomizer may occlude
some of the source radiation and increase the detrimental
effect of some noise sources that are independent of the
level of source radiation. High atomization temperatures
can also contribute to the amount of continuum radiation
observed by the system, and more tepid temperatures can
cause incomplete atomization of the analyte. The sheath

gas flow rate should influence the measurement noise at

152

both low and high flow rates, since the flow laminarity
is destroyed at either extreme.

This optimization was actually carried out with a
multi-minicomputer system. One minicomputer controlled
the EAAA instrument and another executed the simplex
routine. Response and test conditions were transferred
between computers by the Pals asychronous link (i.e., the
operator typed the response or test condition produced
by one computer into the other computer's terminal). The
operator was also required to adjust manually the gas flow
rate. The starting simplex was generated by the simplex
program. Four sample runs were taken for each determina-
tion of the measurement precision, and after 1A iterations,
which required 120 sample runs, the simplex was halted.
The range of values for the starting simplex and final
simplex vertices are shown in Table 10. The flow rate
increased, the height of the vertical observation window
increased, but the atomization temperature remained essen-
tially unchanged. The improvement in overall measurement
precision was not very significant.

The failure of this optimization to converge is likely
due primarily to the statistical noise in the response
measurement. That is, the variance in the variance pre-
dicted for a set of only four determinations. The response
is not only quite noisy, it also contains a plateau in

the region of the optimum for at least the sheath gas

153

 

 

 

 

H.mlm.m m.z o.mtm.z H.m m.HHIm.m m.oa m.m:a.m m.m Hmcfim
.mlm.a :.m H.muu.: H.m z.m|o.n m.> o.m|m.m m.m HmeHCH
owcmm owmgo>< owcmm owmpo>< mwcmm owmpo>< mwcmm mwwpo>< onQEHm

2\m pmzom soapmNfiEOp< soapfimom Hmofipno> opmm 30am mom

 

 

.oapmm mmfloz ow Hmcwfim Ezefixmz how hmzom :oHpmNHEOp¢

new .eofipfimom Heefipee> .epmm seam mew one go cospemfiefipeo mo mpflsmem .OH magma

15A

flow parameter. Later studies showed that over the range
of about 2.5 to 3.5 l/min, the flow rate has little effect
on the measurement precision. It would also have been
better to use some intuition in the decision about the
size and location of the initial simplex rather than
accepting the simplex suggested by the computer routine.
Also of some concern, but probably not important here,

is the fact that some noise is resultant from sources not
under rigid control. As noted in Chapter A this noise is
in the range of l to 3% with some long term drift. This
noise level should only limit the precision of the final
determination of the optimum, and should not be serious

enough to invalidate totally the optimization process.

2. The Determination of the Oppimum Integration

Time Window

 

The attempt to optimize measurement precision by
varying the limits on the time window for the evaluation
of integrated absorbance was a good deal less ambitious
than the optimization just described and met with a great
deal more success. In this optimization, the integration
start time and stop time were varied under the constraint
that the stop time always had to be later than the start
time. Several data files were recorded that contained
a chronological series of 200 absorbance data points for

cadmium collected over a 1.0 5 period. Each of these

155

data files contained 10 replicate sample runs. The pre-
cision was evaluated within the limits of each new integra-
tion time window suggested by the simplex algorithm on
this data set so that no additional on-line data acquisi-
tion was necessary. The data in Table 11 clearly show that
optimizing the integration limits results in a significant
increase in measurement precision only when the amount
of analyte is near or below the detection limit found
when absorbance points during the entire atomization step
are integrated. Previous experiments in this genre, which
had a peak detection algorithm, had not shown any statis-
tically significant increase in the measurement precision
(28). The problem with such algorithms is that it is very
difficult to write one that will perform well for high and
low sample concentrations, and that will perform with any
consistency at or near the detection limit where the noise
in the signal is quite significant. Furthermore, these
algorithms can be expected to be advantagous only at low
concentrations, where the noise sources that are independent
of the photocurrent level become important.

Figure 30 is a typical absorbance peak shape for a 2
HA sample of 10 ppb cadmium, and is useful in the inter-
pretation of the start and stop times of the signal inte-
gration listed in Table 11. To test the validity of the
values found in these optimizations a scheme was devised

in which the optimum time values found by the simplex

.10 .—

ABSORBANCE

 

Figure 30.

156

0.0 . ..

.1 .2 .3 .4 .5 .6 .7 .8 .9 1.0
TIME (SEC)

A Typical Absorbance Peak for 2 pl of 1 ppm Cadmium

Solution

157

.UOHLmQ mchmon :oHpmNHEoum can no cam on»
a» ocm wcHCCHwon on» ohm coHpmpwoch no mpHEHH map con: COHumcHEHmpmo on» you 9mm R *

 

 

 

 

 

MO m.OH OH. OH. OH O
HA 0.0H OH. OH. OH O
OHH m.mH OH. NH. OH O
.ONH m.mH OH. HH. OH O
.Hm m.OH OH. mO. OH O
O.m m.m Hm. OH. OOH m
O.m O.m me. OH. OOOm O
COprpwoch
Hassoz EOEHOOO OH deem upepm HOOOO
HOOOHO OOHmHOeeO HOV meeHe coHOOewOOeH .eeoo OO pew
wcHCHmmH
.moEHB doum ocm ppmpm COHpmeoch mo mCOHpmNHEHon mo mpHSmom .HH oHnt

158

optimization were applied to several test sets. Table 12
contains the results of these tests. In Table 12, the
learning sets were data sets on which the simplex optimiza-
tion was performed, and the test sets were data sets which
were collected under identical experimental conditions.
The optimum time window as determined by simplex optimiza-
tion of each of the learning data sets was then used to
evaluate the data in the test sets. Also shown in Table
12 is the precision for each of the test data sets when
the absorbance integration includes the entire atomiza-
tion period. A 5 to 10 times improvement in the measure-
ment precision resulted when the test data were evaluated
between the simplex suggested limits as compared to the
integration over whole atomization period. Also, the
improvement in precision does not seem to vary significantly
for a given test set when the integrated absorbance is
evaluated in the different time windows found by the
simplex optimization of each of the learning data sets.
The results of these optimizations can only be applied
to a limited extent to actual determinations for several
reasons. The width of the absorbance peak is related to
the amount of the analyte in the sample so that a constant
time window will observe a smaller portion of large peaks
than of small peaks. This will cause curvature in cali-
bration plots, but this is not an unreasonable price to

pay for the observed increase in measurement precision

159

Table 12. Evaluation of Applicability of Simplex Deter-
mined Integration Time Window to Other Data
Sets.

 

 

Test Set (%RSD)

 

Learning
Set F G H I Normal Integration*
A 13 15 19 17 130
B 16 13 15 1A 120
C 20 2O 2O 22 120
D 17 20 21 2O 80
E 13 13 12 13 60

 

 

* % RSD for determination when the limits of integration
are the beginning and end of the atomization heating
period.

160

and sensitivity. The width and time location of the
absorbance peak is also a function of the ashing and
atomization temperatures. The atomizer will reach the
final atomization temperature earlier if the ash tempera-
ture is high, so that the absorbance peak will occur earlier
in time. Also the rate at which analyte atoms leave the
atomizer is temperature dependent so that the width of
the peak depends on the final atomization temperature.
These results point out the value of limited integra-
tion of the absorbance peak. Because of the interaction
of the integration start and stop times with the ash
temperature, atomization temperature, and sample amount,
it is highly desirable to have these parameters available
for optimization. These results lead to the conclusion
that an on—line optimization of the measurement precision
would have a very good chance of success. However, the
number of experimental data points needed would be large
and the acquisition of this data would be prohibitively
time consuming in anything but a totally automated environ-

ment.

F. CONCLUSIONS AND PERSPECTIVES

 

The software required to implement the simplex tech-
nique on the EAAA instrument was neither complex nor

lengthy, and the wetware (operator knowledge) required

161

was significant, but not unreasonable. The unique features
of this work include: 1) the simplex optimization tech-
nique has never been applied to EAAA spectroscopy, 2)

this is the first on-line application of the simplex that
allowed the user any flexibility in the choice of param-
eters to be varied in the optimization. The on-line op-
timization of the integrated absorbance signals of cadmium
and manganese proved that the technique was an efficient
method of finding the global optimum if and only if a
judicial choice of the initial simplex was made. The
optimization of the integrated absorbance signal by vary-
ing the delay time, sample amount, horizontal position,
and vertical position demonstrated the tendency of the
simplex to converge on an "optimum" value for a factor even
if it is insignificant. Restarting the simplex in another
region of the response surface not only helps to decide

if the optimum located was a local optimum or the global
optimum, but will also point out the insignificance of a
factor by converging on totally random optima for that
parameter. The attempted optimization of a combinatorial
response function demonstrated that as the response func-
tion becomes more complex, the response surface is likely
to take a shape less favorable for the operation of the
simplex algorithm. Optimization of the measurement pre-
cision was shown to be an experimentally difficult task

because a good estimate of the variance requires a large

162

population of individual measurements. It would not be
difficult to set up the optimization of the signal-to-
background ratio, since the background value could be
easily evaluated from the array which contains the data
for the correction of atomizer continuum background radia-
tion. The problem with this type of response function
evaluation is that the HCDT source drift causes an im-
precise determination of the background level. It would
be more desirable to evaluate the standard deviation of
this set of noise measurements and assume that this is a
good approximation of the noise in the signal measure-
ment. This procedure is still somewhat susceptible to
HCDT drift, but software could be generated to minimize
the effect of this drift. A direct optimization of the
precision of the measurement would be very valuable and
could be applied to the determination of practical samples
such as the determination of manganese in chlorophyll
(86).

There have been at least as many suggested improvements
to the simplex optimization technique as there have been
researchers who have used the technique. Most suggestions
have been directed toward the desensitization of the
algorithm to the effects of noise on the response surface
(37,58,63). This work has pointed out the critical nature
of the size, location, and orientation of the starting

simplex on the success of the optimization. I propose

163

that the starting simplex be generated by selection of

the sum of the vertices of a regular polyhedron of n+m
dimensions, where n is the number of factors to be optim-
ized and m is an arbitrary integer greater than 1. The
response at each vertex of this figure is then evaluated,
and the n best vertices and the worst vertex retained.
These are then used to form the actual starting simplex
for the simplex algorithm. In two dimensions this pre-
liminary figure might be an equilateral pentagon. Nor-
mally the two best vertices will be adjacent vertices and
the worst vertex will be opposite the face formed by the
two best vertices. When the two next worse vertices are
discarded, the remaining figure is a triangle elongated

in the direction of the optimum. Thus the simplex can

be moved rapidly toward the optimum on the first move.
Often the region in which the simplex starts is the
noisiest section of the response surface that it will
have to traverse. Thus, it is critical to move it rapidly
in the correct direction, so that the simplex does not fall
prey to its natural statistical tendency to contract pre-
maturely in the presence of noise in the response. The
temptation to retain more than n+1 vertices in the working
simplex should be restrained, since increasing the number
of vertices will make the simplex move more sluggishly

over the response surface.

REFERENCES

3.

10.

11.

12.

13.

1A.

15.

16.

17.
18.

References

A. Walsh, Australian Patent No. 230Al (1953).
A. Walsh, Spectrochem. Acta, 1, 108 (1955).

C. T. J. Alkemade and J. M. W. Milatz, Appl. Sci. Res.
32. 289 (1955).

C. T. J. Alkemade and J. M. W. Milatz, J. Opt. SOC.
Am": _A_5_, 583 (1955).

W. H. Wollaston, Phil. Trans. Roy. Soc. London, Ser
A, 23, 365 (1802).

T. T. Woodson, Rev. Sci. Instr., AA, 308 (1939).
B. V. Lvov, Spectrochim. Acta., 11, 761 (1961).

R. Woodriff and G. Ramelow, Spectrochim. Acta., 23B,
665 (1968). —

R. Woodriff and R. Stone, Appl. Optics, 1, 1337 (1968).

R. Woodriff, R. W. Stone, and A. M. Held, Appl. Spec-
trosc., 22, A08 (1968).

R. Woodriff, B. R. Culver, and K. W. Olsen, Appl.
Spectrosc., EA, 530 (1970).

R. Woodriff and D. Shrader, Anal. Chem., AA, 1918
(1971).

G. K. Pagenkopf, D. R. Neuman, and R. Woodriff, Anal.
Chem., AA, 22A8 (1972).

R. Woodriff, B. R. Culver, D. Shrader, and A. B. Super,
Anal. Chem., A5, 230 (1953).

H. Massmann, Spectrochim. Acta., 23B, 215 (1968).

H. Massmann, in Flame Emission and Atomic Absorption
Spectrometry, Vol. 2, J. A. Dean and T. C. Rains, Eds.,
Marcel Dekker, New York, 1971.

Perkin-Elmer Corporation, Instrument News, AA, A (1970).

H. L. Kahn and S. Slavin, A. A. Newsletter, 19, 125
(1971).

16A

19.

20.

21.

22.

23.

2A.

25.

26.

27.
28.

29.

30.
31.
32.

33.
3A.
35.

36.
37.

165

Perkin-Elmer Corporation, Analytical Methods for Atomic
Absorption Spectroscopy Using the HGA Graphite Furnace,
Norwalk, Conn., March 1973.

T. S. West, and X. K. Williams, Anal. Chim. Acta., A5,
27 (1969)-

R. G. Anderson, I. S. Maines, and T. S. West, Anal.
Chim. Acta, 5;, 355 (1970).

J. F. Alder and T. S. West, Anal. Chim. Acta, 5;, 365
(1970).

K. W. Jackson, T. S. West, and L. Balchin, Anal. Chem.,
35. 2A9 (1973).

A. Montaser and S. R. Crouch, 25th Pittsburgh Conf.
Anal. Chem. Appl. Spectrosc., Papers 199 and 116 (197A).

A. Montaser, S. R. Goode, and S. R. Crouch, Anal. Chem.,
.36. 599 (197A).

A. Montaser and S. R. Crouch, Anal. Chem., A6, 1817
(197“)-

A. Montaser and S. R. Crouch, Anal. Chem., Al, 38 (1975).

D. N. Baxter, Ph.D. Thesis, Michigan State University,
East Lansing, MI, 1977.

A. Montaser, Ph.D. Thesis, Michigan State University,
East Lansing, MI, 197A.

S. Augusta, CRC Crit. Rev. Anal. Chem., A, 155 (197A).
E. P. Box, Biometrics, lg, 16 (195A).

. E. P. Box and K. B. Wilson, J. Roy. Statist. Soc.
-. 1;. l (1951).

G

G

B

G. E. P. Box, Appl. Statisc., g, 81 (1957).

C. W. Lowe, Trans. Instr. Chem. Engrs., Ag, T33A (196A).
W

. Spendley, G. R. Hext, and F. R. Hinsworth, Techno-
metrics, A, AAl (1962).

J. A. Nelder and R. Mead, Computer J., l, 308 (1965).

M. W. Routh, P. A. Schwartz, and M. B. Denton, Anal.
Chem., Ag, 1u22 (1977).

166

38. K. w. Lam, Clin. Chem., g;, 89 (1977).
39. D. J. Legget, Anal. Chem., A9, 276 (1977).

A0. J. K. Haken, M. S. Wainwright, and R. J. Smith, J.
Chromat., 133, l (1977).

Al. E. Baumgarten, F. Weinstrauch and H. Hoffkes, J. Chromat.,

A2. G. L. Ritter, S. R. Lowry, C. L. Wilkins, and T. L.
Isenhour, Anal. Chem., A1, 1951 (1975).

A3. T. R. Brunner, C. L. Wilkins, T. F. Lam, L. J. Soltz—
berg, and S. L. Kaberline, Anal. Chem., A8, llA6 (1976).

AA. D. E. Long, Anal. Chim. Acta, AA, 193 (1969).

A5. S. N. Deming and S. L. Morgan, Anal. Chem., A5, 278A

A6. S. N. Deming and P. G. King, Research/Development, g5,
22 (197A).

A7. S. L. Morgan and S. N. Deming, Anal. Chem., A6, 1170
(197“). T"

A8. P. G. King, S. N. Deming, and S. L. Morgan, Anal. Lett.,
8. 369 (1975).

A9. M. J. Holue, D. E. Long, and D. Smette, Anal. Lett.,

50. F. P. Czech, J. Ass. Offic. Anal. Chem., 5g, 1A89
(1973).

51. F. P. Czech, J. Ass. Offic. Anal. Chem., 5Q, 1A96
(1973).

52. L. R. Parker, Jr., S. L. Morgan, and S. N. Deming,
Appl. Spectrosc., g2, A29 (1975).

53. R. D. Krause and J. A. Lott, Clin. Chem., 29, 775 (197A).

5A. W. K. Dean, K. J. Heald, and S. N. Deming, Science,
189, 805 (1975).

55. F. Darvas, J. Med. Chem., AZ, 799 (197A).

56. D. E. Fonner, J. R. Buck, and G. S. Banker, J. Pharm.
Sci., g2, 1587 (1970).

57.

58.
59.

60.

61.

62.

63.

6A.

65.

66.

67.

68.

69.

70.

71.

72.

73.

7A.
75.

167

W. E. Rippetoe, E. R. Johnson, and T. J. Vickers, Anal.
Chem., A1, A36 (1975).

R. R. Ernst, Rev. Sci. Instr., 9, 998 (1968).

E. R. Johnson, C. K. Mann, and T. J. Vickers, Appl.
Spectrosc., 59, A15 (1976).

P. G. King, Ph.D. Thesis, Emory University, Atlanta
(197A).

O. V. Thomas, L. Kryger, and S. P. Perone, Anal. Chem.,
fig. 761 (1976).

M. W. Routh, P. A. Swartz, and M. B. Denton, Anal. Chem.
39. 1A22 (1977).

E. R. Johnson, Ph.D. Thesis, Florida State University,
Tallahassee, Florida (1975).

K. M. Cellier, and H. C. T. Stace, Appl. Spectrosc.,
g9, 26 (1966).

M. L. Parsons and J. D. Winfordner, Appl. Spectrosc.,

N. V. Mossholder, V. A. Fassel, and R. N. Knisley,
Anal. Chem., A5, 161A (1973).

J. L. Malakoff, J. Ramirez-Munoz. and A. Scott, Anal.
Chim. Acta, 33) 515 (1968).

S. R. Crouch, D. N. Baxter, E. H. Pals, and E. R. John-
son, Anal. Chem., 59, 291A (1978).

Digital Equipment Corp., OS/8 Handbook, Maynard, Mass.
Digital Equipment Corp., 197A, Chapter 8

S. P. Keller and J. F. Eagleston, J. Chem. Ed., A5,
317 (1971).

E. C. Toren, Jr., R. N. Carrey. A. E. Sherry, and J.
E. Davis, Anal. Chem., AA, 339 (1972).

H. E. Keller, G. E. Courtois, and J. E. Keller, Chem.
Instr., A, 269 (1972).

G. S. Cembrowski, D. B. Cottrel, and E. C. Torren,
Computers and Chemistry, A, A5 (1975).

R. E. Dessy, Anal. Chem., 32, llOAA (1977).

L. A. Leventhal, Killobaud, ll, 2A (1977).

76.

77.

78.

79.

80.

81.

82.

83.

8A.

85.

86.

168

F. J. M. Massen, F. D. Posma, J. Balke, Anal. Chem.,
56. 114145 (1971:).

V. Sacchetti, G. Tessari, and G. Torsi, Anal. Chem.,
39, 1175 (1976).

C. J. Molnar and J. D. Winfordner, Anal. Chem., A5,
1807 (197A).

S. R. Goode, A. Montaser, and S. R. Crouch, Appl.
SpeC-. 21. 335 (1973).

S. R. Crouch, A. Montaser, and S. R. Goode, in "In-
formation Chemistry: Assisted Chemical Research
Design", (8. Fujiwara and H. B. Mark, eds.), University
of Tokyo Press, Tokyo, 1975, pp. 107-12A.

"Sigma Stepping Motor Handbook", (Sigma Instruments,
eds.), Braintree, Mass., 1972.

R. G. Anderson, I. S. Maines, and T. S. West, Anal.
Chim. Acta, 5;, 355 (1970).

D. N. Baxter, E. H. Pals, E. R. Johnson, and S. R.

Crouch, 27th Pittsburgh Conf. Anal. Chem. Appl. Spectrosc.,

Paper 169, Cleveland, OH, March 1976.

R. D. Reeves, B. M. Patel, C. J. Molnar, and J. D.
Winefordner, Anal. Chem., A5, 2AA6 (1973).

M. D. Joseph and E. R. Johnson, unpublished program,
1977.

J. T. Gano, E. H. Pals, and S. R. Crouch, Current
Research (1978).

APPENDIX A

Selected Program Listings

169

GOOnOOOOOOOOOOOGOOOOOOO0000000

170

' PROGRAM FUNCTION: MAIN PROGRAM FOR GBA EXPERIMENTS

MODIFIED TO ALLOW SIMPLEX OPTIMIZATIONS
PROGRAMMER: E. R. PALS

VERSION: *IAPI (2.03) (3.01)
DATE: 10/10/76 (12/31/76) (Ell/77)
FILENAME: EPMAIN.F4 (EPMNV2.F4) (EPMNSZ.F4)

LOADING INSTRUCTIONS: MUST BE LOADED INTO OVERLAY ZERO.

NOTE: FURTHER INFORMATION IS AVAILABLE IN
THE FORM OF A HELP FILE.

CHANGES VERSION 2.01:
1. DATA AND CALC ARE COMBINED INTO ONE PROGRAM.

“2. COMMON VARIABLE DEFINITIONS ARE UPDATED.

3. ADD OPTION OF USER SUBROUTINE THREE

CHANGES VERSION 2.02:
l. SWITCH REGISTER OPTION FOR PLOTTING IS ADDED

CHANCES VERSION 2.03:

1. GBA SWITCH REGISTER.OPTION 4 IS USED TO FORCE CALL
TO PLOT. ,

CHANGES VERSION 3.0!:

I. MODIFIED TO ALLOW SIMPLEX OPTIMIZATIONS.

. ~-——.—

OOOOO000000000063000OOOOOOOOOOO

171

PROGRAM FUNCTION: MAIN PROGRAM FOR GBA EXPERIMENTS
MODIFIED TO ALLOW SIMPLEX OPTIMIZATTONS
PROGRAMMER: E. H. PALS

VERSION: *lA-I (2.03) (3.01)
DATE: 10/10/76 ( 12/31/76) (fill/77)
FILENAME: EPMAIN. F4 ( EPMNVZ. F4) ( EPMNS2. F4)

LOADING INSTRUCTIONS: MUST BE LOADED INTO OVERLAY ZERO.

NOTE: FURTHER INFORMATION IS AVAILABLE IN
THE FORM OF A HELP FILE.

CHANGES VERSION 2.01:
1. DATA AND CALC ARE COMBINED INTO ONE PROGRAM.

0 2. COMMON VARIABLE DEFINITIONS ARE UPDATED.

3. ADD OPTION OF USER SUBROUTINE THREE

CHANGES VERSION 2.02:
1. SWITCH REGISTER OPTION FOR PLO'ITING IS ADDED

CHANGES VERSION 2.03:
1. GBA SWITCH REGISTER OPTION 4 IS USED TO FORCE CALL
TO PLOT.

CHANGES VERSION 3.01:
1. MODIFIED TO ALLOW SIMPLEX OPTIMIZATIONS.

0000000000000OOOOOOOOOOOOOOOOfiOO

172

METHOD :

I.

5.
6.

CALL INPUT- THIS ROUTINE ALLOWS INITIALIZATION

OF INSTRUMENT PATAMETERS. THE ROUTINE WILL NOT

ALLOW EXIT UNLESS: ALL PARAMETERS ATE WITHIN

LEGAL LIMITS. A DARK CURRENT VALUE HAS BEEN TAKEN

THE SAMPLER AND POSITIONER INITAILIZED. IT ALSO

STARTS TE REAL TIME CLOCK.

CALL FLOW- THIS ROUTINE HANDLES CHANGING THE

POSITION AND SAMPLE SPECIFICATIONS. WHEN AN

ASSIGNED TASK HAS BEEN COMPLETED THE OPERATOR

MAY ENTER COMMENTS AND CLOSE THE FILE. EXIT

MAY BE BACK TO MAIN WITH INSTRUCTIONS

TO GO EITHER TO THE DATA TAKING ROUTINE OR TO

TO GBA MONITOR( SUBROUTINE INPUT).

CALL DATA- THIS PROGRAM EXECUTES CALLS TO POS.SAMPD.DACo
ADC. AND CLOCK ROUTINES TO ACQUIRE READINGS OF 100”.
SIGNAL ATTENUATED BY SAMPLE. AND BACKGROUND EMMISION OF GBA.
THE ABSORBANCE RESULTS ATE THEN CALCULATED. THE DATA

MAY THEN BE PACKED TO PASS TO THE PLOT ROUTINE. 0R WRITTEN
ON ONE OR BOTH OF TWO OUTPUT FILES. AND/0R

OUTPUT TO THE LINE PRINTER. (COMBINES FUNCTIONS OF
VERSION 1 PROGRAMS DATA AND CALC)

IF REQUESTED CALL SUBTHR- THIS IS AN OPTIONAL

USER SUBROUTINE. IT IS NOT NECESSARY FOR CORRECT PROGRAM
EXECUTION AND IF IT IS NOT REQUESTED IT NEED NOT BE
LOADED INTO THE LOAD MODULE.

IF REQUESTED CALL PLOT AND PLOT THE RESULTS

ON THE ADDS AND/OR THE LINEPRINTER.

RETURN TO STEP2 TO FIND OUT IF TASK IS FINISHED.

ODD

Cit-0

0

IO

0 06300

173

PROGRAM MAIN EXECUTABLE CODE FOLLOWS

COI’H‘ION
CC’I‘L‘ION
COIU‘ION
COI‘ZTICI

COPE-ION
COIZI’JN
COI‘II‘IOI‘I
C OPE-ION
CO 13202"?
COILT ION

PLTI'.LPT.RAD.I{DF.WAIT. DEST.ASHT.ATMT.DELT. DESP

ASIIP , ATI‘ZP , SCAN . VER. BOR, VERI . HORI . VEPL. HORL

NI‘ISCAN . NINIRUN , SAM? . SAMPSZ

FIRSTI . ISAIIC, NSCANJ-EPOS, VPOS. IRUN. ISAI'IP

AI'INT. ISAI‘ICA.TAG. XIOIDK.XIDK

FREQ, INTST. INTSP

/A/ ISAI‘IPL( 4-0) . NRUN, IGOTO

/B/DATAI( 100) . DATX‘2( 100) , AKIN, AMAX, IDIR

/CBLI{/IIODEI . SUB! . SUB2. F ILEl . F ILE2. DARKT.P INIT.SINIT.SU'33
/ [DEB/ITER. VIN IT( 5 . 4) , PM. ISVAR( 4) , ITYPE. LSI‘IP. KRUN

LOGICAL PLTT.LPT,RAD.HDF,WAIT.SCAN,SA1‘ZP.FIRST1.SUB3.LSIZ?
IS'GIN BY CALLING INPUT TO SET UP PARAIETEPS
CALL INPUT

II W CALL SII-IPLEX FROG
IF( .NOT.LSIIP) GO TO 2

III-LII IF OPTIMIZATION PqEQUESTED

CALL SMXOPT

GO T01

NOW CALL FLOWS- IS THE TASK FINISHED. WHAT IS TO BE DONE NEXT?
CALL FLOW

CO TO(1.1,3).ICOTO

CALL DATA EXERCISE THE HARDWARE

ALSO. CALCULATE THE RE ULTS

CALL DATA

IF REQUESTED GO TO USER PROGRAM IN LEVEL ONE

IF(SUBG) CALL SUBTHR

IF REQUESTED PLOT THE RESULTS

IF(PLTT.OR.(SWITCH(4.).EQ.1.)) CALL POT

CO T02

END

DO00000000COOCOCOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO??O??O?O??OO

VER:

1714

BLOCK DATA SUBROUTINE TO INITIALIZE COMMON FOR GBA
PROGRAMMER: E. H. PAIS

2.10

COMMENTS: AS VARIABLES ARE ADDEI) TO COMMON THIS PROGRAM MUST
BE UP DATED PER PAGE 8-119 OS/B MANUAL.

LOADING REQUIREMENTS: THIS SUBPROGRAM MUST BE LOADED
IN LEVEL ZERO AND CANNOT BE MADE PART OF A LIBRARY.

VARIABLE DESCRIPTION :

PLTT-
LPT-
RAD-

HDF-
WA I T-

FIRSTI-

SCAN-
SAMP-
SUB 1-
sm-
SUB3-
F ILEI-

FILE2-

. DABKT—

PINIT‘-
SINIT-

DEST-
DESP-
ASHT-
ASHP-
ATMT-
ATMP-
VER-

HOR-

VPOS-
HPOS-
VERI -'

HORI -
VERL-
HORL-

SAMPSZ-

LOG I CAL VARI ABLES-

PLOT RESULTS

OUTPUT RESULTS TO LINE PRINTER

USE RADIATION PROGRAMING (VS. POWER PROG.)

DURING ATOMIZATION STEP

USE HYDROGEN DIFFUSION FLAME DURING ATOMIZATION
PAUSE AFTER PLOTI'ING TO ALLOW DATA INSPECTION

OF MANUAL DELIVERY FO SAMPLE

USED TO INDICATE TO FLOW WHETHER INITIALIZATION OF
VARIABLES IS NECESSARY

A ONE OR TWO DIMENSIONAL SCAN IS REQUESTED

USE OF THE AUTO SAMPLER FOR SAMPLE DELIVERY IS REQUESTED
CALL USER SUBROUTINE ONE FROM DATA BEFORE CALCULATING
RESULTS

CALL USER SUBROUTINE TWO FROM DATA AFTER CALCULATING
RESULTS

CALL USER SUBROUTINE SUBTHR FROM MAIN AFTER RETURNING
FROM DATA

WRITE INTEGRATED ABSORBANCE. ETC. TO FILE SPECIFIED
AS DEVICE 5 TO FRTS

WRITE BACKGROUND CORRECTED ABSORBANCE FOR EACH

DATA POINT TAKEN

DARK CURRENT VALUE HAS BEEN TAKEN

POSITIONER HAS BEEN INITIALIZED

SAMPLER HAS BEEN INITIALIZED

’REAL VARIABLES-
' DELT-

DELAY TIME BETWEEN SAMPLE "RUNS (SESJ

'DESOLVATION TIME (S)

DESOLVATION POWER ( VOLTS FROM DAC OUTPUT)

ASH TIME (S)

ASH POWER (V)

ATOMIZATION TIME(S)

ATOMIZATION POWER( V)

INITIAL VERTICAL POSITION (MM)

INITIAL HORIZONTAL POSITION (MM)

LAST (PRESENT) HORIZONTAL POSITION (MM)

LAST (PRESENT) VERTICAL POSITION (MM)

WHEN SCANNING DISTANCE THAT VERTICAL POSITION ‘IS
INCREMENTED (MID

AS ABOVE FOR THE VERTICAL POSITION (MM)

MOST EXTREME LEGAL VERTICAL POSITION IN A GIVEN SCAN (MM)
MOST EXTREME LEGAL HORIZONTAL POSITION IN A GIVEN SCAN (MM)
VARIABLE USED TO INDICATE SAMPLE SIZE WHEN USING MANUAL
SAMPLE DEPOSITION

DOOOOOOOOOOOOOOOOOOOODDOOOOOOOOOOOOOOOOOOO00000000000

I NT‘SP-

175

AMNT- AMOUNT PT SAMPLE DELIVERED BY AUTOSAMPLER (UL)
XIDK- AVERAGE OF 100 DARK CURRENT READINGS (0-4095)
XIOIDK- AVERAGE OF 100 READINGS OF 100%T READONGS
ABSINT- INTEGRATED ABSORBANCE FOR A SAMPLE RUN
' FREQ- DATA ACQUISITION RATE (HZ)
AMIN- MINIMUM ABSORBANCE READING IN A PARTICULAR SAMPLE RUN
AMAX- MAXIMUM ABSORBANCE IN A PARTICULAR SAMPLE RUN
DATA1- ARRAY CONTAINS 100 TIME REPRESENTATIVE VALUES
FOR THE SAMPLE ATTENUATED TRANSMITTANCES
DATA2- ARRAY CONTAINS 100 BACKGROUND TRANSMITTANCES
TAGI- CONTAINS A2 FORMAT TAG FOR DATA OUTPUT STAT'MENTS
INTEGER VARIABLES-
NMSCAN- NUMBER OF SCANS REQUESTED
NSCAN- NUMBER OF SCANS COMPLETED
NMRUN- NUMBER OF RUNS REQUESTED
NRUN- NUMBER OF RUNS COMPLETED
ISAMC- SAMPLE SPECIFICATION PRESENTLY BEING EXECUTED
IRUN- NUMBER OF RUNS REQUESTED FOR A PARTICULAR SAMPLE
NUMBER AND AMOUNT
ISAMCA- NUMBER OF SAMPLE DELIVERIES COMPLETED ON THE PRESENT
SAMPLE SPECIFICATION
ISAMP- SAMPLE CUP NUMBER FROM AUTOSAMPLER THAT WAS REQUESTED
IGOTO- CONTROLS PROGRAM FLOW IN MAIN UPON RETURN FROM SUBROUTINE
IDIR- GIVES THE DIRECTION (POSITIVE OR NEGATIVE) OF THE
LAST HORIZONTAL MOVE
ISAMPL- ARRAY CONTAINS SAMPLER SPECIFICATION WORDS EACH OF
WHICH IS A CODED AND PACKED WORD WHICH INDICATES
THE SAMPLE NUMBER. AMOUNT. AND NUMBER OF DELIVERIES
REQUESTED
MODE- INDICATES DEVICE FROM WHICH INSTRUCTIONS ARE .TO BE READ
0= ADDS TERMINAL
4-: DEVICE (FILE) SPECIFIED AS NUMBER 4 TO FRTS
SIMPLEX VARIABLES-
LSMP- LOGICAL VARIABLE INDICATES SIMPLEX MODE OF OPERATION
IT'YPE- INDICATFS MDE OF CALCULATING RESPONSE
ISXVAR- INTEGER ARRAY WHICH CONTAINS THE NUMBERS WHICH RELATE'TO
WHICH VARIABLES ARE TO BE OPTIMIZED.
IDEM- DIMENSIONALITY OF THE SPACE (NO. OF FACTORS)
V(I.J)- VALUE OF FACTOR J AT VERTEX I
ALIM(3.J)- LOWER UPPER LIMITS. PRECISSION IN EACH-FACTOR J
IRET(I)- NO. OF MOVES VERTEX I HAS BEEN RETAINED
IMV- TYPE OF MOVE
IWRST- INDEX OF NEW WORST POINT
INTST- NUMBER OF POINT WHERE INTEGRATION STARTS

NUMBER OF T'HE POINT WHERE INTEGRATION STOPS

176

BLOCK DATA

VARIABLES USED BY‘MANY OF THE ROUTINES ARE IN BLANK COMMON;
COMMON PLTT.LPT,RAD.HDF.WAIT,DEST.ASHT.ATMT{DELT.DESP
COMMON ASHP.ATMP,SCAN.VER.HOR.VERI.HORI.VERL.HORL

COMMON NMSCAN,NMRUN.SAMP.TAGI

COMMON FIRSTI.ISAMO.NSCAN.HPOS.VPOS.IRUN.ISAMP

COMMON AMNT. ISAMCA,ABSINT.XIOIDK.XIDK

COMMON FREQ.INTST.INTSP

COMMON /A/ISAMPL(40).NRUN.IGOTO

COMMON /B/DATA1(IOO),DATA2(I00).AMIN.AMAX.IDIR

COMMON /CBLK/MODE,SUBI.SUB2,FILE1.FILE2.DARKT,PINIT}SINIT,SUB3
COMMON /SPBLK/SP1.SP2.SP3,SP4,SP5,SP6

COMMON IISIM/ IDIM.V(5.4).Rlfi).ALIM(3.4).IRET(5).IMV.IWRST
COMMON /IDEB/ITER4VINIT(5.4).PM.ISXVAR(4).ITYPE.LSMP.KRUN.IL
DATA AMNT/1.0/

DATA MODE/OI

DECLARE LOGICAL VARIABLES AND INITIALIZE OPTIONS

LOGICAL PLTT,LPT.RAD.HDF.WAIT,SCAN.SAMP.FIRSTI.FILE
LOGICAL SUB!.SUBZ.FILEI.FILE2,DARKT,PINIT,SINIT.SUB3.LSMP
DATA PLTT.LPT.RAD.HDF.WAIT.SCAN.SAMP/l.l.1.0.0.0.0/

DATA FIRSTl,SINIT.SUB3/l,0.0/

DATA SUB!.SUB2.FILE1.FILE2,DARKT.PINIT.LSMP/0.0.0.0.0.0.0/
INITIALIZE HEATING TIME PERIODS

DATA DEST.ASHT,ATMT,DELTVIO.,10...5,10./

INTIALIZE POWERS

DATA DESP.ASHP,ATMP/.25,.25.].0/

INITIALIZE SCAN PARAMETERS

DATA VER.HOR.VERI.HORI.VERL.HORL/l.,0.,1..l.,12.,4./

DATA NMRUN.NMSCAN/I.l/

INITAILIZE DATA TAKING FREQUENCY'TO 200 HZ

~DATA FREQ.ABSINT/200..O./

INITIALIZE OTHER VARIABLES
.DATA VPOS.HPOS,NRUN.ISAMP.AMNT70.,O.,1.I.2./

(DATA INTST.INTSP/I.50/

DATA IGOTO/S/

DATA SP1,SP2,SP3.SP4,SP5,SP6/6*O./
DATA TAGI/'DA’/

END

OOOOOOOOOOOOOOOOOOOOOOOOOOOOO0000000000000OOOOOOOOODOOOOOOOOONDOOO

177

'PROGRAM.NAME: SUBROUTINE INPUT

VER: 3.01

PROGRAMMER: E. H. PALS
CHEMISTRY DEPT., MSU
E. LANSING. MICH 48824

CALLING FORM: CALL INPUT
CALLING REQUIREMENTS: ALL GBA VARIABLES MUST BE STORED

IN COMMON AND THE INITIAL VALUES DEFINED IN
THE BLOCK DATA SUBROUTINE.

.LOADING REQUIREMENTS: MUST BE LOADED IN LEVEL 1.
‘EXTERNAL ROUTINES: EPLIMT.ANYCHRWEPBKV2.EPSAP,EPSAMP

EPPOS,CLOCK.EPDAC,EPSPV2.EPDARKrEPADC
EPADBM,EPPOSD,EPSAPD,PARINS

VARIABLE DESCRIPTION: SEE PROGRAM MAIN AND BLOCK DATA SUBROUHTNE

PURPOSE: THIS SUBROUTINE IS ENTERED WITH A DEFAULT SET
OF PARAMETERS WHICH MAY BE CHANGED AS DESIRED BY SETS OF
COMMANDS 2 TO 6 CHARACHTERS IN LENGTH. SOME OF WHICH
REQUIRE A NUMERICAL ARGUMENT. UPON COMPLETION CONTROL
IS PASSED BACK TO THE MAIN PROGRAM.

A MORE COMPLETE DESCRITION OF THE COMMAND SET FOLLOWS.

CHANGES VERSION 31D:

1. COMMON DEFINITIONS ARE CHANGED TO COINCIDE WITH
THE COMMON DEFINITIONS IN THE OTHER SUBROUTINES.

2. COMMANDS ARE ADDED WHICH CAUSE THE SAMPLER To
BE INITAILIZED.

3. INPUT SPECIFICATIONS FOR.SAMPLER.DELIVERY ACTIONS
MAY NOW BE ENTERED.

4. THE REAL TIME CLOCK IS STARTED IN THIS
ROUTINE BEFORE EXITING.

5. *xxNOTE*** NOT ALL THE COMMAND STRING
DOCUMENTAION AND HELP COMMAND LISTINGS HAVE BEEN
BROUGHT UP TO DATE (IO/11/76).

CHANGES *IE:

I. CHANGE ISAMPL To SAPINT SUB NAME

2. CORRECT TYPING ERRORS

CHANGES VERSION: 2. 01
I. COMMANDS ADDED TO TAKE THE PLACE OF FILE,SETUP SUBROUTINE.

.2. ADDED COMMANDS TO ALLOW BATCH STREAMING OF INPUT

SPECIFICATIONS.

‘3. ADDED COMMANDS TO ALLOW'EXECUTION OF TWO USER.SUBROUINES.

4. ADDED COMMAND TO ALLOW OUTPUT OF COMMENT TO LINE.PRINTER.
CHANGES VERSION 2.02:

I. BATCH MODE COMMAND ERROR CAUSES MODE TO GO TO LOCAL

2. DATE AND PROGRAM VERSION NUMBER.ADDED TO OUTPUT FILES
CHANGES VERSIONS 2.03 AND 2.04:

I. FILE OUTPUT FORMS ARE MODIFIED

CHANGES VERSION 2.06:

I. MODIFIED TO READ GBA SWITCH REGISTER NOT COMPUTER.SW'REG.
CHANGES VERSION 2.07:

1. FILE OUTPUT STRUCTURE CHANGED TO SHAPE 810013.6

2. BACKGROUND RUN DESOLVATE,DELAY STEPS MODIFIED

CHANGES VERSION 2.09:

1. LIMIT CHECKS MOVED TO FLOW'ROUTINE FROM INPUT.

OOOOQOOOOOOOOOOO00000000000OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO

178

CHANGES VERSION 2.10:

I. COMMAND MSG AND / ADDED TO FACILITATE BATCH STREAM OPERATION
2. COMMAND TO CHANGE FILE DATA TAG ADDED

CHANGES VERSION 3.0]:

I. EXPANDED TO INCLUDE SIMPLEX ROUTINES.

METHOD :

I. PICK UP THE CURRENT DATE FROM THE SYSTEM
2. ISSUE A PROMPT AND RING TERMINAL BELL
3. READ A COMMAND STRING IN A6 FORMAT AND STORE AS VARIABLE 'ARG'
4. IF IN BATCH MODE ECHO COMMAND TO ADDS TERMINAL
5. CHECK ARG VERSUS EACH INDIVIDUAL COMMAND POSSIBLE.
ARG EQUALS A LEGAL COMMAND? .
YES- EXECUTE THE REQUESTED TASKIOR.TASKS. GO TO 6
NO- ECHO A QUESTION MARK.AND THE ARC. GO TO 2
6. WAS ARG THE GO COMMAND?
NO- GO TO 2
YES- A. IS THERE A PARAMETER LIMIT VIOLATION?
NO- GO TO B
YES- PRINT ERROR MESSAGE. GO TO 2
B. HAVE POSITONER. DARK CURRENT. AND
AUTOSAMPLER BEEN INITIALIZED?
NO- PRINT ERROR MESSAGE GO TO 2
YES- GO TO C
C. SCANNING?
NO- SET VARIABLES SO THAT FLOW’RECOGNIZES
NOT SCANNING
YES- GO TO D
D. PRINT LINE PRINTER.HEADING IF REQUESTED (LPT)
E. SET CLOCK RATE AND CLEAR TIME TO 0 SEC
F. WAIT FOR POSITIONER TO REACH DESTINATION
‘THEN SEND TO SAMPLE DELIVERY POSITION
G. SET FIRST CALL INDICATOR TO TRUE
F. RETURN TO MAIN
COMMAND DESCRIPTION: THERE ARE THREE TYPES OF COMMANDS.

'THE 'SPECIAL COMMANDS" CAUSE AN IMMEDIATE EXECUTION

OF SOME FUNCTION BUT DO NOT CHANGE ANY VARIABLE VALUES.
THE 'CHANGE OF OPTION' COMMANDS SET THE VALUE OF A
LOGICAL VARIABLE TO .TRUE. OR .FALSE..

‘THE I'CHANGE NUMERICAL ARGUMENT' COMMANDS WHEN CALLED

IMMEDIATELY REQUEST A NEW VALUE FOR THE VARIABLE
WHICH IS IMPLIED IN THE COMMAND. THESE COMMANDS

-§HOULD BE TYPED FOLLOWED BY A CARRIAGE RETURN. AT THIS

TIME GBA MONITOR.SHOULD RESPOND WITH A ' = ' AND
THE NEW VALUE OF THE VARIABLE MAY BE ENTERED.

SPECIAL COMMANDS-

.GO BEGIN EXECUTION WTTH PRESENT ARGUMENT VALUES.
.RS RESTART INTERRUPTED TASK -NO VARIABLE INITIALIZATTON

.HL GIVE PRESENT STATUS OF ALL VARIABLES

.HLTIME GIVE HEATING TIMES

.HLPWR GIVE HEATING POWERS

.HLOPT GIVE STATUS PROGRAMMING OPTIONS

.HLSCAN GIVE SCAN VARIABLES STATUS .

.HLCOM GIVE LIST OF ALL GBA MONITOR COMMANDS

.HLLPT GIVE STATUS OF ALL VARIABLES ON THE LINE PRINTER
.HLPOS GIVE STATUS OF POSITION VARIABLES

.HLRUN GIVE TOTAL NUMBER OF RUNS REQUIRED FOR COMPLETION

mines or OPTION comps-
. PLT PLOT on ADDS
.NPLT no nor PLOT on ADDS

000000000000

96
I"

00000000000000000000000000000000002000000000000000000

179

OUTPUT DATA TO LINE PRINTER

DO NOT OUTPUT DATA TO THE LINE PRINTER

USE RADIATION PROGRAMMING FOR ATOMIZATION
USE POWER PROGRAMMING FOR ALL HEATING STEPS
USE HYDROGEN DIFFUSION FLAME

DO NOT USE HYDROGEN DIFFUSION FLAME

WAIT AFTER EACH COMPLETE RUN FOR OPERATOR INTERVENTION
DO NOT WAIT: GO TO NEXT RUN IMMEDIATELY
ALLOW SCANNING

DO NOT ALLOW SCANNING

USE AUTO SAMPLER

DO NOT USE AUTO SAMPLER

CHANGE OF NUMERICAL ARGUMMENT COMMANDS-

. DEST
. ASHT
. ATMT
. DELT
. DESP
. ASHP
. ATMP
. VER

. HOR

. VERI
. HORI
O VEMJ
. HORL
. NMRUN

. NMSCAN

.TIT'LE
.AMNT

CHANGE THE DESOLVATE TIME TO THE NEXT VALUE .ENT'ERED
CHANGE THE ASH TIME TO THE NEXT VALUE ENTERED
CHANGE THE ATOMIZE TIME TO THE NEXT VALUE ENTERED
CHANGE THE DELAY TIME TO THE NEXT VALUE ENTERED
CHANGE THE DESOLVATE POWER TO THE NEXT VALUE ENTERED
CHANGE THE ASH POWER

CHANGE THE ATOMIZE POWER

CHANGE THE VERTICAL STARTING POSITION

CHANGE THE HORIZONTAL STARTING POSTION

CHANGE THE VERTICAL INCREMENT

CHANGE THE HORIZONTAL INCREMENT

CHANGE THE VERTICAL LIMIT

CHANGE THE HORIZONTAL LIMIT

CHANGE NUMBER OF RUNS PER POSITION
CHANGE NUMBER OF SCANS

READ IN 72 CHARACHTERS OF EXPERIMENT ID
CHANGE THE SAMPLE SIZE

*NOTES ON ADDING COMMANDS
I. THE COMMAND MUST BE 6 CHARACTERS OR. LESS
2. DECIDE WHAT TYPE OF COMMAND IT IS:

A. SPECAIL I- REQUIRES MORE THAN ONE EXECUTEABLE
COMMAND. BUT DOES NOT CHANGE THE VALUE OF A VARIELE

B. SPECAIL 2- LIKE A. BUT ONLY ONE EXECUTEABLE COMMAND

C. LOGICAL- CHANGES A LOGIC VARIABLE'S VALUE

D. NUMERICAL- CHANGES VALUE OF NUMERICAL VARIABLE

3. INSERT PROPER CODE AT POINTS MARKED C*SI.C*S2.C*N. OR

C*L DEPENDING ON WHAT TYPE OF COMMAND IS TO BE INSERTED

USE THE EXAMPLE OF A SIMILIAR TYPE OF COMMAND TO DETERMINE THE
EXACT CODE TO BE INSERTED.

4. IF A NUMERICAL OR LOGICAL COMMAND IS INSERTED YOU MUST
ALSO UPDATE THE COMMON.LOGICAL. AND DATA STATEMENTS OF THE
AFFECTED PROGRAM SUCH AS MAIN AND EPBKDT.F4.

ADDITIONAL COMMANDS VERSION *IE AND VERSION 2.0!
.REWIND SET INPUT SPECIFICATION FILE BACK TO BEGINNING

. LOCAL
. BATCH
. USEI

. USE2

. NUSEI
. NUSE2
. OPEN I
. OPEN2

.CLOSEI

SET MODE SWITCH TO READ FROM ADDS TERMINAL

SET MODE SWITCH TO READ FROM INPUT BATCH STREAM FILE

SET TO CALL USER SUBROUTINE FROM DATA TAKING PROGRAM

SET TO CALL USER SUBROUTINE FROM CALCULATION SUBROUTINE
DISREGARD USER SUBROUTINE I

DISREGARD USER SUBROUTINE 2

SET TO OUTPUT INTEGRATED ABSORBANCE DATA TO FILE!

SET TO OUTPUT PEAK SHAPE RAW DATA TO FILE2

WRITE END FILE IN FILEI AND PREVENT FUTHER WRITING ATTEMPTS

000000000000000000000000000000

.CLOSE2
.HDLPT

180

AS ABOVE FOR.FIEE2
SET TO READ FROM KEYBOARD AND OUTPUT COMMENT
TO LINE PRINTER

.HEADI AND HEAD2 READ FROM KEYBORARD OUTPUT TO FILE
.COMI AND COM2 READ FROM.KEYBOARD OUTPUT TO SPECIFIED FILE

.EXIT
.DARK
.BURN
.PSIN
.SPIN
.POSDEM
.SAMPDM
.TUNE
.PARM
.FLI
.NFL2

RETURN TO KEYBOARD MONITOR.ENTER FILES IN DIRECTORY
TAKE 100 DARK CURRENT READINGS AND RETURN'THE AVERAGE
ALLOW TEMPORARY HEATING OF THE BRAID

INITIALIZE THE POSITIONER

INITIALIZE THE SAMPLER

ALLOW USER SECIFIED POSITIONER MOVEMENTS

ALLOW USER SPECIFIED SAMPLER ACTIONS

ALLOW TUNING OF POSITION OF SAMPLER ABOVE‘THE BRAID
OUTPUT INSTRUMENT PARAMETERS TO OPEN OUTPUT FILES
ACTIVATE DATA WRITING COMMANDS TO FILE I

INACTIVATE DATA WRITING COMMANDS TO FILEI

.FL2 AND .NFL2 AS ABOVE BUT FOR FILE 2

.RSBAT
.SAMPT

.SPI
.SP2
.BELL
.MSG
.TAG

RESTART DATA ACQUISTION WHERE INTERRUPTED AND

SET MODE TO REMOTE

SET TO USE AUTOSAMPLER DELIVERIES BUT DO NOT

ASK FOR SAMPLER DELIVERY SPECIFICATIONS

SPARE VARIABLE ONE

SPARE VARIABLE TWO

RINGS TERMINAL BELL REPEATEDLY UNTIL A KEY IS STRUCK.
WRITE MESSAGE TO TERMINAL IF IN BATCH MODE

CHANGE FILE DATA IDENTIFICATION TAG OUTPUT

ADDITIONAL COMMANDS VESION 3.01:

08m
.NSMP

SET TO SIMPLEX OPTIMAMIZATION MODE
TO NOT USE SIMPLEX OPTIMIZATION MODE

IO

50
100

105
106
c-

08:8 1

107

199

110

1101

III

112

181

SUBROUTINE INPUT
COMMON PLT'I'. LPT.RAD.HDF.WAIT.DEST.ASHT.ATMT.DELT.DEP
COMMON ASHP . ATMP . SCAN . VER. HOR. VERI . HORI . VERL. HORL
COMMON NMSCAN.NMRUN.SAMP.TAGI
COMMON FIRSTI . ISAMC.NSCAN.HPOS.VPOS. IRUN. ISAMP
COMMON AMNT. ISAMCA. ABS INT. XIOIDKJIIDK
COMMON FREQ. INTST. INTSP
COMMON /A/ ISAMPL( 4-0) . NRUN
COMMON /CBLK/MODE. SUBI . SUB2.FILEI .FILE2.DARKT.PINIT.STNIT.SUB3
COMMON /SPBLK/SP I . SP2 . SP3 . SP4 . SP5 .SP6
COMMON /IDEB/ITER. VINIT(5.4) .PM. ISXVARI 4) . ITYPE.LSMP.KR. IL
COMMON /ISIM/ IDIM.V(5.4) .R( 5) .ALIM(3.4) . IRET( 5) . IMV. IWRST
LOGICAL PL'IT. LPT. RAD. HDF. WAIT. SCAN. SAMP.FIRSTI

LOGICAL SUBI .SUB2.FILEI .FILE2.DARKT.PINIT.SINIT.SNT.SUB3.LSMP

DIMENSION XID( 11) .USR(11) .TYDATA(7)

DIMENSION FORMTI(2) .FORMT2(2)

DATA KID/’NO TITLE'.9*’ ‘/

DATA USR.FILL/II*' '.1*' '/

DATA FORI'IT'I.FOR.M'12/’(A2.7GIS.6)’.’(A2.IOG13.6)’/
DATA TYDATM'TMAX.HPOS.VPOS.XSAMP.AMNT.AMAX.ABSINT’/
DATA XNUI‘LYNUM/7. .50./

DATA VE/3.0l/

ASSIGN 205 TO ERROR

ASSIGN 50 TO START

GET PRESENT DATE FROM OS/8

CALL DATE( IM. ID. IY)

IY= IY-I9OO
W=1.
G=O.

=-l.
FIRSTI= .T.

WRITE(0. IO)VE

FORMAT(’ GBA VERSION: ’.F6.2)

CALL CLOCK( 500)

CALL CLTIME

BEGIN PROGRAM EXECUTION BY ISSUEING A PROMPT
WRITE(0. 100)

FORMAT( ' : ’ .0)

CALL ANYCHR(7)

PICK UP COMMAND STRING - IF IN BATCH MODE ECHO TO TERMINAL
READ(MODE.105)ARG

IF( MODE.EO..4) WRITE(O.106)ARG

FORMAT(A6)

FORMAT( ’ ' .A6)

ALLOW COMMENTS IN BATCH MODE

IF(ARG.EQ.'/’) GO TO START

INSERT LONG SPECIAL COMMANDS HERE

IO=3

IF (ARG.EQ.5HHLLPT) GO TO 107

IF HL IS REQUESTED GIVE PRESENT STATUS ALL VARIABLES
IF(ARG.NE.2HHL)GO TO 115

10:0

WRITE( IO. 109)XID. IM. ID. IY

FORMAT( ' TITLE: ’ . IIA6. I2. '/‘ . I2. ’/’ . I2./I

WRITE( IO. 1 IO)PLTT. LPT.RAD.HDF.WAIT.SCAN.SAMP.LSMP
FORMAT( ’ PLT‘ .L2. ' LPT’ .L2. ’ RAD’ .L2. ' HDF’ .L2

.' WAIT" .L2.‘ SCAN’.L2.' SAMP’.L2.' SMP ’.L2./)
WRITE( IO. 1 101)SUBI .SU82.SUB3.FILEI.FILE2.SINIT.PINIT.DARKT
FORMAT( ’ SBI’.L2.‘ SB2'.L2.' SB3'.L2.' FL1’.L3.' FL2'.L2.
’ SPIN'.L2.' PSIN'.L2.' DARK’.L2./)

WRITE( IO. 1 I 1)DELT.DEST.ASHT.ATMT

FORMAT(’ DELT'-'-‘.F7.2.' DEST‘=’.F7.2.‘ ASHT".F9.2

.’ ATMT-1'.F7.2./)

WRITE( IO. 112) DESP.ASHP.ATMP '

FORMAT(' DF8P='.F7.3.' ASHP=‘.F7.3.’ ATMP'-".F9.3./)

113

117

120

1202

140

0*82
141

1411

I

1

”HH—

~_H~

182

WRITE( IO. 113JVER. HOR~AMNT.NMRUN.FREQ

FORMAT(' = ’.F8. 4.’ HOR=‘.F8.4.’ AMNT".F9.2.
' NMRUN=‘.16.’ FREQ= '.F6.I./)

IF (.NOT.SCAN) GO TO START
WRITE(IO.114JVERI.HORI.VERL.HORL.NMSCAN

FORMAT(’ VERI='.F7.4.' HORI='.F7.4.' VERL='.F9.4
.' HORL='.F7.4.' NMSCAN=’.I2./)

GO TO START

IF HLRUN REQUESTED CALCULATE NUMBER OF RUNS TO COMPEETE TASK
IF(ARG.NE.5HHLRUN) GO TO 120

N=I

M=l

L31 '

IF(HORI.NE.O.) N=HORL/HORI

IF(VERI.NE.O.) M=VERL/VERI

IF (NMSCAN.GT.O) L=NMSCAN

‘IRUNT=NMRUN*N*M*L

WRITE(O.117)IRUNT

FORMAT(’ TOTAL NUMBER OF RUNS REQUIRED='.I§)

GO TO START

IF COMMAND IS GO FIRST CHECK FOR.LIMIT VIOLATIONS

IF (ARG.EQ.’RS‘) RETURN

IF (ARG.NE.’RSBAT’) GO TO 1202

MODE=4

RETURN

IF (ARG.NE.2HGO) GO TO 140

FIRSTI=.T.

RETURN

GIVE HELP INFORMATION IF REQUESTED

IF(ARG.EQ.’HLOPT’) WRITE(O.1IO)PLTT.LPT.RAD.HDF.WATT.SCAN.SAMP
IF(ARG.EQ.'HLTIME’) WRITE(0.III)DELT.DEST.ASHT.ATMT
IF(ARG.EQ.'HLPWR') WRITE(0.II2)DESP.ASHP.ATMP
IF(ARG.EQ.’HLPOS') WRITE(O.113)VER.HOR.AMNT.NMRUN.FREQ
IF(ARG.EQ.’HLSCAN’)WRITE(0.114)VERI.HORI.VERL.HORL.NMSGAN
IF(ARG.EQ.'HLCOM’)WRITE(0.141)

INSERT SHORT SPECAIL COMMANDS HERE

FORMAT(’ GO. HL. HLTIME. HLPWR. HLOPT. HLSCAN. HLCOM.THLLPT'./.
' HLPOS. HLRUN. PLT. NPLT. LPT. NPLT. RAD. PWR?./

’ HDF. NHDF. WAIT. NWAIT. SCAN. NSCAN. SAMP. NSAMP'./.

' DEST. ASHT. ATMT. DELT. DESP. ASHP. ATMP‘./.

' VER. HOR. VERI. HORI. VERL. HORL. TITLE')
IF(ARG.EQ.’HLCOM') WRITE(0.1411)

‘FORMAT(' NMRUN. NMSCAN. REWIND. LOCAL. BATCH. USEI. NUSEI’./

’ USE2. NUSE2. OPENI. CLOSE]. OPEN2. CLOSE2.'./

’ HDLPT. HEADI. HEADZ. COMI. COM2. DARK. TUNE’./,

’ BURN. POSINT. SPIN. POSDEM.‘SAMPDM. USE3. NUSE3‘./,
’ FL1. NFLI. FL2. NFL2. RS. RSBAT. BELL. MSG')

C*SI.82.L.N INSERT NEW COMMANDS IN LISTING OF COMMANDS ABOVE

C

IF(SINIT)SNT=. T.
IF(.NOT.SINIT)SNT*.F.
THE FOLLOWING IF STATMENTS CHANGE LOGICAL VARIABLE VALUES

IF (ARG.EQ.3HPLT) PLT'I‘=.T'RUE.
IF (ARG.EQ.4HNPET) PLTT=.FALSE.
IF (ARG.EQ.3HLPT) LPT=.TRUE.
IF (ARG.EQ.4HNLPT) LPT=.FALSE.
IF (ARG.EQ.3HRAD) RAD=.TRUE.
IF (ARG.EQ.4HNRAD) RAD=.FALSE.
IF (ARG.EQ.3HHDF) HDF=.TRUE.
IF (ARG.EQ.4HNHDF) HDF=.FAI.SE.
IF (ARG.EQ.4HWAIT) WAIT=.TRUE.
IF (ARG.EQ.5HNWAIT) WAIT=.FALSE.
IF (ARC.EQ.4HSCAN) SCAN=.TRUE.
IF (ARG.EQ.5HNSCAN) SCAN=.FALSE.
IF(ARG.EQ.4HSAMP) SAMP=.TRUE.

IF(ARG.EO»4HSAMP) CALL SAP

14115

C*L

I412

c*SZ.L

I42

HHHI‘HD‘

C*N

“Hi-H

I43

145
I50

170
180
185

183

IF(ARG.EQ.5HNSAMP) SAMP8.FALSE.
IF (ARG.EQ.4HDARK) CALL DARK
IF (ARG.EQ.4HTUNE) CALL TUNE(SNTT

IF (ARG.EQ.6HPOSDEM) CALL POSDEM

IF (ARG.EQ.6HSAMPDM) CALL SAPDEM

IF(ARC.EQ.4HBURN) CALL BURNDM

IF (ARG.NE.4HPSIN) GO TO 14115

CALL POS(Q.G.G)

CALL POS(W.G.G)

CALL POS(G.G.G)

PINIT=.T.

IF (ARG.NE.4HSPIN) GO TO 1412

SINIT=.TRUE.

CALL SAPINT

INSERT NEW LOGICAL COMMANDS BEFORE'THIS COMMENT AND
IN IF STATEMENT 142

IF NOT SCANNING SET SCAN PARAMETERS SO’THAT MAIN PROGRAM
RECOGNIZES THIS

IF(SCAN) GO TO 142

HORI=O.

VERI=O.

HORL=HOR

VERL=VER

SHORT SPECIAL COMMANDS AND LOGICAL COMMANDS

IF ADDED TO THE PROGRAM SHOULD ALSO BE ADDED‘TO‘THE FOLLOWING
LOGICAL IF STATEMENT WHICH CHECKS FOR LEGAL COMMANDS
IF(ARG.EQ.3HPLT.OR.ARG.EQ.4HNPLT.OR.ARG.EQ.3HLPT
.OR.ARG.EQ.4HNLPT.OR.3HRAD.EQ.ARG.OR.ARG.EQ.4HNRAD.OR~
ARG.EQ.3HHDF.OR.ARG.EQ.4HNHDF.OR.ARG.EQ.4HWAITr0R.ARC
.EQ.5HNWAIT.OR.ARG.EQ.4HSCAN.OR.ARG.EQ.5HNSCAN.OR~ARG.EO.
5HHLCOM.OR.5HHLPWR.EQ.ARG.OR.ARG.EQ.6HHLTIME.OR.
ARC.EQ.6HHLSCAN.OR.ARG.EQ.5HHLPOS.OR.ARG.EQ.4HSAMP.OR.
ARG.EQ.5HHLOPT§OR.ARG.EQ.5HNSAMP)GO TO START
IF(ARG.EQ.4HBURN.OR.ARG.EQ.4HTUNE)GO TO START
IF(ARG.EQ.6HPOSDEM.OR.ARG.EQ.6HSAMPDM)GO TO START
IF(ARG.EQ.4HFILE.OR.ARG.EQ.4HDARK) GO TO START
IF(ARC.EQ.4HSPIN.OR.ARG.EQ.4HPSIN)GO TO START

IS THIS A CHANGE NUMERICAL ARGUMENT CONNAND

‘WHEN NEW NUMERICAL ARGUMENT COMMANDS ARE ADDED'THEY

MUST ALSO BE ADDED TO THIS IF STATEMENT .
IF(ARC.EQ.4HDESP.OR.ARG.EQ.4HASHP.0R.ARG.EQ.4HATMP.OR.
ARG.EQ.4HDELT.OR.ARG.EQ.4HASHT.OR.ARG.EQ.4HATMT.OR.
ARG.EQ.5HNMRUN.OR.ARG.EQ.4HDEST.OR.ARG.EQ.6HNMSCAN.OR.
ARG.EQ.3HVER.OR.ARG.EQ.4HAMNT.OR.ARG.EQ.3HHOR.OR.
ARG.EQ.4HVERI.OR.ARG.EQa4HHORI.OR.ARG.EO.4HVERL.OR.
ARG.EQ.4HHORL.OR.ARG.EQ.5HTITLE.OR.ARG.EQ.4HFREQ) GO TO 145
IF(ARG.EQ.’SPI’.OR.ARG.EQ.’SP2'.OR.ARG.EQ.’SP3' '
.OR.ARG.EQ.’SP4’.OR.ARG.EQ»'TAG’) GO TO 145

GO TO 300

WRITE(0.I50)

FORMAT(' ='..)

IF(ARO.EQW4HDEST) READ(MODE.170)DEST

IF(ARG.EQ.4HDELT) READ(MODE.I70)DELT

IF(ARG.EQn4HASHT) READ(MODE.170)ASHT

IF(ARG.EQ.4HATMT) READ(MODE.I70)ATMT

IF(ARG.EQ.4HFREQ) READ(MODE.I70)FREQ

IF(ARG.EQ.4HDESP) READ(MODE.180)DESP

IF(ARG.EQ.4HASHP) READ(MODE.180)ASHP

IF(ARG.EQ.4HATMP) READ(MODE.180)ATMP
IF(ARG.EQ.5HNMRUN) READ(MODE.185)NMRUN
IF(ARG.EQ.6HNMSCAN) READ(MODE.185)NMSCAN

'FORMAT(F6.2)

FORMAT(F5.3)
FORMAT( I4)
IF(ARG.EQ.’SPI’) READCMODE.180)SPI

190
C*N

c-
205

210
c-

c-
300

303
C-
305

310

320
c-
330

340
350

360

370
380

390

391

~400

401

1814

IF(ARG.EOm'SP2') READ(MODE.180)SP2
IF(ARG.EQ.'SP3') READ(MODE.180)SP3
IF(ARG.EQ.’SP4’) READ(MODE.180)SP4
IF(ARG.EQ.'TAG') READ(MODE.105)TAGI

CONTINUE CHECKING FOR NUMERICAL ARGUMENT COMMANDS
IF(ARG.EQ~3HVER) READ(MODE.I90)VER
IF(ARG.EQ.3HHOR) READ(MODE.I90)HOR
IF(ARG.EQ.4HVERI) READ(MODE.I90)VERI
IF(ARG.EQ.4HHORI) READ(MODE.190)HORI
IF(ARG.EQ.4HVERL) READ(MODE.I90)VERL
IF(ARG.EQ.4HHORL) READ(MODE.I90)HORL
IF(ARG.EQ.5HTITLE)READ(MODE.310)XID
IF(ARG.EQ.4HAMNT) READ(MODE.I70) AMNT
FORMAT(F7.5)

INSERT NEW NUMERICAL COMMAND IF CHECKS HERE

GO TO START

IF ILLEGAL COMMAND RESPOND WITH ERROR MESSAGE
WRITE(0.2IO)ARG

FORMAT(’ ?’.A6)

IF IN BATCH MODE SET TO LOCAL MODE IF ERROR.IS DETECTED
MODE=0

GO TO START

THIS PAGE WAS ADDED TO CONTAIN VERSION 2.01 COMMANDS
COMMAND USED TO WRITE A MESSAGE ON THE TERMINAL (BATCH‘MODE)
IF(ARG.NE.'MSG’) GO TO 305

READ(MODE.310)USR

WRITE(0.303)USR

FORMAT(' MSG’.IIA6)

COMMAND TO PUT COMMENT ON THE LINE PRINTER

IF (ARG.NE.'HDLPT’) GO TO 330

READ(MODE.310)USR

FORMAT(11A6)

WRITE(3.320)USR

FORMAT(1H0.11A6)

COMMENTS AND READINGS TO FILES HANDLED HERE
IF(ARG.NE.'HEADI’) GO TO 350

IF(.NOT.FILEI) GO TO 415

‘READ(MODE.310)USR

WRITE(5.340)USR
FORMAT(‘HD’.IIA6)
IF(ARG.NE.’HEAD2’) GO TO 360
IF(.NOT.FILE2) GO TO 415
READ(MODE.310)USR
WRITE(6.340)USR
IF(ARG.NE.’COMI’) GO TO 380
IF(.NOT.FILEI) GO TO 415
READ(MODE.310)USR
WRITE(5.370)USR
FORMAT(‘CC'.IIA6)
IF(ARG.NE.'COM2’) GO TO 390
IF(.NOT.FILE2) GO TO 415
READ(MODE.310)USR
WRITE(6.370)USR
IF(ARG.NE.'CLOSE1’) GO TO 400
IF(.NOT.FILE1) GO TO 415
WRITE(5.39I)

FORMAT(‘EF‘)

END FILE 5

'FILEI=.F.

IF (ARG.NE.’CLOSE2') GO TO 401
IF(.NOT.FILE2) GO TO 415
WRITE(6.391)

END FILE 6

FILE2=.F.

IF(ARG.NE.'OPENI’) GO TO 404

402
403
4031

4033
4034

4035
4036
404

405

406

410

69
70

71

72

73

74

75

c-
413

185

FILEI'.T.

WRITE( 5 .402) FORMTI . XNUM
FORMAT(’FO'.2A6/'FN'.G13.6)
WRITE(5.403)VE.TYDATA
FORMAT('VE'.GI3.6.7A6)
WRITE(5.403I)IM.ID.IY

'FORMAT(’TI’.3I3)

IF(.NOTzLSMPICO To 404 .
NRITE(5.4033)ITYPE.xn.IL,(ISXVANIJ),J=I.IDIM)
TONMATI 'SM’ .° ITYPE‘ .143 IGIUN’.I4.‘ 1m LMT'.I4.' VAR 0PT‘4I4)
WRITE(5.4034)((ALIM(I.J).I=I.3).J=I.IDIM)
FORMAT(‘SL'.3GI3.6)

no 4035 I=I.IDIM+I
WRITE(5.4036)(VINIT(I.J).J=1.IDIM)
FORMAT(’SI'.4GI3.6)

IF(ARG.NE.’0PEN2') GO TO 405

FILE2=.T.

WRITE(6.402)FORMM2.YNUM

WRITE(6.403)VE

WRITE(6.403I)IM.ID.IY

ADDITIONAL COMMANDS VERSION 3.01
IF(ARG.NE.’SMP’) CO TO 406

LSMP=.T.

CALL PARIN

co TO START

IF(ARG.NE.'NSMP’) GO TO 410

LSMP=.F.

GO TO START

IF (ARG.NE.’PARM’) GO TO 413

1FN=5

IF(.NOT.(FILEI.OR.F1LF2)) CO To 415
IF(FILE2.AND..NOT.FILEI) IFN=6
WRITE(IFN.70)PLTT.LPT.RAD.HDF.WAIT.SCAN,SAMP
FORMAT(‘PA'.’ PLTT’.L2.’ LPT’.L2.’ RAD’.L2.’ HDF‘.L2.
' WAIT’.L2.' SCAN‘.L2.' SAMP’.L2)

-WRITE(IFN.7I)SUBI.SUB2.SUB3.FILE].FILE2.SINITSPINIT§DARKT

FORMAT(‘PB’.’ SBI’.L2.' SB2’.L2.' SB3'.L2.' FL1'.L2.
' FL2'.L2.' SINT’.L2.’ PINT'.L2.’DARKT'.L2)
WRITE(IFN.72)DELT.DEST.ASHT.ATMT

FORMAT(‘PC’.’ DELT’.GI3.6.' DEST‘.GI$.6.' ASHT'.GI3.6.

' ATMT" .GI3.6)

WRITE(IFN.73)DESP.ASHP.ATMP.FREQ

FORMAT(‘PD’.’ DESP‘.G13.6.' ASHP'.G13.6.’ ATMP'.GI3.6.

' FREQ‘.G13.6)

WRITE(IFN.74)VER.HOR~AMNT.NMRUN

FORMAT(‘PE'.’ VER’.GI3.6.’ HOR‘.GI3.6.' AMNT’.GI3.6.
’NMRUN'.GI3.6)
IF(SCAN)WRITE(IFN.75)VERI.HORI.VERL.HORL{NMSCAN ~
FORMAT(’ VERI'.GI3.6.' HORI’.GI3.6.' VERL'.G13.6.’ HORL'.
GI3.6.'NSCAN'.GI3.6)

IF(IFN.EQAG) GO TO 413

IFN=6

IF(FILE2.AND.FILE1) GO TO 70

LOGICAL TYPE COMMANDS

IF (ARG.EQ.’REWIND') REWIND 4

IF (ARG.EQ.'LOCAL’) MODE=0

IF (ARG.EQ.’BATCH') MODE=4

IF (ARG.EQ.'USEI') SUBI=.T3
IF (ARG.EQ.’USE2') SUB2=.T.
IF (ARG.EQ.‘USE3') SUB3=.T.
IF (ARG.EQ.'NUSE1') SUB1=.F.
IF (ARG.EQ.'NUSE2') SUBZ=.F.
IF (ARG.EQ.’NUSE3‘) SUB3=.F.
IF (ARG.EQ.'EXIT') CALL EXIT

IF (ARG.EQ.'FL2’) FILE2=.T.

186

IF (ARG.EQ. ’NFL2’) FILE2=.F.
IF (ARG.EQ. ’NFLI’) FILEI'.F.
IF (ARG.EQ.‘FLI‘) FILEI=.T.
IF (ARG.EQ. 'SAMPT’) SAMP=.T.
IF (ARG.EQ.’BELL’) CALL BELL
C- NOW CHECK TO SEE IF IT WAS A LEGAL COMMAND

IF(ARG.EQ. ’REWIND' .OR.ARG. EQ. ’LOCAL’ .OR.ARG.EQ. 'BATCH’ .OR.
ARG.EQ. ’USEI ’ .OR.ARG.EQ. ’USE2’ .OR.ARG.EQ. ’NUSEI ’ .OR.
ARG.EQ. 'NUSE2’ .OR.ARG.EQ. 'OPENI ' .OR.ARG.EQ. 'CLOSEI ’ .OR.
ARG.EQ. ‘OPEN2' .OR.ARG. EQ. ’CLOSE2’ .OR. ARG.EQ. ’HDLPT' .OR.
ARG.EQ. ’HEADI ’ .OR.ARG.EQ. ’HEAD2' .OR.ARG.EQ. 'COMI ' .OR.
ARG.EQ. 'COM2'.OR.ARG.EQ. 'USE3’ .OR.ARG.EQ. ’NUSE3') GO TO 420
IF( ARG.EQ. ‘PARM' .OR.ARG.EQ. 'FLI ’ .OR.ARG.EQ. 'NFLI ’ .OR.
ARG.EQ. ’FL2’ .OR.ARG.EQ. 'NFL2’ .OR.ARG.EQ. ’SAMPT' .OR.

ARG.EQ. 'BELL' .OR.ARG.EQ. 'MSG’) GO TO 420

GO TO ERROR

fl—Hflfl

“H

415 WRITE(0.416)

416 FORMAT(' N01! FILE IS NOT OPENED’)
420 D0 430 IK=I.6

430 USR( 1K) =FILL

GO TO START
END

000000000000000000000000000000000000000

187

PROGRAM FUNCTION: PROGRAM SUPERVISEE PARAMETER CHANGES NECBSARY

FOR MULTIPLE SAMPLE RUN EXPERIMENTS
PROGRAMMER: E. H. PALS
VERSION: *IE-I (2.04)
DATE: 10/9/76
FILENAME: EPFLOW.F4

CALLING FORM: CALL FLOW

EXTERNAL ROUTINES: NONE

SUMMARY: THE PRIMARY FUNCTION OF "THIS ROUTINE IS'TO
KEEP TRACK OF WHERE THE POSITIONER. SAMPLER. ETC. ARE AT
AND WHAT CONDITIONS ARE REQUIRED FOR THE NEXT REQUESTED
SAMPLE RUN. THE ROUTINE ALSO DETECTS THE COMPLETION OF
A TASK AND TAKES APPROPRIATE RE- INITIALIZATION STEPS.
CHANGES VERSIONNHB:

1. ADD THE COMMON AND LOGICAL SPECIFICATIONS

CHANGE VERSION *IC:

1. DEFINE ISAMPL AS LABELED COMMON ARRAY

CHANGES VERSIONS 1E AND 2.01:

1. UPDATE COMMENTS

2. UPDATE COMION STATEMENT AND FIX POSITIONER CALL
3. INSERT SWITCH REGISTER OPTION CHECKS.

4. FIX MULTIPLE RUN WHEN NOT SAMPLING OR SCANNING

CHANGES VERSION 2.02:
I. FORCE TO SET TO LOCAL MODE IF ERROR DURING BATCH STREAMING

CHANGES VERSION 2.03:
1. LIMIT CHECKS MOVED HERE FROM INPUT.
2. OPTION TO WAIT WHEN NOT PLOTTING ADDED.

000000OOOOOOOOOOOOOOOOOODOOOOOO0000000000000

188

"METHOD:

1. FIRST CALL?
YES- INITIALIZE RUN COUNTER AND POSITION POINT'ERS.
NO- GO TO 4

2. SCANNING OR SAMPLING?
YES- INITIALIZE SCAN PARAMETERS.
NO- . GO TO 4

3. SAMPLING?
YES- DECODE SAMPLE SPECIFICATION WORD AND IF IT

IS ZERO GO TO I3 OTHERWISE GO TO 12.

NO- GO TO 4

4. COMPLETED REQUESTED NUMBER OF RUNS YET?
YES- GO TO 5
NO- GO TO 12

5. NOT SAMPLING OR SCANNING?
YES- GO TO TASK COMPLETE ROUTINE
NO- GO TO 6

6. SAMPLING?
YES- GO TO 7

‘2 NO- GO TO 8

7. SAMPLE RUN COUNTER EQUALS REQUESTED NUMBER OF RUNS?
YES- GO TO 8
NO- GO TO 12

8. REACHED VERTICAL LIMIT?
YES- REINITIALIZE VERTICAL POSITION AND GO TO 9
NO- GO TO 12

9. COMPLEMENT HORIZONTAL DIRECTION

IO. REACHED HORIZONTAL LIMIT IN POSITIVE DIRECTION?
YES- REINITIALIZE GO TO 11
NO- GO TO 12

II. COMPLETED REQUESTED NUMBER OF SCANS?
YES- GO TO 13
NO- GO TO 12 '

12. READ SWITCH REGISTER AND EXECUTE REQUESTED FUNCTIONS.
SWB- GO TO 13 (INTERUPT TASK BEFORE COMPLETION)
SW5- WRITE SELECTED GENERAL PARAMETERS ON TERMINAL
SW6- WRITE SELECTED SCAN VARIABLE STATUS TO TERMINAL
SW7- WRITE SELECTED SAMPLING VARIABLE STATUS TO TERMINAL

I3. QUERRY OPERATOR TO ALLOW EXIT TO GBA MONITOR OR TO

RESTART DATA ACQUSITION WITH PRESENT CONDITIONS.

1000
1010

I201

121

I 25
122

189

' SUBROUTINE FLOW

COMMON PLTI‘.LPT.RAD.HDF.WAIT.DEST.ASHT.ATMT.DELT.DFSP
COMMON ASHP.ATMP.SCAN.VER.HOR.VERI.HORI.VERL.HORL
COMMON NMSCAN.NMRUN.SAMP.SAMPSZ

COMMON FIRST! . ISAMC.NSCAN.HPOS.VPOS. IRUN. ISAMP

COMMON AMNT. ISAMCA. TAG. XIOIDK. XIDK

COMMON FREQ

COMMON /A/ISAMPL( 40) .NRUN. IGOTO

COMMON /B/DATAI( 100) .DATA2( 100) .AMIN.AMAX. IDIR

COMMON /CBLK/MODE.SUBI .SUB2. FILE] .FILE2.DARKT.PINIT. SINIT.SUB3
LOGICAL PLTT'.LPT.RAD.HDF.WAIT.SAMP.SCAN.FIRSTI
LOGICAL DARKT.SINIT.PINIT

ASSIGN I 'IO'DECODE

ASSIGN 2 TO RETRN

ASSIGN 3 TO EXIT

ALLOW WAITING HERE WHEN IN WAIT MODE AND NOT PLOTTING

IF (.NOT.(.NOT.PLT'I‘.AND.WAIT))GO TO 1100
WRITE“). 1010)

FORMAT(’ CONTINUING) OR TO GBA MONITORU‘D? ’.0)
READ(O.II)AA

IF(AA.EQ.’C’) GO TO 1100

IF(AA.NE.’M’) GO TO 1000

MODE = 0

IGOTO = I

RETURN

DETERMINE IF INITAIL VARIABLE SETUP NECESSARY

IF(.NOT.FIRSTI) GO TO 100

IGOTO=3

IF(LIMIT(VERL.0. .26.01. ’VERL') .GT.0) IGOTO’I
IF(LIMIT(HORL.-I2.7.+l2.7.’HORL').GT.0) IGOTO=I

INSERT VARIABLE LIMIT VIOLATION CHECKS HERE

IF (PINIT.AND.DARKT.AND.(SINIT.OR..NOT.SAMP)) GO TO 121
WRITE(0.1201)

FORMAT( ’ NON EITHER DARK CURRENT. SAMPLER. OR POSITIONER’
’NOT INITIALIZED' ./)

IF NOT SCANNING SET SCAN VARIABLES TO COINCIDE WITH THIS
IGOTO = 1

IF( IGOTO.EQ. URETURN

IF(SCAN) GO TO 125

HORI=0.

VERI=0.

HORL=HOR

VERL=VER

IF REQUESTED WRITE READING ON LINE PRINTER OUTPUT

IF( LPT) WRITE(3. I22)

FORMAT( 1H0. ’SAMP' .2X. 'AMOUNT’ .3X. 'HPOS' .SX. ’VPOS'

.UX. 'TMAX' .3X. ’ABSMAX‘ .SX. ' INTABS')

MAKE SURE ALL FLAGS SET BEFORE EXITING

CALL CLOCK( 500)

CALL CLTIME

WAIT FOR POSITIONER TO INITAILIZE

CALL POS(W.0. .0.)

XNULL=0.

HPOS=HOR

VPOS=VER

'SEND POSITIONER TO SAMPLE DELIVERY POSITION

CALL POS(XNULL. HULL. HULL)

MAKE SURE RAD FLAG INITIALLY CLEAR
CALL FLAG(- I..0..0. )

NRUN=0

G=0.

$000

~C§OC5OC§

000

I03
I02

190

IDIR=I

ISAMC=0

NSCAN=0

HPOS=HOR.

VPOS=VER
IF(.NOT.(SCAN.OR.SAMP)) GO TO 2
IF(HORL-HPOS)9.8.9

HORL=HORL-I.
IF(VERL-VPOS)16.17.16

VERL=VERL-I.
IF (.NOT.SAMP) GO TO RETRN
ISAMCA=0

IF SAMPLE SPECIFICATION EQUALS ZERO GO TO EXIT ROUTINE

ISAMC=ISAMC+I
IXX=ISAMPL(ISAME)
IF (IXX)3.3.4

DECODE SAMPLE SPECIFICATION WORD

Y3FLOAT(ISAMPL(ISAMC))/l00000.

IRUN=IFIX(Y)

ISAMP=IFIX((Y‘FLOAT(IRUN))*I00.)
AMNT=((Y‘FLOAT(IRUN)-(FLOAT(ISAMP)/100.))81000.)+.005

IF NOT FIRST CALL BEGIN CHECKING RUN VARIABLES
CHECK RUN COUNTER

NRUN=NRUN+I
IF(NMRUN-NRUN)101.IOI.2

NRUN=0

IF(.NOT.(SAMP.OR.SCAN)) GO’TO 3
IF(.NOT.SAMP)GO TO 102
IF(SAMP.AND.SCAN) GO TO 19

CHECK SAMPLE RUN COUNTER

ISAMCA=ISAMCA+I
IF(IRUN-ISAMCA)103.103.2

ISAMCA=0

IF(SAMP.AND..NOT.SCAN) GO TO DECODE

CHECK VERTICAL POSITION

VPOS=VPOS+VERI
IF(VERL-VPOS)20.2.2
VPOS=VER

COMPLEMENT’HORTZONTAL DIRECTION

IDIRF-IDIR

CHECK HORIZONTAL POSITION

HPOS=ABS(HPOS)+HORI
IF(HORL-HPOS)23.2.2
HPOS=-HPOS

GO TO 2

HPOS=HOR

CHECK NUMBER OF SCANS
NSCAN=NSCAN+I

IDIR=I

110

”GOO

398
399
400
40 l
402
403

404
405

I-COOOO

191

IF(NMSCAN-NSCAN) 110. 110.2
NSCAN=0

IF( .NOT.SAMP) GO TO EXIT
GO TO DECODE

ALLOW TASK INTERUPTION OR RUN COUNTING. PROGRESS REPORT

IGOTO = 3

XPT‘S=ATMT*FREQ

IF(LIMIT(DEST‘.1)..20..‘DEST’).GT.0) IGOTO=!
IF(LIMIT(ASHT.0..30..’ASHT').GT.0) IGOTO=!
IF(LIMIT(ATMT.0..5.0.‘ATMT’).GT.0) IGOTO=]
IF(LIMIT( DELT,5. .60. . ’DELT' ) .GT.0) IGOTO= I
IF(LIMIT(DESP.0..10.00.’DEBP’).GT.0) IGOTO=!
IF(LIMIT(ASHP.0..I0.00.'ASHP’).GT.0) IGOTO=!
IF(LIMIT(ATMP.0..I0.00.’ATMP').GT.0) IGOTO=]
IF(LIMIT(VER.0..26.01.'VER').GT.0) IGOTO=I
IF(LIMIT(HOR.-l2.7,+l2.7.'HOR’).GT.0) IGOTO=!
IF(LIMIT(VERI,-26.0I.+26.0I.'VERI').GT.0) IGOTO=!
IF(LIMIT(HORI.-26.0l.+26.0l.’HORI').GT.0) IGOTO=!
IF(LIMIT(AMNT.0..200..'AMNT’).GT.0) IGOTO=!
IF(LIMIT(FR.EQ.I..1000..'FREQ’).GT.0) IGOTO=]
IF(LIMIT(XPT‘S.I..1000..’NUMPT‘S’).GT.0)IGOTO=I

IF (IGOTO.EQ.3) GO TO 398

MODE-=0

RETURN

IF (SWITCH(0.).EQ.0.) GO TO 399

MODE=0

GO TO 3

IF (SWITCH(I.).EQ.0.) GO TO 401

WRITE(0.400) FIRSTI.SCAN.SAMP.NMRUN.NRUN.HPOS.VPOS
FORMAT( ' FIRST] SCAN SAMP NMRUN NRUN HPOS VPOS‘./
.L4.L7.L5.I5.I6.F6.2.F5.2)

IF(SWITCH(2.).EQ.0.) GO TO 403

WRITE(0.402) IDIR.NMSCAN.NSCAN.VER. HOR.VERL.HORL
FORMAT(’ IDIR NMSCAN NSCAN VER HOR VERL HORL’,/
.I4.2I6.3X..4F5.2)

IF(SWITOH(3.).EQ.0.)GO TO 405

WRITE(0.404) ISAMC. ISAMCA. IRUN. ISAMP.AMNT

FORMAT(' ISAMC ISAMCA IRUN ISAMP AMNT'./
.I4.I6.16.I5.9X.F5.2)

'FIRSTIP- .F.

I GOTO= 3
RETURN

ALLOW EXIT TO GBA MONITOR

WRITE(0.15)

FORMAT( ’ EXIT TO GBA MONITOIK M) OR DATA TAKING( D) ')
READ( MODE. 1 I) AA

FORMAT( Al)

IF( AA.EQ. 'M' ) IGOTO= I

IF( AA. EQ. ’D’ ) IGOTO=-3

FIRST]8 .T.

RETURN

END

OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO

192

COMBINED ROUTINE TO ACQUIRE DATA AND CAIEULATE RESULTS

‘FUNCTION: THIS ROUTINE HANDLES DATA ACQUISTION

PROGRAMMER: E. H. PALS

FILNAME: EPDATA.F4
DATE: 9/10/76 (4/11/77)
VERSION: *IB (2.10)-T

CALLING FORM: CALL'DATA

PURPOSE: THIS ROUTINE CONTROLS THE TAKING AND STORING
OF RAW DATA FROM ONE SAMPLE RUN AND CALCULATES
INTEGRATED ABSORBANCE.

CHANGES VERSION *IB:
1. FLAG ROUTINE CALLED TO SET RAD FLAG. EPCURT NO LONGER
A SUBROUTINE.

CHANGES VERSION 2.0!:

I. DESOLVATE BEFORE TAKING 100%T DURING POSITIONER MOVE
2. ADD OPTION TO CALL USER SUBROUTINE WHEN DATA TAKEN
3. USE FAST A/D ACQUISTION ROUTINES

CHANGES VERSION 2.02:

I. FORMAT OF FILE WRITING CHANGED

CHANGES VERSION 2. I0:

I. DATA FILE OUTPUT FORMAT IS CHANGED TO ALLOW USER
DEFINITION OF DATA TAG ALSO OUTPUT ORDER IS ALTERRED.

CHANGES VERSION T:

I. MODIFIED TO ALLOW BURN OFF OF SAMPLE BETWEEN SAMPLE
AND BACKGROUND RUNS .

'EXTERNAL ROUTINES REQUIRED:

I . FREQS( FREQ) .SET( ARRAY. NCON. IERR) .GO
FILE-EPADCR.RA AND EPADBM.RA
FREQS- STARTS T'HE ADC CLOCK AT REQUESTED FREQUECY
SET- PASSES REQUESTED NUMBER OF CONVERS IONS DATA ARRAY
AND IERR INFORMS USER OF ILLEGAL CALLS
GO- START‘S THE DATA TAKING PROCESS SET UP BY FREQS AND GO
2. DAC( VOLT) FILE- EPDAC. RA
OUTPUTS VOLT VOLTS AT REFERENCE
INPUT OF GBA POWER SUPPLY. ‘
3. POS(I.HPOS.VPOS) ‘ FILE- EPPOS.RA
I=-I: HARDWARE RESET OF POSITIONER
1= 1: WAIT FOR POSITIONER TO REACH DESTINATION
I: 0: START MOVE TO POSITION HPOS.VPOS.
4. SAMP(DATAI.DATA2) FILE- EPSAMS.RA
DELIVER AMNT MICROLITERS OF SAMPLE
SOLUTION NUMBER ISAMP. IF ISAMP<0
WAIT FOR COMPLETION BEFORE RETURNING.
5. CLOCK ROUTINES FILE- CLOCK.RA
CLOCK( FREQ)- START CLOCK AT FREQ FREQUENCY
CLTIME- CLEAR CLOCK COUNTER
TIME(SEC)- RETURN TIME IN SEC SINCE CLTIME
WAS LAST CALLED.

000000000000000000000000000000000000

I.
2.
3.
4.

5.
6.
7.

8.
9.

10.
II.
‘2.
13.
I4.
‘5.

16.
I7.
18.
I9.

20.
21.
22.

23.

193

'METHOD:

WAIT FOR POSITIONER TO REACH SAMPLE DELIVERY POSITION
WAIT FOR DELAY TIME TO ELAPSE
IF REQUESTED DEPOSIT SAMPLE
LATCH DESOLVATE POWER TO DAC AND WAIT FOR
DESOLVATE TIME TO ELAPSE
TAKE 100 PTS 0F IOOZT AND CALCULATE AVERAGE
ASH AT ASH POWER FOR REQUESTED ASH TIME
SET UP GBA ADC CLOCK TO REQUESTED FREQUENCY. POINT RESULTS TO
ARRAY XVAL (DONE BY CALLING FREQS) '
SET RADIATION PROGRAMING FLAG IF REQUESTED
LATCH TO DAC THE ATOMIZATION POWER AND WAIT FOR ATOMIZATION
TIME TO ELAPSE
CLEAR RADIATION PROGRAMMING FLAG. SET DAC TO 0 VOLTS
WAIT FOR DELAY TIME TO ELAPSE
RE-EXECUT'E STEPS 4.6-10 BUT DEPOSIT RESULTS IN BACK NOT XVAL
CLEAR CLOCK TO 0 SECONDS ( I.E. START DELAY TIME)
START POSITIONER TOWARD SAMPLE DELIVER POSITION.
ALLOW EXECUTION OF A USER SUBROUTINE AND IF ISKP SET=O IN THE
SUBROUTINE GO TO STEP l9. OTHERWISE CONTINUE (THIS ALLOWS AN
ALTERNATE CALCULATION ROUTINE)
INITIALIZE VARIABLES FOR CALCULATIONS
CALCULATE THE INTEGRATED ABSORBANCE
FIND ABSORBANCE MAXIMUM.MINIMUM AMD TIME OF MAXIMUM ABS
ALLOW EXECUTION OF A USER DEFINED SUBROUTINE. IF ISICP IS SET
IN THE SUBROUTINE TO 0 EXIT TO MAIN. IF ISICP 8 I GO TO 22
IF ISKP NOT 0 OR I CONTINUE TO 20
WRITE RESULTS ON LINE PRINTER IF REQUESTED
WRITE RESULTS 0N EITHER OF TWO FILES
PLOT REQUESTED?
NO- RETURN TO MAIN
YES‘ PACK A REPRESENTATIVE SET OF T'RANSMIT'I‘ANCE
VALUES INTO DATA ARRAYS FOR FUTURE PLOT'I‘ING
RETURN TO MAIN

H000 000

000

000

000

91

23
25

1914

SUBROUTINE DATA

COMMON PLTT.LPT.RAD.HDF.WAIT}DEST.ASHT.ATMT.DELT}DESP
COMMON ASHP.ATMP.SCAN.VER.HOR.VERI.HORI.VERL.HORL
COMMON NMSCAN.NMRUN.SAMP.TAGI

'COMMON FIRSTI.ISAMC.NSCAN.HPOS.VPOS.IRUN.ISAMP

COMMON AMNT.ISAMCA.ABSINT.XIOIDK.XIDK

COMMON FREQ

COMMON /B/DATAI(IOO).DATA2(IOO).AMIN.AMAX.IDIR

COMMON /CBLK7MODEI.SUBI.SUB2.FILEI.FILEZ.DARKT.PINIT.SINIT.SUB3
COMMON /SPBLK/BRNT.BRNP

LOGICAL PLTT.LPT.RAD.HDF.WAIT.SCAN.SAMP.FIRSTI

LOGICAL F.SUBI.SUB2.FILEI.FILE2

DIMENSION BACK(IOOO).XVAL(IOOO)

DATA lSKP/2/

WAIT FOR POSITIONER.TO REACH DESTINATION

XNULL=O.
WAITP=I.
CALL POS(WAITP.0..0.)

WAIT FOR DELAY TIME TO ELAPSE

CALL TIME(XT)
IF (XT.LT.DELT) GO TO 10

DEPOSIT SAMPLE IF REQUESTED

IF(.NOT.SAMP) GO TO 90
XSAMP=FLOAT(ISAMP)
AMWTTfiAMNT

CALL DAC(DESP)

CALL SAMPD(XSAMP.AMNTT)
CALL DAC(XNULL)

DESOLVATE BEFORE TAKING IOOXT

CALL CLOCK(500)

CALL CLTIME

CALL DAC(DESP)

CALL TIME(XT)
IF(XT.LT.DEST) 'GO TO I5
CALL CLTIME

CALL DAC(0.)

GET I(0)+I(DARK). AVERAGE 100 PTS AND RETURN

CALL POS(XNULL.HPOS.VPOS)
CALL POS(WAITP.HPOS.VPOS)
Frnso:noo.

CALL FREQS(FFREQ)
NCON=IOO

CALL SET(BACK.NCON.IERR)
CALL GO

SUM=O.

DO 91 I'I.IOO
SUM=SUM+BACK(I)
XIOIDK=SUM/IOO.
NPTS=IFIX(ATMT*FREQJ
START HEATING STEPS

F=.T.

DO 60 J=I.2

CALL CLTIME

CALL DAC(ASHP)

CALL TIME(XT)

40

49

00¢!
90

00‘
0|

0000000000000

000

100

195

IF (XT.LT.ASHT) GO TO 25
START ACQUIRI NG DATA

SETF=I.

CLEAR3-I.

FQ=FREQ

CALL FREQS(FQ)

IF(F) CALL SET(XVAL.NPTS.IERR)
IF( .NOT.F) CALL SET( BACK.NPTS. IERR)
IF(RAD) CALL FLAG(SETF.O..O.)
CALL DAC(ATMP)

CALL GO

F=.F.

CALL FLAG( CLEAR.CLEAR.0.)
XNULL=0. ,

WAIT FOR 10 SECOND DELAY
CALL DAC( XNULL)

IF (J.EQ.2) GO TO 65

CALL CLTIME

CALL FLAG( CLEAR. CLEAR.CLEAR)
CALL DAC( BRNP)

CALL TIME( XT)

IF(XT.LT.BRNT) GO TO 49

CALL DAC( XNULL)

CALL CLTIME

CALL TIME( XT)

IF (XT.LT.3.) GO TO 50
CONTINUE

START DELAY CLOCK

CALL CLTIME

START POSITIONER BACK TO ORIGIN
CALL POS(XNULL..OI.XNULL)

THIS PORTION OF THE ROUTINE WAS ORIGINALLY AN INDEPENDENT
CALCULATION SUBROUTINE. CONSULT EPCALC.F4 VERSION I
FOR ORGINAL VERSION OF THIS ROUTINE.

ALLOW INTERPOSISITION OF A USER SUBROUTINE FOR MASSAGING

' THE RAW DATA OR FOR ALTERNATE CALCULATION METHODS

IF (SUBI) CALL SUBONE( ISKP.XVAL.BACK)
IF (ISKP.EQ.O) GO TO 105

SET UP VARIABLES FOR ABSORBANCE CALCULATIONS

ABSINT=I.
AMIN= 10. **IO
AMAX=~AMIN
Q=XIOIDK-XIDK
T'MAX=O.

NOW CALCULATE THE INTEGRATED ABSORBANCE

DO 100 I'I.NPTS

T'DIF=Q/(XVAL( I)-BACK( I)+Q)

IF (AMAX.LT.TDIF) AMAX‘TDIF

IF (AMIN.GT.T'DIF) AMIN=TDIF

IF (AMAX.EQ.TDIF) TMAXBFLOAT(I)/FREQ
ABSINT= ABSINT‘RTDIF
ABSINT=ABS(ABSINT)
ABSINT=ALOGIO(ABSINT)

c-
105

I IO
C-

I20
125

I35

I40

I45
I46

I47
I48

I51

196

AMAX=ABS(AMAX)
AMAX=ALOGIO(AMAX)
AMIN=ABS(AMIN)
AMIN=ALOGIO(AMIN)

ALLOW USER SUBROUTINE FOR DATA TREATMENT AND/OR ANALYSIS

ISICP=2

IF( SUB2) CALL SUBTWO( ISKP.XVAL.BACK)
IF (ISICP.EQ.O)RETURN

IF (ISKP.EQ.I) GO TO 135

IF REQUESTED OUTPUT THIS DATA TO LINE PRINTER

IF(.NOT.LPT) GO TO 120
WRITE(3. IIO) ISAMP.AMNT.HPOS.VPOS.TMAX.AMAX.ABSINT
FORMAT( ’ ’ . I3.2X.F8.2. IX.F7.3.2X. F7.3. IX.F?.2.F8.2..‘2X.FIO.2)

IF REQUESTED WRITE DATA ON OUTPUT FILE

IF( .NOT.FILEI)GO TO 135
WRITE(5. I25)TAGI.TMAX.HPOS.VPOS.XSAMP.AMNT.AMAX.ABSINT
FORMAT(A2.7GI3.6,/)

IF PLOT REQUESTED SET UP DATA ARRAYS IN COMMON TO PASS DATA

IF (.NOT.FLT'I‘.AND.(SWIT‘CH(4.) .NE. 1.))GO TO 145
FINC=(FLOAT( NPTS) )/IOO.

P13 1 O

DATA1(I)=XVAL( I)

DATA2( I)=BACK( I)

DO I40 I=2. 100

PT=PT+F INC

IPFFLOAT( PT)

DATA“ I)=XVAL( IPT)

DATA2( I) =BACK( IPT)

ALLOW OUTPUTTING SHAPE TO FILE STRUCTURE) DEVICE

IF(.NOT.FILE2) RETURN

DO 146 I=I.NPTS

XVAL( I)8ALOGIO(ABS(Q/(XVAL( I)-BACK( I)+Q)))
DO I47 I=I. NPTS. 10

WRITE(6. I48) TAGI. (XVAL(K).K=NPTS. NPTS+IO)
FORMAT(A2. IOGI3. 6)

WRITE(6. I51)

FORMAT(‘ER')

RETURN

END

00000000000000000000000000000000000000000000000

197

FUNCTION: PLOTTER ROUTINE FOR GBA SAMPII RUN
PROGRAMMER: E. H. PALS

VERSION: *IB-I (2.02)
DATE: 9/20/76 ( I2/3I/76)
FILE: EPPLOT.F4 (EPPTV2.F4)

CALLING FORM: CALL POT

SUMMARY: THIS ROUTINE TAKES THE RAW DATA COLLECTED
IN DATA] AND DATA2. CALCULATES ABSORBANCES AND THEN
PLOTS THE RESULTS ON THE ADDS. IF THE WAIT OPTION
WAS REQUESTED. OPTIONS ARE AVAILABLE TO GET A
HARDCOPY PLOT ON THE LPT. OR A PLOT OF

THE TRANSMITTANCE OF THE BACKGROUND RUN.

EXTERNAL ROUTINES: THIS ROUTINE USES ALL THE

ROUTINES IN ERJ'S ADDS PLOT'I‘ING PACKAGE.

CHANGES VERSION *IB:

I . TAKE DATA OUT OF COMMON

CHANGES VERSION 2.01:

I. RESTRICT PLOTTING OPERATION TO THE 100 PTS IN
ARRAYS DATAI AND DATA2 .

METHOD :

I. CALL PLINT TO SET UP PLOTTING LIMITS
2. CALL AXIS TO DRAW AXIS
3. CALL LABEL TO LABEL AXIS TIME AND ABSORBANCE
4. CALCULATE INDIVIDUAL ABSORBANCES FOR DATA STORED
IN PASSED ARRAYS
5. CALL PLT TO PLOT EACH POINT AS IT IS CALCULATED
6. WAIT REQUESTED?
NO- RETURN TO MAIN
YES- A. PRINT HARDCOPY IF REQUESTED
B. IS PIOT OF BACKGROUND REQUESTED?
' NO- RETURN TO MAIN
YES- GO TO 7
7. FIND MAXIMUM AND MINIMUM TRANMIT'I‘ANCES
8. PASS MAX ANS MIN. DRAW AXIS AND LABEL
9. CALCULATE AND PLOT INDIVIDUAL TRANSMITTANCES
IO. WAIT FOR USER TO REQUEST CONTINUING

10
20

35
70
80

50

00

90
I00
I 10

COMMON NMSCAN . NMRUN . SAMP . SAMPSZ

198

SUBROUTINE POT
COMMON PLTT. LPT. RAD. HDF . WAIT. DIST. ASHT. ATMT. DELT. DESP
COMMON ASHP . ATMP . SCAN . VER. HOR. VERI . HORI . VERL. HORL

COMMON FIRST] . ISAMC.NSCAN.HPOS.VPOS. IRUN. ISAMP
COMMON AMNT. ISAMCA.ABSINT. X10 IDK. XIDK

COMMON FREQ

COMMON /B/DATAI( 100) .DATA2( 100) .AMIN.AMAX. IDIR
DIMENSION PLTDAT( 224) .XTXT( I) .YTXT(2) .YYT'XT(2)
LOGICAL PLT'I‘.LPT.RAD.HDF.WAIT.SCAN.SAMP.FIRSTI
DATA XTXTV’ TIME '/. YTXT/‘ABSORBANCE’I

DATA YYTXT/‘BKG %T'/

SET UP PLOT'I'ING LIMITS INTIALIZE

AATMT‘=ATMT

AAMIN=AMIN

AAMAX=AMAX

CALL PLINT( 0. .. AATMT. AAMIN. AAMAX. PLTDAT)

CREATE AXIS

CALL AXIS( 1.1.1)

LABEL AXIS

CALL LABEL( XTXT. 6 . YTXT. 10)

Q=XIOIDK-XIDK

NPTS= IF I X( ATM'NFREQ)

DO 10 I: I . 100

X= ATMT*FLOAT( I)/100.

Y=ALOGIO( ABS( Q/( DATA“ I)-DATA2( I)+Q) ) )

CALL PLT( - I . X. Y. PLT'DAT)
WRITE(O.20)AMIN.AMAX.ABSINT

FORMAT( ’ AMIN’ .GI3.6.’ AMAX’.G13.6. ' ABSINT’ .'GI3.6)

IF WAIT IS NOT REQUESTED RETURN IMMEDIATELY

IF (.NOT.WAIT) RETURN
WRITE(0.80)

FORMAT( ’ BACKGROUND ADDS PLOT? ’ .3)

READ(0.50) ARC
FORMAT(AI)
IF(ARG.EQ.’N’) GO TO 100
IF(ARG.NE.’Y’) GO TO 70

’ THIS PORTION OF THE ROUTINE PLOTS‘THE BKG

NOTE THAT 0 T=XIDK AND IOOFXIOIDK-XIDK
TRMAX=-10.**10

TRMIN=-T'RMAX

DO 85 I=I.NPTS

TT=DATA2( I)

IF (TRMAX.LT.TT) T'RMAX=TT

IF (TRMIN.GT.TT) TRMIN=TT
TRMAX=((TRMAX-XIOIDK)/Q)*100.
TRMIN=( ( TRMIN-XIOIDK) /Q)*100.
CALL PLINT(O. .ATMT.TRMIN.TRMAX.PLTDAT)
CALL AXIS( I. I. I)

CALL LABEL(XI'XT.6.YYTXT.6)

DO 90 I=I. 100

T‘=( (DATA2( I)-XIOIDIO /Q)8100.
X=ATMT¥FLOAT( I)/100.

CALL PLT(-1.x. T. PLTDAT)
WRITE(O. IIO)

FORMAT( ’ CONTINUE? TYPE Y' .0)
READ(0.50)ARG

IF(ARG.NE.’Y') GO TO 100
RETURN

END

3199

PROGRAM FUNCTION: 8 CODE FOR FAST ADC TO F4 CONVERSIONS
PROGRAMMER: E. H. PALS

VERSION: #16 (2.01)

DATE: 11/15/76 (12/7/76)

FILENAME: EPADBM.RA

EXTERNAL ROUTINES: THIS ROUTINE Is USED IN CONJUNCTION WITH
THE RALF PORTION OF THE PAST COVERSION PACKAGE (EPADRA.RA).

LOADING INSTRUCTIONS: MUST BE LOADED IN LEVEL 0.

ERROR CONDITIONS: IF THE REQUESTED CONVERSION FREQUENCY

IS TO FAST FOR DATA STORAGE BY THIS ROUTINE AN ERROR *HALT*
‘WITH THE ACCUMULATOR =I WILL RESULT. THIS

SHOULD NEVER OCCUR FOR A FREQUENCY REQUESTED OF IKHZ OR
LESS. HIGHER FREQUENCIES ARE POSSIBLE IF THE NUMBER OF
DATA POINTS REQUESTED IS SMALL.

METHOD: THIS ROUTINE IS THE COMBINED BMODE PIPORTIONS

OF THE INDIVIDUAL ROUTINES EPFREQ.RA AND EPADC8.RA. NOTES
CONCERNING EACH OF THESE PORGRAMS ARE LISTED TOGETHER
HERE FOR CONVENIENCE.

CHANGES VERSION *IB EPADC8.RA:

I. EAE INSTRUCTION SCA=7441 NOT 7411.

2. CORRECT DATA FIELD COUNTER DO NOT NEGATE POINTER.
CHANGES VERSION *IC:

1. REWRITE SUBROUTINE CALL TO CORRECT PROBLEM IN
CROSSING DATA FIELD BOUNDRIES.

CHANGES VERSION *IB EPFREQ.RA(8MODE CODE):

1. BECAUSE OF THE NEW MODE OF DATA STORAGE IT IS NOT NECESSARY
TO OPEN SYSTEM PAGE ZERO TO HOLD A CROSS PROGRAM

COMMUNICATION FLAG.

CHANGES VERSION 2.01:
I. THE BRAID POWER IS SET TO ZERO BEFORE EXITING.

NOTE: THIS PROGRAM IS WRITTEN FOR AN A/D CONVERTOR THAT

IS WIRED FOR A STRAIGHT BINARY OUTPUT. THE NUMBERS RETURNED

ARE POSITIVE FLOATING VARIABLES BETWEEN 0. AND 4096. THE
EXTENDED ARITHMETIC ELEMENT IS USED TO AID IN THE CONVERSION

TO A FORTRAN 4 VARIABLE COMPATIBLE FORM. BRIEFLY. THE AC-

AND MQ ARE CLEARED THEN THE A/D VALUE LOADED INTO THE AC

THE AC AND MQ ARE THEN SWAPPED AND NORMALIZED. THE STEP COUNTER
IS USED TO FORM THE F4 EXPONET AND THE AC IS USED

TO FORM THE FIRST WORD OF THE F4 VARIBLE MANTISSA.

////////////////////////////////////////////////
ADC8 IS ALL 8 MODE CODE
THIS PORTION PICKS UP VALUES SENDS TO F4/
////////////////////////////(///////////////////

DEFINE I/O INSTRUCTIONS AND EAE INSTRUCTIONS
SCA=7441 IEAE STEP COUNTER.OR?D INTO ACC

CAM=7621 /CLEAR AC AND MQ
SWP=7521 /SWAP THE AC AND MO

\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\

\\\

////
////
/

/
AGAIN.

CHECK.

NMI=7411

SKPAD=6532
READB=6534
CLRFG=6524

FIELD] ADCB
0

IOF
CAM
TAD AR
AND N7
CLL RTL
RAL
TAD N620l
DCA .+1
0

AD .-I
DCA CCDF
JIZP CHECK

200

/FORHALIZE TO F4 LIKE VARIABLE
/SKIP ON A/D FLAG

/READ A/D BUFFER

/CLEAR A/D FLAG

/SAVE RETURN

/DON‘T BOTHER ME I'M BUSY
/CLEAR AC AND HQ

/PICK UP DATA FIELD

/MASK

/MOVE TO

/PROPER BITS

/SKELETON OF CDF INSTR
/SET TO EXECUTE AS

/IN LINE CODE

/SAVE FOR FUTURE REFERENCE
/ELSEWHERE

/DON'T CHECK IF TO SLOW ON ENTERING

THE FOLLONING CODE IS RE-EXECUTED UNTIL ALL DESIRED VALUES
HAVE BEEN OBTAINED.

SKPAD

JNP CHECK
CLA

6111

IAC

BLT
SKPAD

JIIP .‘I
CAN
READB
CLRFG

SW?

NMI

DCA TEMP
SCA

CIA

TAD N27
DCAZ ARfI
ISZ ARfI
SKP

JHS SUB
TAD TEMP
DCAZ AR+I
ISZ ARfI
SK?

JMS SUB
DCAZ AR+I
ISZ AR+I
SK?

JHS SUB
ISZ IIIIUI'I
J”? AGAIN
ION

CDF CIF 0
JMPZ ADCB

/GOINC TO SLOW?

/NO. CONTINUE

/CLEAR AC FOR POWER LEVEL =0
ICLEAR BRAID POWER SUPPLY LEVEL TO OFF
/YES. LOAD I INTO ACC
/***ERROR HALT WITH AC=I***
/GOT A VALUE?

/NO. WAIT

/CLEAR AC AND HQ

/YES. READ IT

/AND CLEAR THE FLAG

/SWAP AC AND HQ SO EAE WONT THINK.NEG
/USE EAE TO NORMALIZE

/STORE TEMPORARILY

/PICK UP NUM OF STEPS

/NEGATE

/ADD 27(OCTAL)

/DEPOSIT EXPONET

/INC POINTER. END DF?

/NO. CONTINUE

/YES. GO CHANGE THE DATA FIELD
/GET MANTISSA

IDEPOSIT IN F4 MANTISSA WORDI

/DEPOSIT 0 IN 2 WORD F4 MANTISSA

/DONE YET?

/N0. GET ANOTHER POINT
/ALL RIGHT I'M DONE
/YES. GET SET TO RETURN

THIS CODE CHANCES THE DATA FIELD

201

 

/
SUB. 0 /SAVE RETURN LOCATION
CLA CLL
TAD N10 /TO INCREHENT CHANGE DATA FIELD INSTRUCTION
TAD CCDF /PICK UP LAST CNIYGE DATA FIELD INSTRUCTION
DCA .+l /SET TO EXECUTE AS IN LINE CODE
CCDF, 0 /DO IT
CLA CLL
JHPN SUB /RETURN POINTING TO NEXT DATA FIELD
ENTRY AR
AR, 0000 /NILL CONTAIN DF IN BITS 9-11
0030 /WILL CONTAIN THE ADDRESS
N7. 030? /FOR HASKING DATA FIELD
N10. 0010 /T0 INCREHENT DATA FIELD
N620]. 6201 /SKELETON OF CDF INSTR.
TEMP, 0000 /TEHPORARY MANTISSA STORAGE
N27. 0027 /NEEDED TO CORRECT EXPONET
ENTRY MHNUHB ‘
MHNUHB, 0000 /ROOH FOR PASSING NUHBER OF CONVERSIONS
0030 .
IEHEL COO /8HODE VARIABLE IS HERE
/ .
/////////////////////////////////////////////////////////
/ 8 MODE CODE FOR SETTING ADC HARDWARE CLOCK /
/////////////////////////////////////////////////////////
/
/ BIT PATTERNS FOR THE RATE ENABLE WORD
/ RATE=(M)*(10EY)
/ AC6 MUST BE SET TO ENABLE CLOCK CHIP
/
/ AC? 0 0 l I
/ AC8 0 l O I
/ .........................
/ H: 0 I 2 5
/
/ AC9 I 0 O O O I l I
/ ACIO O I I 0 O I 1 0
/ ACII 0 I 0 l O I O I
/ ................... -
/ = 4 3 2 I 0 -I -2 -3
/ I/O DEFINITIONS
RATE=652I /LATCH RATE WORD AND START IF AC6=0
CLRFG=6524 /CLEAR.A/D FLAG
ENTRY #FREQ
#FREQ. 0 /SAVE FOR RETURN
IOF /CRITICAL CODE HERE
CLRFG /MAKE SURE FLAG INITIALLY CLEAR
CLA CLL
TAD “8+2 /PICK UP M
RTL
RAL /PUT IN PROPER POSITION
TAD E8 /PICK UP CLOCK ENABLE BIT
TAD EXP8+2 /PICK UP EXPONET
RATE /LATCH OUT AND START CLOCK
CLA CLL /SET AC TO 7777
ION /ENABLE INTERRUPT SYSTEM
CDF CIF O /RETURN
JNPZ #FREQ
/DEFINE 8 MODE CONSTANTS
ENTRY EXPB
EXPB, 0:0;O; /EXPONENT HERE
ENTRY H8
H8, 030:0 IMANTISSA HERE

E8. 0040 /CLOCK ENABLE BIT

\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\

202

PROGRAM FUNCTION: THIS mum WITH THE BIDDE MDULE EPADDMJIA
CONTROL THE GBA A/D CONVERTOR.

PROGRAMMER: E. H. PALS

VERSION: 31A
DATE: 12/7/76
FILENAME: EPADCR.RA

LOADING INSTRUCTIONS: MAY BE LOADED IN ANY LEVEL.
EXTERNAL ROUTINES: NONE OTHER THAN F4 LIBRARY
ROUTINES AND THOSE IN THE BMODE PACKAGE (EPADBM.RA).

METHOD! THIS MODULE IS A PACKAGE OF THREE SUBROUTINE-
CALLING FORIB: CALL. FREQS(FFREQ)
CALL SET( ARRAY. HUM. IERR)
CALL GO
VARIABLE DEFINITIONS: FREQ-CONVERS ION FREQUENCY
ARRAY- NAME OF DATA STORAGE ARRAY
NUM- NUMBER OF CONVERS IONS REQUESTED
IERR- RETURN VALUE )0 INDICATES ERROR
FOR FURTHER EXPLANATIONS CHECK INDIVIDUAL
PROGRAM LISTINGS THAT FOLLOW.
NOTE: SUBROUTINES GO AND FREQ WERE ENTIRELY COMPOSED
BY THE AUTHOR. SET IS AN EDITED VERSION OF THE RALF LISTING
FILE PRODUCED BY COI‘IPILING A FORTRAN 4 SOURCE FILE. THIS
SOURCE FILE MAY BE FOUND ELSEWHERE IN THE GBA DOCUMENTATION.

NOTE: THE BMODE CODE OF THIS PACKAGE COULD HANDLE

UP TO 4096 REQUESTED CONVERSIONS AND SET COULD EASILY BE

MODIFIED T0 ACCOMDATE THIS HOWEVER THE LIMIT HAS BEEN ARBITRARILY
SET AT 2000 CONVERSIONS MAXIMUM.

ERROR CONDITIONS: IF HUM IS (0 IERR IS IETURNED AS 1. IF M2000
IERRISRETURNASZ. IERRISRETURNEDASOWHENTHEREARE

_NO ERRORS. AN ERROR *HAL'DR OCCURS WITH I IN THE ACCUMULATOR

IF THE REQUESTED CONVERSIONS FREQUENCY IS TO FAST FOR PROGRAM
EXECUTION.

PROGRAM FUNCTION: START GBA ADC AT REQUESTED
FREQUENCY

PROGRAMMER: E.H.PALS

VERSION: *IB
DATE: 10/5/76
FILENAME: EPFREQ.RA

CALLING FROM: CALL FREQS(FFREQ)
WHERE - FFREQ IS FLOATING VALUE OF
FREQUENCY CONVERSION mum.

NOTE: THE FREQUENCY OF CONVERSION REQUESTED IS ROUNDED DOWN
'IO MULTIPLES OF 1.. 2.. AND 5.

EXTERNAL ROUTINES: NONE. BUT THIS ROUTINE
IS COMI'IONLY USED WITH SET AND GO ’IO PICK UP VALUFS

\

FROM THE ADC.

EXTERN EXPO / IN EXTERNAL BMODE MODULE
EXTERN MB ITHIS ONE ALSO

EXTERN J‘FREQ NAME OF BMODE ENTRY POINT

SECT FREQS
J A IFRQST . ‘

BPFREQ. F 0.

XRFREQ.

FRQTMP,
EXP,

FRORTN.

*FRQST.

\\\

MDET.

FIVE.

OUT.

LARGE.

SMALL.

Q'fl'fl'fl"!

JA
BASE
STARTD
FLDA
FSTA
FLDA
SETX
SETB
BASE
FSTA
LDX
FLDAX
FSTA
STARTF

1083
FRQRTN.0
0
XRFREQ
BPFREQ
BPFREQ
BPFREQ
191
BPFREQ,1
FRQTMP

203

‘RALF CODE CALCULATE VALUES FOR.SETTING RATE

FCLA
FSTA
FLDAX
FSTA
JEQ
FSUB
JLT
FSUB
JGE
FADD
FSUB
JGE
FLDA
JA
FSUB
JGE
FLDA
JA
FLDA
ALN
FSTA
FLDA
ALN
FSTA
FCLA
TRAP4
FCLA
JA
FLDA
FADDM
FLDA
FDIV
FSUB
JLT
FADD
FSTA
JA
FADD
JA
FLDA
FSUB
FSTA
FLDA

EXP
FRQTMP
TEMP
OUT
F01
SMALL
F09
LARGE
F010
F02
TWO
FCI ‘
OUT
F03
FIVE
F02
OUT
F03

0

MB
EXP

0
EXPO

IFREQ

FRQRTN
FCI
EXP
TEMP
F010
F010
LDET
F010
TEMP
LARGE
F010
MDET
EXP
F01
EXP
TEMP

ISET EXP 3 0

/LOAD FREQ

/SAVE IT

/IF ARG=0 DISABLE ADC
/LESS THAN 1 NZ?

IYES. FIND NEG. EXP
/NO, GREATER THAN 1H2?
/YES. FIND POS. EXP
lRESTORE VALUE

/LESS THAN 2?

/NO . CONTINUE CHECKING
lYES. SET M?2

/LESS THAN FIVE?

/NO. GO SET TO'5

/YES. SET MF2
/BECOMES Mi5 IN 8 CODE

/PASS TO 8 CODE

/PASS TO 8 CODE

/* TO DETERMINE EXP>0

IRESET. AND GO DETERMINE M
/* TO DETERMINE EXP<0

\\\\\\\\\\\\\\\\\\

cxn,
'RET.

IBASE.
ARRAY.

1m,

IGOBAKQ
IARGS.

'TMP.
’LIT,

FMUL
FSUB
JCT
FADD
FSTA
JA

FADD
JA

F I.
F 2.
F 3.
F 9.

F 10.
PROGRAMMER: E.

PROGRAM'FUNCTION:

F010
F01
SDET
F01

SMALL
F01
MDET

DATA ACQUSITTON.

DATE:

11/30/76

2014

/RESET, AND GO DETERMINE M

H. PALS

SET UP ROUTINE.FOR.FAST

FILENAME: EPSET.RA

VERSION:
METHOD:

MODULE T0 EXECUTE PROPERLY.

81A

THIS PROGRAM IS A MODIFIED VERSION OF THE
RALF LISTING FILE OUTPUT OF FORTRAN IV COMPILER.
THE ORGINAL FORTRAN FILE IS SAVED UNDER.THE READING
OF EPSET.F4.

NOTE: THIS ROUTINE NEEDS EPAD08.RA WHICH IS AN BMDDE

IT IS IN THIS ROUTINE

THAT THE EXTERNALS REFERENCED IN THIS PROGRAM
ARE DEFINED.

'EXTERN AR
'EXTERN MHNUMB
EXTERR FLOAT
SECT SET
JA #ST
ORG .+10
“TEXT +SET +
SETX IXR
SETB IBASE
JA 0+8
ORG .+6
ORG .+3
ORG .+3
ORG .+3
ORG IBASE+30
FNOP

JA IRET
FNOP

0:0

ORG .+3
ORG .+0003
ORG .+0011
0000

0000

0000

0001

2000

0000

0002

2000

0000

0002

3000

ITELL LOADER VARIABLE IS NOT IN THIS MDULE
/SAME FOR.THIS VARIABLE

cm.
:ST.

#00001.

#00002.

#00003.

'LHL
IBASE
'GOBAK

'GOBAK30

IXR
vBASE
0,1
IBASE
OARGS
rBASE.1+
AR
ARRAY
IBASE.1+
NUH
IBASE.1+
IERR

0002.0
0003.0
ILIT*0000
IERR
0004.0
NUM
'LITW0017
'00001
ILIT+0003
IERR
0005.0
NUM
'LITW0000
000002
lLIT+0006
IERR
0006.0
IERR
lLIT*0000
#00003
#NE

ORTN
0007.0
NUM

300004

FLOAT
.+0004

0

XNUM
0

MMNUMB
0010.0
'NE
ORTN
0012.0

205

/PASS THIS.ADDRESS TO BCODE

xnmxz IT'NEGATIVE
ICHANGE.TO BMODE COMPATABILTTY
IPASS VALUE‘TO OCODE

\\\

ISTGO.

EXTERN 'NE
J A 'RTN

THIS SECTION
EXTERN ADCB

SECT GO

J A 'STGO
0

BASE 0
STARTD

FLDA 30
FSTA IGORN
FCLA

STARTF

TRAP4 ADCB
FCLA

J A IGORN

206

INITIATES THE DATA ACQUSITION PROCESS

LOADER
. SYMBOL

ABS
ADCRTC
ADCB
ADPLOT
AINT
ALOG
ALOGIO
ANYCHR

ARGERR
AXIS
AXISS

BACKB

BURNDM
CBLK
CGET
CHAR

CLOCK
CLTIME
CPUT
DAC
DARK
DATA
DATE
EXIT
EXPB
FLAG
FLOAT
FLOW
FREQS
FUNC

GRAPH
GRAPHS
HEAD
IABS
IDEB
IFIX
INPUT
INT
IPLOT
ISIGN
ISIM
LABEL
LABELS
LIMIT
LOOKUP
MMNUMB

ONQB
ONQI
OUT
PARIN
PLINT
PLINTS
PLT
PLTS
POS
POSDEM

V23 09 /14 I77
VALUE LVL OVLY

13417
17372
11267
10600
32000
17242
17515
16133
16714
10670
00204
25445
32745
13615
11713
10101
32400
14756
20165
12600
20066
15654
16066
20175
10200
32400
20400
17066
00223
10720
11000
15353
20400
15143
24366
15634
27175
35400
32400
17372
15011
17242
20400
17242
17273
20027
16460
26336
32400
32400
17027
10677
10723
11510
11400
12400
32400
23117
32400
23464
32400
10400
33104

enou—a—oooooeeae—csea—coconut—o—o—eaooauaooooooocoeooa—ooooeoueooo

00
00
00
00
00
00
00
00
00
00
00
03
11
00
00
00
03
00
00
00
00
00
00
00
00
02
02
00
00
00
00
00
01
00
04
00
03
01
10
00
00
00
00
00
00
00
00
03
12
00
00
00
00
00
00
00
06
03
11
03
13
00
03

207

POSITN
POS IT‘S
POT
PT
PTS
RESPND
ROTATE
RUN
SAMPD
SAP
SAPDEM
SAPINT
SET
SFL2
SIGN
SMXNM
SMXOPT
SMXSUB
SPBLK
SUBONE
SUBTHR
SUBTWO
SWITCH
TIME
TUNE
TUNES
MAIN
36000

00
00
01
02
03
04
00
00
01
02
03
04
05
06
07
10
11
12
13
00
00
01

an»uowaaaaoaaaan---~o

27402
35400
20400
24501
34400
32400
16765
32400
16370
32714
32400
11600
15372
32400
17754
22136
20400
20400

16672
13000
13000
13000
10000
16105

33121

11671
13000

' 18T FREE LOCATION
LVL OVLY LENGTH

20352
10416
03342
07660
07351
11116
00370
01260
00275
01162
01224
00737
01102
01616
01163
01651
01236
00637
01015
00744
00347
00205

sconce-Queen-oaooaaoaoam—u—a-

00

Ex:
EX:
Ex:

208

APPENDIX B

Simple Hardware Control Approach for

Sequencing Chemical Instruments

Reprinted with permission from Analytical Chemistrx, Vol.
50, Page 291A, February 1978, Copyright 1978 by the American
Chemical Society.

209

S. R. Crouch,‘ D. N. Baxter,”
E. H. Pals, and E. R. Johnson3

Department at Chemistry
Michigan State University
East Lansing, Mich, 48824

210

Simple Hardware Control Approach
for Sequencing Chemical Instruments

In modern analytical instrumenta-
tion. digital controllers have become
more and more complex and flexible
as instruments have become more
highly automated. The controller di-
rects the sequencing of various instru-
ment functions. For example. in an
automated gas chromatography sys-
tem. the controller might be required
to send signals to an automatic sam-
pler for introduction of the sample
into the column. to control the oven
temperature and its temperature pro-
gram. to control the sensitivity of the
detection system. and to control the
detection and integration of the vari-
ous chromatographic hands as they
elute from the column. Such control-
lers may be simple fixed order sequen-
cers or may require branching to skip
certain functions if certain instrument
conditions are set by the user li.e.. un-
less the temperature program button
and rate have been set. the Gf‘ con-
troller would skip this part of its se-
quencel.

In modern instruments. controllers
take the form of hardwired logic sys-
tems. minicomputers or microproces-
sors. The microprm‘essor is certain to
become the controller of choice in the
future for many operations. because
the program (sequence) can be altered
without changes in hardware. Also.
even the most complex control opera-
tions. with multiple and nested
branches. can be readily carried out
under the supervision of a micropro-
cessor. However, the microprocessor
requires the instrument designer to
be re-educated in terms of software
and processor capabilities. This edu-
cation process may require months to

 

' To whom correspondence should be ad-

dressed.

? Present address. Eastman Kodak (‘o..
Kodak Park. Building 82. Rochester. NY.
14650.

" Present address. Department of (‘hem-
istry. \\ ayne Statel niyersity. Detroit.
Mich. JR'JHL’.

years in order to get working systems
into the laboratory with "intelligent“
instrument controllers.

A few years ago our research group
became heavily involved in controllers
when we decided to undertake a
project that involved the spatial pro-
filing of atomic concentration above
a filamentotype electrothermal atom-
izer using atomic absorption spec-
trometry. We desired to automate
fully the operations of sample intro-
duction to the filament. movement of
the filament in both x and y directions
away from the AA optical axis to pre-
determined destinations. temperature
programming of the filament. data ac-
quisition of backgroundworrected
peak area. and automatic return to po-
sition "zero" for pickup of a new sam-
ple. The filament position controller
was intended to move the atomization
cell freely over a total displacement
of] in. in eitheriif'twodimensions.
through the use of stepper motordriv-
en micrometer translation stages. The
controller was required to function in
either a “local" mode. in which the de-
sired filament position could be set by
the user via front panel BCl) switches.
or in an "on line" mode in which the
controller received binary information
from a minicomputer and functioned
as a hardware slave to the processor.
In addition. we required the controller
to convert BCD position code to bina-
ry Code in the local mode. to use posi-
tion instead of displacement as the
input tie. to remember where it is
and to determine the direction to
move and how far to gol, and to be ca-
pable ofa hardware reset to an opti-
cally defined reference position. Final-
ly. we required the controller to cor-
rect for backlash in the leadscrews of
the translation stages in that if a re-
quested movement was in the positive
direction (away from the optical refer-
ence). the controller would intention-
ally move the cell to a known distance
beyond the desired destination and
then return to the true destination. in

Contrast to movement in the negative
direction. which would proceed direct~
ly to the destination without such ov-
ershoot. In this manner. no matter
whether the net movement was posi-
tive or negative. the cell would be
moved last in the negative direction.
thus leaving the leadscrews tightly
meshed in that direction of move-
ment.

Because of the complexity of our
application. we did not look forward
to relearning combinatorial and se-
quential logic methods. component
minimization techniques. and other
forgotten digital design methods. Mi-
croprocessor approaches were consid-
ered but later abandoned because the
interleaving of real-time tasks each
with their own timing requirements
put the problem on a level of complex-
ity where our then meager micropro-
cessor experience could not hope to
reach in a reasonable time. Fortunate-
ly. the sinking feeling we had original-
ly experienced was short-lived when
we discovered the simple approach to
the design of complicated controllers
described in this paper. Even for our
complex application. this systems-
level approach to controller design en-
abled us to design in about 3 h a unit
requiring over 50 10's per motor con-
troller. Construction and debugging
of the position controller required ad-
ditional time. of course. The speed of
our design was made possible by the
use of a method. originally described
by Richards (1 ). that allows all the
complex sequential operations to be
treated by using only a simple flow
chart description at the beginning and
by implementing the flow chart with
3-4 MSl lC's.

To understand the approach. we
present in the first section of this
paper the general concepts and princi-
ples of the “state" of a controller (2)
and the simple implementation that
results from the Richards (Hap-
proach. We then present a chemical
example of a "fall-through" sequencer

 

 

 

211

 

'III’OII
YOAISIII
FDIC Y I OI

IO

.L_.

”It"! I

 

Pl” one
thauann

'Dltilo. '—'..1AYI he;
7'.

 

 

 

 

'IIFOII
YOAISVII

rue'c‘r‘ro. L—«Deti\rl not

 

 

 

Figure 1. Basic flow chart units for transfer conditions and transfer functions. N:

state number

that controls an automatic fraction
collector used in elution chromatogra-
phy. 1n the last section we describe a
branching sequencer and illustrate its
use as an electrothermal atomizer po-
sition controller.

Concept ot the State

Automatic sequencing of measure-
ment and control operations is highly
desirable for modern chemical instru-
ments. We assume here that the oper-
ations to be sequenced are ON-OFF
operations that can be controlled by
logic level signals. The sequencer must
then produce logic level outputs for
each controlled operation and must
be responsible for the timing relation-
ships of the appropriate logic level sig-
nals. There are a variety of approaches
to the design of sequencers (3). How-
ever. the systematic approach present-

WI

 

 

 

 

ed here allows even the most complex
sequencing applications to be handled

with ease.

To help understand this approach,

several terms must be defined. This
can be done most conveniently with
the aid of an example. Consider the
controller for the thermostat on a
temperature bath. The method de~
scribed by Richards (1) considers it

to have three states: State 0 - thermo-

stat off-heater off ; State 1 . heater

on: and State 2 = heater off. in State
0 the thermostat controller waits for
the condition of the thermostat power

switch to be on. The on condition of
the switch then is referred to as a

transfer condition. When the transfer

condition for State 0 is true. i.e.. the

switch is on. the controller advances
to State 1. The process of this advance

causes a transfer function to occur.

I "OI Illt I
II ISO.

 

The heater is turned on. The second
state (State 1) tests to see if the bath
temperature exceeds the limit set on
the thermostat. When the bath tem-
perature does exceed the limit. the
controller steps to State 2. which
causes another transfer function to
occur. The heater is turned off. The
transfer condition for State 2 is a bath
temperature below the thermostat
limit. When this condition is met. no
transfer function is required except
to return to State 0 and repeat the
cycle.

Viewing such a trivial example in
terms of states, transfer conditions.
and transfer functions may seem over-
ly complicated. but it is useful for two
reasons. First. the method is also ap-
plicable to much more complex con-
trol applications. Second. with this ap-
proach the circuit design is greatly
simplified by use of a standard circuit
for the heart of the controller.

The application of this approach in-
volves first converting the states.
transfer conditions. and transfer func-
tions into a standard flow chart for-
mat. The two basic flow chart units
for each state are shown in Figure 1.
The unit used depends on what is to
occur at that particular state. Either
unit contains a diamond-shaped box
that represents the state and contains
the transfer condition. Depending on
the application. the basic unit will also
contain one or two rectangles that rep-
resent the transfer functions which are
to occur when the transfer conditions
are met or not met. in this method the
transfer function on the “no" leg (if
one exists) is available only if the
transfer condition not being met
causes a move to a State I other than
the present one. This transfer func-
tion, if it is used. is called a secondary
transfer function. as opposed to the
primary transfer function on the
“yes" leg. Figure 2 shows the flow
chart for the bath thermostat exam-
ple. No secondary transfer functions
are required for this application;
therefore. the controller is an example

 

 

 

 

 

 

 

 

‘reecrtea e

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 2. Sequencer flow chart for bath thermostat example

 

 

212

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

OOIA .
I .I7 I ' I -1’
"Put. ‘35:”. ggcg‘gg O—OOICOIDAIV
I use-ant ’ "”1""
' 'UUCIIOI‘
C
CLOCI—I SIAILI I'm ..3 ._.p
to ureut (inlet ""’
on 0' ALL __I
JUI' FUNCTIONS
I . .‘LICI HCLIC‘ . '
—'. I 00"”? IIQDLC I —.
II'UI —- 2 I —. 'RIIAIY
'Ilaulfth ' a g" g g" ' IIAIOFII
COIOIIIDU. e IULTI'L‘I‘I O‘CODI. e '”.c"°..
a '“I'I.' O
-III'I I'I F—H’
—- ll et—D

 

 

 

 

 

 

Figure 3. General schematic diagram at controller

of a fall-through sequencer (no condi-
tional branches).

The “heart" of a sequencer based
on this approach can be implemented
with the standard integrated circuits
shown in Figure 3. From the flow
chart. the number of states in the se-
quence and the various transfer condi-
tions and functions are known. If no
secondary transfer functions are nec-
essary. the secondary n-bit decoder
can be eliminated. Thus. the heart of
a sequencer consists of a k-bit coun-
ter. an n-bit multiplexer. and one or
two n -bit decoders. where n is the
number of states in the sequence and
n S 2*.

The fit -bit counter keeps track of the
state of the controller. Logic level sig-
nals representing the. various transfer
conditions are connected to the n -bit
multiplexer. The outputs of the state
counter determine which transfer con-
dition is supplied by the multiplexer
to one or two n -bit decoders. The mul-
tiplexer output also allows the state

 

 

 

 

counter to increment to the next state
on the next clock pulse after a “true“
transfer condition. The presence of a
“true" transfer condition at the multi-
plexer output causes the appropriate
primary transfer function (as selected
by the state counter) at the output of
the decoder to go true. However. only
a pulse occurs on the selected decoder
output line. because the next clock
pulse increments the state counter to
the next state. If the transfer condi-
tion is false. either a secondary trans-
fer function can be generated (at the
secondary decoder output ). or the se-
quencer merely waits in the same state
until the transfer condition becomes
true. To provide the capability for the
sequencer to skip certain states under
the appropriate conditions. a state
counter capable of parallel loading is
used. When a jump or branch in the
sequence is necessary. the load (LD)
line of the state counter is activated.
and the new state is loaded into the
counter through the data inputs.

 

An eight-state sequencer can be
readily implemented with a 74163
counter. a 74151 multiplexer. and one
or two 7442 decoders. At current sin-
gle unit prices. they can be obtained
for just over $3.00.

Fall-Through Sequencer

The fall-through sequencer is the
simplest type. In this sequencer the
only decision to be made is whether
to remain in the present state or pro-
ceed to the next state. ln a closed-loop
operation this decision is based on re-
ceiving an “all done" signal (flag) from
a previously actuated circuit. In an
open-loop operation the change of
state may be immediate or may occur
upon receiving a signal that a time
delay has elapsed.

A chemical example of a fall-
through sequencer is a controller for
an automatic fraction collector used
in elution chromatography. Typically.
the fraction collector has a rotatable
tray that contains vials sequentially
rotated beneath the column outlet
valve. When a vial is directly beneath
the outlet valve. the valve is opened
for a preset time. then closed. and the
next vial rotated into the receiving po-
sition. These motions are repeated
until the desired number of fractions
has been collected.

In Figure 4 the decisions and ac-
tions involved in controlling this ap-
paratus are represented in flow di-
agram form. along with the external
inputs used to make the decisions. As-
sume that the operator has initialized
the apparatus. by rotating the first
vial into position beneath the valve.
and loaded the number of fractions
desired into a counting register. Also,
assume that the controller is initially
in State 0. Condition A is satisfied
since the counter is nonzero; therefore.
the controller generates Function A.
which is to open the delivery valve and
to start the delay timer. Now the con-
troller moves to State 1 and repeated-
ly queries the delay timer until it sig-
nals that enough solution has been de-

 

I
CI

 

 

 

 

 

 

'UICIIOI A 'WCII" O
3.73.“; 5:33:
._I
SIAII I'll!
DILM' 1”?
VIII. IDVII.

 

 

 

 

 

 

 

 

 

 

 

 

 

   

'UICI I 0' C
NCQIIII‘I
VIAC‘IIOI
CWIIII
10'

‘ IA! ‘

 

 

 

Figure 4. Sequencer tlow chart for automatic traction collector

CORD] 1 I 0"

CAI'IACTIOI COUITII I. IOIIEIO

CUHIII IILAY I. 00"
CCIIICIOSIIVCN I'IUCI

rt
TIT?

213

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

cs
A egg-1" . 'UICIIOIS
.. I u ' r'troese nus. sun H
s sin stictess nus. novs sit
a t. trustee um. ester-cur
m—cu
cu e>—r1’
t" .A 0e 'c u
uses: ' II
a ID—F'E
e s:
: ‘9
sh
cs . r a II c .
ce——li e s 1
cc “l: nutrinsxse
s
s
s s
1

 

 

 

Flame 5. Schematic diagam of traction collector controller

livered (Condition B). When Condi-
tion 8 has been satisfied, the control-
ler closes the valve and starts the vial
tray moving (Function B). Now the
controller moves to State 2 and waits
for a position indicator microswitch
beneath the outlet valve to be tripped
(Condition C). When Condition C is
met. the controller must decrement
the fraction counter register, stop the
tray. and effect a jump back to con-
troller State 0 (Function C). This se-
quence is then repeated until the frac-
tion counter register is decremented
to zero. When this happens. Condition
A can never be met. The apparatus
halts since it becomes hung up in
State 0.

This controller can be quite easily
implemented with three IC's: a 74163
counter. a 74151 multiplexer. and a
7M2 decoder. Of course. additional
packages may be necessary for such
things as delay timing. but these three
IC's handle the generation of all the
signals necessary to initiate the ac-
tions at the appropriate times. Figure
5 shows how these lC‘s are intercon-
nected and also the input and output
signals required in this particular ap-
plication. When power is turned on,
an initialize pulse (INIT) is required
to clear the fraction counter register
and to assert the CLEAR (CLR) input
of the state counter. This forces the
state counter to State 0. The Q. Q3.
and Qc outputs of the state counter
are all LO (zero). which sets the multi-
plexer’s Y output L0. This LO signal
is fed to the state counter's ENABLE
P (EN P) input. which inhibits the
counter from counting clock pulses at
its CLOCK ( C LK) input. Thus. it re-
mains in State 0. The multiplexer's W
output (the complement of the Y out-

put) is directed to the decoder’s D
input. Output W is HI. and examina-
tion of the truth table for the decoder
indicates that the 0-7 decoder outputs
must be H] when D is HI (LO true
logic is used at these outputs so that

a function is generated only when the
output goes L0). This means that ini-
tially none of the functions is being as-
serted. When the operator gives the
“GO" command. the number of frac-
tions desired is loaded into the frac-
tion counter register. and Condition

A is satisfied. Now the multiplexer's
W output goes L0. and the decoder
points to a decimal 0. whi_ch asserts
the Function A signal (FA).

Function A triggers external circuit-
ry to open the delivery valve and start
a time delay circuit. In addition. the
multiplexer Y output is H1. so that the
state counter's ENABLE P input is
asserted. and on the first clock pulse
the counter is incremented once (State
1). Now the multiplexer shifts from
selecting data channel 0 (Condition
A) to data channel 1 (Condition B).

If the required time delay has not yet
elapsed, Condition B is not yet satis-
fied so that the multiplexer Y output
falls. locking the state counter in State
1. and the W output goes HI. concludo
ing function A and inhibiting the de-
coder from generating any functions.

When the required time delay has
elapsed. Condition B is satisfied. and
a similar sequence occurs in the state
counter. multiplexer. and decoder.
The multiplexer's W output enables
the decoder. and its A. B. C, and D in-
puts point to decimal 1. Thus. output
1 (Function B) is asserted. This
triggers the valve to close and starts
the tray moving. The multiplexer Y
output goes H1. allowing the counter

to count one pulse and move to State
2. Now the state counter requests the
multiplexer to address the status of
Condition C.

When the new sample vial reaches
the position beneath the outlet valve.
a microswitch is tripped and Condi-
tion C is satisfied. Again this activates
the decoder. and Function C is assert-
ed. Function C stops the tray and dec-
rements the fraction counter register.
Also. since this is the last function in
the sequence. it is used to assert the
CLEAR input of the state counter and
return the controller to State 0.

Unless this was the last fraction to
be collected. the fraction counter reg-
ister will be nonzero when the state
counter enters State 0 so that Condi-
tion A is immediately satisfied. Thus.
as soon as the Function C pulse has
cleared the state counter to State 0.
Function A will be generated. and on
the next clock pulse State 1 will be
achieved. This entire delivery se-
quence will continue until the fraction
counter register reaches zero. which
prevents Condition A from being sat-
isfied. Then the apparatus will halt
because the controller becomes hung
up in State 0 until the operator inter-
venes to start a new collection se-
quence.

Branching Sequencer

In many cases. it is desirable to have
a sequencer capable of branching; that
is. one in which sections of the basic
state sequence may be skipped. re-
peated. or otherwise altered as re-
quested by the instrumentation under
control. An example of such a sequen-
cer is the electrothermal atomizer po-
sition controller mentioned previous-
y.

The flow diagram of its sequence
is shown in Figure 6. The system waits
in State 0 until data representing a
new location are received. and a com-
mand to move is given. Upon receipt
of this command. the overflow flag is
cleared. and the system determines
if it is being operated under local or
computer control. If control is local.
the desired destination has been en-
tered in BC D format from front panel
thumb-wheel switches. which requires
a BCD to binary conversion. A check
is made to see if the system is free of
overflow; if so. another check is made
to see if BCD to binary conversion is
complete. If conversion is not com-
plete. the registers are clocked. and
the checks for overflow and complete
conversion are repeated. If an over-
flow should occur. the sequence will
abort to State 0 with the overflow flag
set. (Overflow results from entering
a destination outside the defined lim-
its of the system.) If no overflow oc-

curs. then upon completion of the con-

version. the stepper motor direction
is determined, and the stage begins to
move. If control is from the computer.
the BCD to binary conversion is un-
necessary. Thus. this section of the se-
quence is omitted.

The stage moves until the destina-

   

21“

tion is reached. At this point the sys-
tem checks to see if the direction of
movement was positive. If so. the
stage continues to move 10 additional
increments positive. and when fin-
ished. reverses direction and comes
back 10 increments negative. lf move-
ment was originally negative. the se-
quence is concluded immediately. In
this manner. gear play is removed
from the mechanism by assuring that
movement always occurs last in the
negative direction.

To implement a branching sequen-
cer such as this. it is necessary to add
only one additional decoder 1C to the
basic controller circuit. The actual
wiring is shown in Figure 7. The ac.
tions of this controller are identical to
the simpler three-chip system pre-
viously discussed as long as the basic.
nonbranching sequence is followed.

If a branch is called for. however (if
the selected data channel of the multi-
plexer is LO when a potentially bran-
chable state is reached). the following
events occur. The L0 state of output
Y from the multiplexer causes the
function pulse to be generated by the
second decoder instead of the first de-
coder. This function pulse is used by
the instrument to perform tasks in the
same manner as any function pulse

 

from the first decoder. In addition.
this pulse is also fed back to the state
counter where it serves simultaneously
to provide binary data representing
the state to which to branch as well as
to load those data into the state coun-
ter. For example, in the positioner se-
quence. if the system is under comput-
er control when State 1 is entered. the
LO level at the data Channel 1 of the
multiplexer disables the first decoder.
but causes the second decoder to gen-
erate Function BZ. Function B2 is
used in the circuitry to update the di-
rection of movement and start the
motor. just as Function 01 of the first
decoder does under local control. But
it is also present at the C and Load (L)
inputs of the state counter so that as
it occurs. the state counter is forced
to skip directly to State 4 through par-
allel loading. All additional functions
that are part of other such branches
are similarly directed back to the
Load and appropriate data inputs of
the state counter. by use of simple
gates as shown to OR them together
as required.

By adding additional multiplexer
and decoder chips to the controller.
it is possible to produce sequencers of
even greater complexity in which the
branches consist not only of single

 

 

 

 

 

rescue- I!
O"
ensure-
"I“
It!!!
IMHO! It
stern ‘
III!!!"

 

 

 

 

  

 

0.81!” O

 

In!!!” t!

 

 

 

I'M'Iu Cll
F

 

 

I'm?!” '(l
—

 

II? N

 

 

 

 

 

 

 

 

 

 

   

 

 

 

Figure 6. Sequencer flow chart tor «mm positioner

yer-cue- si

Jaim-

     

215

  
 
 

      
    
 

I1

 
   

AT!
COUIYII

IULTIPLIIII

    
 
 

chfloll

UICDDII

 

Figure 7. Schematic diagram at filament positioning controller

function pulses. but also of additional
states or even complete subsequences
reminiscent of nested computer sub-
routines (1). Similarly. it is also possi-
ble to substitute basic chips different
from those employed here to produce
sequences of more than eight basic
states However. such systems can
quickly become unwieldy In their
complexity and might very well be
better performed using microproces-
sor-managed controllers even if the
user has no previous experience with
such devices (4—6).

Conclusions

We have described in this paper a
hardware approach for designing in-
strument sequencers. We feel that this
approach greatly reduces design time,
yet allows complex sequences to be
readily implemented The increasing
education of chemists as to the poten-
tial. the software, and the interfacing
of microprocessors will certainly lead
to a level of controller complexity at
which a microprocessor is the sequen-
cer of choice. Even so. the concepts
presented here should enable simple
controllers to be designed and imple-
mented readily by chemists unfamiliar
with microprocessors or for applica.
tions in which it is undesirable to “tie
up" the processor in simple. unchang-
ing control tasks

In addition. the hardware approach
is advantageous for high- speed control
functions and for applications that in-
volve the interleaving of tasks with
quite dissimilar timing requirements.
In fact. we have found In our own lab-
oratories that hardware sequencers
are important adjuncts to instrumen-

tation systems supervised by mini-
computers or microcomputers. Be-
sides the electrothermal atomizer po-
sitioning systems. we have used this
approach to design a minicomputer
data acquisition system and a commu-
nications network between several mi-
croprocessors and a minicomputer.
The simplicity and power of the ap-
proach should prove valuable to other
chemists involved in the automation
of chemical instrumentation.

Acknowledgment

We are grateful to T. V. Atkinson
for his assistance in the use of a com-
puterized graphics facility for the pro-
duction of our figures.

Relerences
(l) C L. Richards.£lerlronics, 46(3). 107

(l 97)3

(2) T. R.B esklee.‘ Digital Deslign with
Standardl M51 and LSI". 17 20.
“9’71ey-lnterscience. New ork N. Y..

(3)19H.5 V. Malmstadt. C. G. Enke andS.
RC rcou .“Electronic Measurements
for Scientists". pp 6627
York. N.Y. 179 .

(4) D R. McGlynn.“Micr rocesso rs"
Wiley- lnteracience Ne! 'ork. N Y..

(5) B. Soucek “Microprocessors and Mi-
crocom uters .“'iley- lnterscience. New
York. M
(6 ) T. R. Bla kleslee.‘ Digital Design with
Standard MS! and LSl' -232
“9'i7ley- -lnterscience New York?0 N.

.Benjamin New

'..B was recipientn briflh a summer and full-
year ACS Analytical Fellowship under the span-
somhip of (Procter & Gamble Co. Cincinnati.

 

S. R. Crouch

 

D. N. Baxter

 

E. H. Pals

 

E, R. Johnson