. . .
. . 0.1.1:“. 01:“) L
.111. . 3.; .5
.1. 2,215» . .2 3.59;... l x. .
i. a: :1... s.
.1 5.. . . .1...

 

 

 

 

 

I.\IA>-plv‘ a; .
r1}. 5:4 . .ti
1:

;:;suz3frs:;-‘::.‘

.PFI: l.‘ '0‘?!- ‘XI'.7 0‘ . ‘l‘
a 95...... .10.!“ 0". 'nvt4rrl.tt. p .‘urvlr :zkonlll
...£..I,nn.....np... {.2 Janna .1........i...1.....1:
.. . .31.. 1.....ltsirxslnu
’. It. I. ’1' .l.u"-‘ala|.".fi
. LL... .13..

91.3.1.3

hK.,1urv.l.rt-.E?.1vv.11:35:11! .
.1.0.!335-!.ID¢.!O;:lsuuifi.!.cbt.1al1l.aiu
..I\... ‘79:! inf-ail :1!!! .A..\...L§Er¢.11{r
9'. v .t. 1:1!!!{C—1nuf. 1.11.1..1I 1.5119. 1.
. 3-1)... .3 ...1
J... 1 q. I 3.:
111‘ .try.¢1v‘va ins... :1
. 19.5.231‘45,
0 A 1.9.! lulv4Lf.I:I:|
"51‘

. - .
5‘5... '1'.

. a

.s . .Ii
11' .. .1 {it

1. 11.)»
1:. . zit-.3...»

1 n1s1f11|llllcn
In. c;.l-e)11.£11ltf
1.. I:A....L?f!1.:...r.
. :rltlo'4lk .

 

.ILIADI!‘ .11
r.‘ tttt..su

    
  
  

       

  
       

:.ll0.v.v\ufih I)... v

 
     
            

 

     

  

. if:
- Y’ODOI'.)I’.I|\DZ(:. .10.: EfSEJL III1111|I\| I? |
. . shun-fi‘antxllnall’phllluil'l t:n.!)}.l.1.l\s.l. luv”), .fi gill 11.» I, 3 1t 1
. .. 3 3.11.1. -ulairlx‘lsll31..!;iilalfflnzlsf .{lfttnu I1:|lll1-1 1.. .
. .93“ I‘I’:O l .:-‘n’ll 'I.‘E3.¢.“5§Cz’ '5. élnii",’ it u 7' vl 0.! J? .311 ' 4 1' 1| I 1 r I I\.! I. Q
.. in!!! ES . .. . . 1 (1.. .5 . 1 I! l I!
.. .1001 .1 . 5.0.1.».1363‘! x11..v...afi\lntal|bnlvxrf tnfliuhyvtllziuflhuaflfl 1.311513 .411 .. 1.15311}: 1 . .w unit‘ii

. . 1111...}... 3.1.1:». 1):...3. 1:35:51..3391131121xxuilclhwxnwunfiuuuh . 1 . , 1.. . 1 VIII. 0 I
. .l ‘5. of- 1 A In... Jg.fl..g..uflu oauvIWInhufl .12."... r§uJfiSYuIJN§HuwbnllJlirzg . , . . o. _ .0 . .5. r1! .

....Ilv '1. .II 1 I ‘H. . xx! 13:9... 11 . . vii . (Ll. .III Ir.- .

11L. . 2 ... «a. 1571...»... 0.11 1.1.151. 13...? o. a . Egg . . Irfi‘ot.».t..1v.f. .. I II

 

iunu. m...um,...mu.1flhue.§.l haMWduJM-flmrflmflhfinmuvflflu z. .7: 2, L . , . .

 

‘ Yflwfli

Illlllllllllllllllllll‘lllllllllll 2 m s . w

3 129300 779

 

 

 

 

 

F: ‘
LIBRARY

Michigan State
University

K I

 

 

This'is to certify that the

thesis entitled

CONSTRUCTION OF A DEMONSTRATION PROTOTYPE OF AN
EXPERT SYSTEM FOR SCHEDULING PART-TIME STUDENT
WORKERS IN A UNIVERSITY RESIDENCE HALL FOODSERVICE

OPERATION
presented by

POHSIANG TSENG

has been accepted towards fulfillment
of the requirements for

, a o 0
Masters degree in Inst1tut1onal Adm1n.

 

 

dew/C Jana/«27W

Major professor

Datefla? JV] I90/0

0-7639 MS U is an Affirmative Action/Equal Opportunity Institution

PLACE IN RETURN BOX to remove We checkout from your record.
TO AVOID FINES return on or before due due.

“ DATE DUE DATE DUE DATE DUE

I
-| ‘3
W fl
|__L______=_______]:

__lg._é
WWW—l

MSU to An Afflrmdlve ActioNEquel Opportunity lnetltuflon
cmm1

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

CONSTRUCTION OF A DEMONSTRATION PROTOTYPE OF AN EXPERT
SYSTEM FOR SCHEDULING PART-TIRE STUDENT WORKERS IN 3

UNIVERSITY RESIDENCE HALL FOODSERVICE OPERATION

BY

Pohsiang Tseng

A THESIS

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

MASTER OF SCIENCE

Department of Food Science and Human Nutrition

1990

wv
I e

V f '

4,545)”

"o
.x

' ABSTRACT

CONSTRUCTION OF A DEMONSTRATION PROTOTYPE OF AN EXPERT
SYSTEM FOR SCHEDULING PART-TIME STUDENT WORKERS IN A
UNIVERSITY RESIDENCE BALL FOODSERVICE OPERATION

BY

Pohsiang Tseng

The purposes of this study were to (1) obtain objective
time data on manual scheduling of part-time student workers
in a university residence hall foodservice operation, (2)
construct an expert system demonstration prototype for
scheduling part-time student workers in a university residence
hall foodservice operation to demonstrate feasibility of use
of expert systems in foodservice management, (3) develop a
method for evaluation of the expert system, and (4) evaluate
the demonstration prototype of the expert system in a
laboratory environment. Time data recorded by residence hall
schedulers for three terms were used to determine total time
needed for scheduling manually per term; mean times for
scheduling one student worker per term at three residence
halls ranged from 16.66 to 62.76 minutes. The demonstration
prototype has completed weekday scheduling in a laboratory
environment. This result suggested that expert systems
technology could effectively be applied to schedule part-time

student workers in a university foodservice operation.

Dedicated to My Parerlts.
Mr. and Mrs. Tseng

ii

ACKNOWLEDGMENTS

The author wishes to express her sincere thanks and
deepest gratitude to her adviser, Dr. Carol A. Sawyer for her
guidance, assistance, and encouragement in her graduate study.
The author also wishes. to thank Mr. Robert Hauser, the
knowledge engineer of this project, for his understanding,
assistance, and advice.

Genuine appreciation is extended to the members of
Master's Degree committee, Dr. John A. Partridge, Dr. Jon
Sticklen, and Ms. Burness G. Wenberg, for reviewing and
commenting on this manuscript.

Sincere thanks and appreciation for their help are given
to Mr. Mike Gardner who is the Food Service Manager at Brody
Complex, Mr. Justin Terry and Douglas Fox who are supervisors
at Brody Complex, Mr. Scott Pierpont who used to be the
supervisor at Mason-Abbot Hall, and Ms. Anita Starks who is .
the supervisor at Shaw Hall. The author is also grateful to
her friends and colleagues, especially to Ms. Renee Polega
and Mr. Mung-Sen Sheu, for their help and encouragement.

Last but not least, the author wishes to express deepest
appreciation to her parents, Mr. W.K. Tseng and Mrs. C. H.
Tseng, and her host parent, Janice Cummings, for their love

and support.

iii

TABLE OP CONTENTS

PAGE

LIST OFTABLES .................................... viii
LIST OF FIGURES ..............J.................... x

1.0 INTRODUCTION ...................................... 1
2.0 LITERATURE REVIEW ................................. 7
2.1 Labor Scheduling .............................. 7

2.1.1 Labor Scheduling in Warehouse/

Distribution Facilities, Supermarkets,

and Grocery Stores .................... 7
2.1.2 Labor Scheduling in Underground

Mine planning ......................... 9
2.1.3 Labor Scheduling in the public

Accounting Firm ....................... 10
2.1.4 Single Shift Schedules with Variable

Demands ............................... 10
2.1.5 Labor Scheduling in the Foodservice

Industry .............................. 11
2.1.6 Summary 13

2.2 Time Study ...OOOOOOOOOOOOOOOOOOOOOOO0.00...... 14

2.2.1 The First Phase of Work Measurement .... 14
2.2.2 The Second Phase of Work Measurement ... 15
2.2.3 summary ......OOOOOOOOOOOIOOOOI.0.00.... 19

2.3 Expert Systems ................................ 20

2.3.1 Rule-Based Expert Systems .............. 21
~2.3.2 Generic Tasks: Expert Systems Beyond

Rules ................................. 22
.3 Introduction to DSPL ................... 24
.4 The Advantages of Expert Systems ....... 26

iv

2.4 Building an Expert System .....................

2.4.1 The Identification Phase ...... ..... ....
2.4.2 Knowledge Acquisition ............. .....

2.4.2.1 The elicitation of domain
knowledge

2.4.2.2 The analysis of elicited
knowledge ....................

2.4.2.3 Knowledge representation ......

2.4.3 Testing and Evaluation of Expert
SYStems ......OOCOOOOOOOOOOO0....O...0.

2.4.3.1 Establishing criteria .........
2.4.3.2 Evaluation approaches .........

3.0 MATERIALS AND METHODS .............................

3.1 Collection of Time Data on Manual Scheduling
of Part-Time Student Workers .................

3.1.1 Developing the Worksheet for -

collection of Time Data ...............
3.1.2 Determining the Sampling Frame .........
3.1.3 Conducting the Collection of Time Data .
3.1.4 Analyzing the Data .....................

3.2 Developing a Demonstration Prototype of an
Expert system for Scheduling Part-Time
Student Workers in a Residence Hall
Foodservice Operation at Michigan State
University (MSU) .............................

3.2.1 The Characteristics of Manual

Scheduling of Part-Time Student

Workers ...............................
3.2.2 Rationale for Choice of Conventional

Software vs. Expert Systems ...........
3.2.3 Evaluating Suitability of DSPL for

the SPRF Domain
3.2.4 The Development of a SPRF Expert

System Demonstration Prototype ........

3.2.4.1 Identifying the Scope of the
SPRF PrOblem ......OOOOOOOOOO.
3.2.4.2 Knowledge Acquisition .........

3.2.4.2.1 Interviewing the
expert

27

28

30

31

33

34

35

36
38

41

42

42
43
51
52

52

52

‘55

58

59

62
64

65

‘ 3.2.4.2.2 Analyzing knowledge
' elicited from
interview
3.2.4.2.3 Representing
knowledge ..........

3.2.4.3 Testing and Evaluation of the
SPRF Expert System ...........

3.2.4.3.1 Evaluation criteria .
3.2.4.3.2 Evaluation approaches
400 RESULTS AND DIstSION 0.00.0000...OOOOOOOOOOOOOOO.

4.1 Manual Method to Schedule Part-Time Student
Workers ......................................

4.2 Collection of Time Data on Manual Scheduling
of Part-Time Student Workers .................

Scheduling Methods Differ Among Halls ..
Scheduling Methods Differ Among
Schedulers in One Ball ................
The Quantity of Returning vs. New
Student Workers .......................
Human Error on Collection of Time Data .
Other Factors ..........................
Summary ................................

4
4

UMP

.h

2
2
2
2
2
2

chubb
mow-h

4.3 The Development of a SPRF Expert System
Demonstration Prototype ......................

4.3.1 Status of the SPRF Expert System
Demonstration Prototype ...............

4.3.2 Testing and Evaluation of the SPRF
Expert System Demonstration Prototype
in a Laboratory Environment ...........

5.0 CONCLUSIONS Q..........OOOOOOOCOOOCCOOO0.......0...

5.1 Implications of This Study for the
Foodservice Industry .........................
5.2 Difficulties with the SPRF Expert System
Demonstration Prototype Development ..........

5.2.1 Inherent Limitations of Expert Systems .

5.2.2 Limitations of the Expert System
Building T001 "DSPL"OOOOOOOO0.00.0.0...

vi

66

68

69

69
73

76

81
85
88
89
89
93

93
94

94

98

100

109

109

110

110

111

5.3 The Potential Benefits of Using the SPRF
Expert system ..OOOOOOOOOOOOOOOOOOOOOOOOOOOOOO 112

5.3.1 Ensuring Consistent Performance ........ 113
5.3.2 Reducing Supervisor Time for

Scheduling ............................ 113
5.3.3 Reducing Labor Costs ................... 114
5.3.4 Preserving and Expanding Expertise ..... 114

RECOMMENDATIONS FOR FURTHER RESEARCH .............. 115

6 O 1 ResearCh Prototype O O O O O O O O O O O O O O O O O O O O O O O O O O O O 115
6 O 2 Field PrOtOtype O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O 117
6.3 Production Model and Commercial System ........ 118

GLOSSARY OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO 120
APPENDICES 0.000000.00000000000000000000000000000000 122

A. Data Obtained from Collection of Time Data on
Manual Scheduling of Cafeteria Part-Time
Student Workers ............................... 122
B. Knowledge Representation in DSPL ............... 128
C. FSA (Foodservice Scheduling Assistant: the SPRF
Expert System Demonstration Prototype)
Interface and Scheduling Strategies ........... 130
D. Master Schedule (A Sample Print-out Before
Completion of User Interface) ................. 134
E. Student Worker List (A Sample Print-out Before
Completion of User Interface) ................. 135

LIST OFREFERENCES .OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO 136

vii

LIST OF TABLES

Table

1.

10.

11.

12.

Work function activities in a university residence
hall vegetable pre-preparation unit ...............

Elements for carrot preparation in a university,
residence hall vegetable pre-preparation unit .....

The characteristics of quality software ...........

Worksheet for collection of time data on manual
scheduling of part-time student workers ...........

Code number and elements for term scheduling ......
Code number of elements for final scheduling ......
Code number of elements for volunteer scheduling ..

Criteria for the evaluation of optimal manual and
SPRF expert system scheduling of part-time

student workers in the university residence

hall foodservice operation ........................

A check-list for evaluation of the SPRF expert
system demonstration prototype in a laboratory
enViroment OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO

Number of students and times used to manually
schedule cafeteria student workers at Mason-Abbot
Hall, MiChigan State univerSity O O O O O O O O O O O O O O O O O O O

Number of students and times used to manually
schedule cafeteria student workers at Shaw Hall,
Michigan state University .........................

Number of students and times used to manually

schedule cafeteria student workers at Brody
Complex, Michigan state University ................

viii

Page

16

17

37

44
45

47

70

74

78

79

80

13.

14.

15.

16.

17.

Number of students and times used to manually

schedule cafeteria student workers at Mason-

Abbot Hall, Shaw Hall, and Brody Complex,

Michigan State university ......................... 86

Factors accounting for the variability of results
generated from the collection of time data at three
residence halls during three terms ................ 95

The results of evaluation of the SPRF expert
system demonstration prototype in a laboratory
enVironment OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO 101

The contrast between manual scheduling of

part-time student workers (MSPS) and scheduling

by the SPRF expert system demonstration prototype

on the number of open work shifts of the

completed weekday lunches schedule at Brody

Complex, Michigan State University, Winter

Term 1990 ......................................... 104

The contrast between manual scheduling of

partftime student workers (MSPS) and scheduling

by the SPRF expert system demonstration prototype

on the number of open work shifts of the

completed weekday schedule at Brody Complex,

Michigan State University, Winter Term 1990 ....... 108

ix

LIST OF FIGURES

Figure Page
1. The evolution of an expert system ........... ...... 5

2. The structure of knowledge base and inference
engine OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO 22

3. The process of developing a SPRF expert system
demonstration prototype in DSPL .......... ....... .. 61

4. Final exam student employment schedule sheet ...... 9O

5. Application for student foodservice employment

Sheet OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO 91

6. The scheduling structure implemented in the SPRF
expert system demonstration prototype ............. 99

. Chapter I

INTRODUCTION

In warehouse/distribution facilities, supermarkets,
retail stores, mining operations, hospitals, and accounting
firms, computerized labor scheduling has been viewed as one
way to improve manual labor scheduling. Computerized labor
scheduling has successfully reduced labor-related expenditures
through a combination of efficiency improvements and personnel
cuts, enhanced employee morale and the firm's operations
(Fensholt, 1988; Britton, 1987; Desai, 1987; Burns and.Carter,
1985).

Published information on computerized labor scheduling
in the foodservice industry is limited. Some software
application programs have been identified. In 1984, the CBORD
Group (Ithaca, NY), a leading’ supplier' of institutional
foodservice software, developed Labor Scheduling System, a
multi-functional labor scheduling application program in a
price of around $1000. This system could be used to aid in the
planning of employee work schedules and to analyze the
resulting labor costs. In 1986, SUPERBNED (Management
Robotics, Inc.), in a price of $1395, was developed to
determine work force and scheduling employees. Other similar
software in the price range of $995 to $2000 was: CIDER System.

~(Dining Data System; Benicia, CA), LABORSERVE (Practorcare

Inc.; San Diego, CA), Labor Scheduling (RestaurantComp:

2
Larkspur, CA), and People-Planner Scheduler (Information
Marketing Businesses Inc.: Cambridge, MA) . These programs were
characterized by the ability to schedule full-time employees;
they were not capable of handling part-time employees with
"intermittent" availability (e.g. an employee was available
from 9:00 am to noon and 2:00 pm to 4:00 pm).

In the foodservice industry, large numbers of student
workers were hired on a part-time basis by various segments,
especially in colleges and universities. Residence hall
foodservice personnel supervisors, such as those at Michigan
State University (MSU), had found labor scheduling to be time-
consuming. Each student worker must be matched with work
shifts that did not conflict with his/her class schedule.
Since personnel supervisors were also responsible for other
duties, such as supervising full-time employees and part-time
student workers, part-time student supervisors were employed
to perform some scheduling tasks.

The current study focused on the computerized scheduling
of part-time student workers in residence hall foodservice at
MSU. Results of this study would contribute to the further
development of computerized part-time labor scheduling for the
foodservice and other industries.

Conventional computer software as well as expert systems
have the potential to be used to construct computerized labor
scheduling programs. Conventional software, to date, has not

been equipped with the features described above. The strong

3

point of conventional software in general has been the ability
to use highly complex algorithms to sort through information
(words, numbers, data) in thousands of database records and
then read. them 'to determine a final answer. The labor
scheduling problem was not amenable to an algorithmic
solution. 0n the other hand, expert systems were heuristic in
nature. That is, they required the use of rules of thumb to
achieve acceptable solutions for sufficiently narrow problems
(Roadmmer’ and.‘White, 1988). In. addition, expert systems
represented knowledge in a structured way that showed the
relationships among knowledge, which could make the process
of problem solving more efficient (reducing the time spent in
processing unrelated knowledge: Waterman, 1986).

From this point of view, expert systems would seem to be
a better choice for computerized scheduling of part-time
student workers in the residence hall foodservice than '
conventional software. Thus, the expert system technique was
used to develop a computerized program for scheduling part-
time student workers in a residence hall foodservice at MSU.

DSPL (Design Structures and Plans Language), an expert
system building tool, was developed to perform problem solving
of routine design in which standard.methods of completing the
task were fixed and well-known (Brown, 1987) . Scheduling part—
time student workers was not a typical routine design, but it

may stretch the boundaries of routine design (Kern et al.,

1989). Thus, DSPL was chosen to construct this expert system.

4

Testing and evaluation, important elements in the
development of expert systems, enabled a feedback process to
take jplace ‘whereby the comments served as a basis for
iterative refinements of the expert system (Liebowitz, 1986).
During the development of an expert system, testing in a
laboratory environment was usually needed before the system
could. be released for field testing (test in the user
environment: Waterman, 1986). Therefore, methods for testing
and evaluation of the expert system in both the laboratory and
user environment should be developed at the time the system
is being designed.

Also, to be able to compare the expert system to the
manual system, a study was done to determine the amount of
time needed for scheduling student workers manually. Data
generated from time study were used to evaluate the completed
expert system.

The evolution of an expert system could be divided into
five stages: demonstration prototype, research prototype,
field prototype, production model, and commercial system.
(Figure 1: Waterman, 1986) . Thus, the objectives of this study
were as follows:

1. To obtain objective time data (e.g. the amount of

time used in term scheduling) on manual scheduling
of part-time student workers in a university

residence hall foodservice operation.

 

Demonstration Prototype: The system solves a
portion of the problem
undertaken, suggesting
that system development
is achievable.

{

Research Prototype: The system displays

- credible performance
on the entire problem
but may be fragile due
to incomplete testing
and revision.

i

Field Prototype: The system displays
good performance and
has been revised based ,
on testing in the user
environment.

1

Production Model: The system exhibits
high quality, reliable,
fast, and efficiency in
the user environment.

Commercial System: The system is a productio

model being used on a
regular commercial basis.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

' Waterman (1986).

Figure 1. The evolution of an expert system (Waterman,

1986).

6

To construct an expert system demonstration
prototype for scheduling part-time student workers
in a university residence hall foodservice operation
to demonstrate feasibility of*use of expert systems
in foodservice management.

To develop a method for evaluation of the expert
system.

To evaluate the demonstration prototype of the

expert system in a laboratory environment.

Chapter II

LITERATURE REVIEW

The literature review for the current study was divided
into five parts: labor scheduling, time study, expert systems,

and building an expert.system.

MW
Discussion of the labor scheduling problem was divided
into five parts: labor' scheduling' warehouse/distribution
facilities, supermarkets, and grocery stores; underground mine
planning labor scheduling: the public accounting firm labor
scheduling; single shift scheduling with variable demands; and

labor scheduling in the foodservice industry.

. _1-o gedu, :- ,g . -,._ - 1'st 'ou '., F. i ' '-s

u a s

In warehouse/distribution facilities, supermarkets, and
grocery stores, the major labor scheduling problem supervisors
had faced was unable to use workers effectively (Fensholt,
1988). First, supervisors often did not know exactly how much
time each_task should take. Therefore, they were unable to
ensure that the right number of people were located in the
right.place:at the right time, and that the appropriate amount
of work was assigned to each worker. Often employees found

themselves with "leftover" empty hours and subsequently job

8
efficiency decreased. Also, after supervisors spent many of
their own working hours plotting labor schedules manually,
they had an insufficient amount of time remaining to follow
up on the performance of employees.

Based.on.the observations above, measuring and recording
the amount of time each task should take may be a solution for
such a labor scheduling problem. However, two problems may
arise. First, it was difficult to identify all of the various
tasks and then to give each worker the required number of
working hours when a large number of employees and tasks were
involved. Second, the time requirement for tasks varied with
the number of orders and customers. When the number of
customers or orders increased, so did the time requirement of
tasks. Therefore, to react to changing demands by re-adjusting
schedules in short periods of time was very difficult
(Fensholt, 1988).

Computerized labor scheduling programs have been used in
real world applications (warehouses, supermarkets, and grocery
stores); two examples were the Baum system (EriC' C. Baum
Consulting Firm; Chicago, IL) and COMPU-SKED system (Compu-
Sked Co.: Fort Lee, N.J.). These two programs could
automatically calculate exactly how many people were needed
per shift and print out individual picking orders that told
each employee just how many tasks he/she was responsible for
and how much time they had to finish each task. In this way,

a warehouse/distribution facility has been able to reduce its

9
labor-related expenditures by at least 25% through a
combination of efficiency improvements and personnel cuts.
About 200 Compu-sked installations, at Roundy's, Rice
Supermarkets (Houston), Bells (Rochester, N.Y.), Crook's
(Nashville) and others, had reported a fast return on their
investment through a decrease in labor costs because labor

hours (300 to 500 labor hours) were cut (Fensholt, 1988).

J... ,-._ ,. _, ,.- . . ,. , ,- ~_.,, ,.

The workplace of the average underground miner was filled
with unseen hazards, tiring work tasks, and long periods of
isolation. It was important to make an equitable arrangement
(schedule) for all mining workers according to their
experience, job skill required, their performance, acceptable
levels of risk, etc. However, to make an equitable arrangement
for mine managers or superintendents with so little time
before shift changes was difficult (Britton, 1987).

An expert system (called CHOOZ and written in Turbo
Prolog software), developed by Tanoma Coal Co. N.V., was to
automatically search and compile work crews, premium work
lists, and suggest strategies on deploying a work force for
its maximum productive effort and achieving an equitable
arrangement. At the same time, CHOOZ was also expected to
provide varying levels of information based on acceptable
risk, which would be valuable to the ever changing physical

environment of underground mining (Britton, 1987).

10
2.1.3 Luuoz Sgneguling in the Public Acgouutiug Eiru

Scheduling was an important administrative, planning, and
operational function of the public accounting firm. When a
scheduling staff was unable to consider all scheduling factors
appropriately (e.g. staff utilization, manpower requirements,
personnel skills, individual development), the firm's
resources would be mis-allocated through the overuse of some
staff and the underuse of others. Such inefficiency was
frustrating for all involved. Administrators complained about
the lack.of chargeable time. Underemployed.professional staff
were dissatisfied because they were not challenged, while
overemployed staff became burned out due to the» heavy
workload.

Since the foundation of a public accounting firm was its
professional staff, a CAP firm incorporated the microcomputer
into the manual scheduling process to enhance the firm's
operations and reduce employee dissatisfaction. This system
would provide a higher outlook on enhancing operations and

employee satisfaction (Desai, 1987).

In recent years, labor negotiations for seven-day-week
organizations such as hospitals, mining companies and chemical
industries have involved an increased emphasis on improving

shift schedules. A study by Burns and Carter (1985) had given

an exact lower boundary on the number of workers required to

‘ 11
satisfy contractual commitments, such as ensuring that each
employee received at least A out of every B weekends off,
everyone worked exactly five days per week, and no one worked
more than six consecutive days. A linear time algorithm was
presented, which generated .schedules satisfying all the

primary objectives (Burns and Carter, 1985).

WWW
Published information on computerized labor scheduling
in the foodservice industry is limited.— Some software
application programs have been identified. In 1984, the CBORD
Group (Ithaca, NY), a leading' supplier' of institutional
foodservice software, developed Labor Scheduling System, a
multi-functional labor scheduling application program in a
price of around $1000. This system.could.be used to aid in the
planning of employee work schedules and to analyze the
resulting labor costs. In 1986, SUPERSIBD (Management
Robotics, Inc.), in a price of $1395, was developed to
determine work force and schedule employees. That is. this
program could determine how many people were needed on each
shift based on "sales" (presumed indicator of amount of
business. Then this program filled in employees on each shift
(e.g. 6:00 am to 3:00 pm was one shift) according to their
skill levels, maximum number of work day per week, maximum
number of shifts per week, etc.). Other similar software in

the price range of $995 to $2000 was: CIDER System (Dining

12
Data System: Benicia, CA), LABORSERVE (Practorcare Inc.; San
Diego, CA), Labor Scheduling (RestaurantComp: Larkspur, CA),
and People-Planner Scheduler (Information Marketing Businesses
Inc.; Cambridge, MA).

However, these programs were characterized by the ability
to schedule full-time employees: they were not capable of
handling part-atime employees with "intermittent" availability
(e.g. an employee was available from 9:00 am to noon and 2:00
pm to 4:00 pm).

In the foodservice industry, large numbers of student
workers were hired on a part-time basis by various segments,
especially in colleges and universities. Residence hall
foodservice personnel supervisors, such as those at Michigan
State University (MSU) , had found labor scheduling to be time-
consuming. Each student worker must be matched with work
shifts that did not conflict with his/her class schedule.
Personnel supervisors were also responsible for other duties,
such as personnel management, determining the amount of work
force for next term, and adjusting the number of shifts.
Personnel supervisors could be better able to use their time
and abilities on the really tough problems if they could be
relieved from handling scheduling problem which, for the
experts, was routine (Briggs and Doney, 1988).

Therefore, efforts should be made to computerize the
scheduling of part-time student workers in university

foodservice operations.

13
2 . a

Even though labor scheduling problems varied among
different industries, computerized labor scheduling had been
most often thought of as the best way to improve current
manual labor scheduling. Existing applications had proved that
computerized labor scheduling was more beneficial than manual
scheduling in terms of reducing labor-related expenditures
through a combination of efficiency improvements and personnel
cuts, reducing employee dissatisfaction, and enhancing the
organization's operations.

However, current applications were characterized by
scheduling full-time employees. They were not capable of
handling large numbers of ‘part-time employees with
"intermittent" availability (e.g. an employee was available
from 9:00 am to noon and 2:00 pm to 4:00 pm).

Part-time employment has grown over most of the post-WWII .
period; more than 15% of all adults in non-farm civilian jobs
work less than 35 hours a week (Otten, 1990). Therefore,
efforts should also be made to schedule part-time employees
by extending current available software or developing new
programs._

Residence hall foodservice personnel supervisors, such
as those at Michigan State University (MSU), had found labor
scheduling to be time-consuming. To be better able to use
their 'time and abilities on. the really tough jproblems,

personnel supervisors should be relieved from handling

14
scheduling problem which, for the experts, was routine.
Therefore, computerized scheduling of part-time student
workers in the university residence hall foodservice operation

would be needed in the future.

W

Decisions regarding human work activities were often
based on subjective beliefs rather than objective data. Work
study could provide objective data about activities concerning
people, facilities, and equipment as a means to improve those
activities (Currie, 1977).

Two distinct, yet interdependent, approaches were used
in work study: (1) method study that referred to the way in
which work was done and (2) work measurement that pertained
to the time required to complete a particular task (Block et
a1., 1985). One purpose of this study was to determine the
amount of time needed for scheduling student workers.
Therefore, in the section below, attention was given to the
work.measurement approach. Discussion of the work measurement
approach is presented in two phases (Sections 2.2.1 and

2.2.2).

2.2.; The Figst Phase of Wozk Measureuehu

The first phase of work measurement was to identify work

function activities of the object studied. (e.g. labor

15
scheduling) and then to classify them into steps referred to
as elements (Mundel, 1978; Block et al., 1985).

In the study by Block et al. (1985) , work function
activities were described as a group of similar activities
that may be recognized by sight and may be considered
homogeneous. Work function activities were categorized into
three major groups: direct work, indirect work, and delays.
Direct work functions were defined as any essential activity
that. contributed. directly' to the jproduction of the end
product. Indirect work functions were any catalytic activity
that contributed to the production of the end product. Delays
were all times when an employee was scheduled to be working
but. was not engaged in either* direct or indirect. work
functions. The time study of a vegetable pre-preparation unit,
an example, is shown in Table 1 (Block et al., 1985).

An element consisted of a unified group of motions, such
as taking hold of an object, moving it, and placing it (Table
2). Each element was the smallest practical unit for time
measurements and had the well-defined end point for timing

(Mundel, 1978).

2.3.2 The Second Phase of Work Measurement

The second phase of work measurement was to choose a work
measurement technique. Work measurement techniques that had
been used were survey, work sampling, Master Standard Data

(MSD) quantity food production code, and MM-3 dietary

0

Table 1. Work function activities in a university residence
hall vegetable pre-preparation unit' '

 

Direct Work Function

Pre-preparation

carrots

celery

lettuce, head and leaf

onions, yellow mature

all other fruits and
vegetable

 

Indirect Work Function

Transportation of Food

Transportation of
Equipment or Supplies

pot and pan washing
Ahousekeeping
instruction

 

Delay Time

 

4

 

Personal delay

meal time
break time

Idle Time

 

° Block.et al. (1985)

 

17

Table 2. Elements for carrot pre-preparation in a university
residence hall vegetable pre-preparation unit'

 

Work Function

Element

 

Pre-preparati

 

Ol'!

 

1.
2.
3.
4.
5.
6.

8.
9.
10.
11.
12.

move out of storage
open sack

remove from sack
peel and nub

wash

place in container
cover

move into storage
remove onto rack
weigh

remove from container
cut into sticks

 

 

' Block et al.

(1985)

18
methodology (Mundel, 1978: Zemel and Matthews, 1982: Olsen and
Meyer, 1987).

The survey work measurement technique involved collecting
data from predetermined samples with a questionnaire or
worksheet, then analyzing data and calculating labor time
spent in each practical unit (Mundel, 1978: Olsen and Meyer,
1987). Two steps for the development of the worksheet were as
follows:

1. Develop a description of the elements of each work
function and check these against the requirements
for good elements (e.g., having easily detected and
defined. beginning/end points for timing, well?
unified group of motions) for time study elements.

2. Adjust the elements as necessary and then detail the
descriptions of the elements one by one.

Work sampling involved nonsequential observations at
random times; the standard number'of minutes that a qualified,
properly trained, and experienced person required to perform
a specific task when working at a normal pace could then be
determined (Olsen and Meyer, 1987).

MSD quantity food production code was the simulation of
production based on predetermined time values for specific
conditions. Scaled layout and standardized production formulas
were used to predetermine production time (Zemel and Matthews,

1982).

19
In MM-3 dietary methodology, data needed to calculate
production time included number of batches of product produced
per week, number of portions per week, and distances traveled.
These data were entered onto forms in the 1414-3 Dietary
Methodology Manual to estimate production time (Zemel and

Matthews, 1982).

2.2,3 Sunnanx
The labor intensity of hospital foodservice, coupled with

constantly rising labor costs, demanded that managers
carefully allocated labor resources to the production of menu
items to contain costs (Zemel and Matthews, 1982). In the
public sector or institutional foodservice segment, access to
information and research opportunities had been far greater
than from the private sector (Olsen and Meyer, 1987).
Therefore, work measurement had been a common approach used
in the published research on productivity in institutional
foodservice operations.

While studies of food production in the foodservice
industry had been helpful in gaining the understanding of
productivity, other functions performed by foodservice workers
and supervisors, such as service and scheduling workers, had
largely been ignored (Olsen and Meyer, 1987). Thus,
information was limited on the amount of supervisor time
required to schedule student workers in residence hall

foodservice. Such information would be useful to gain an

20
insight into the related labor (supervisor) costs and

usefulness of facilitating scheduling by computer.

212_EKP§IE_§!§§§m§

Expert systems were characterized by a collection of
general strategies that used knowledge in such a way that the
complexity inherent in certain tasks was minimized (Firdman,
1986). Thus, expert systems were also viewed as very
specialized computer programs which were created by using
extensive, high quality, specific knowledge about some narrow
problem area.

Expert systems differed from other computer programs in
that they were characterized by goal-directed behavior
(Firdman, 1986). Given the goals and strategies which
exploited typically heuristic knowledge about the problem
domain for goal achievement, expert systems were able to
explore very large problem spaces to make plausible
alternative options that contributed to goal achievement
effectively and efficiently. For example, the design of an
air-cylinder (AC) consisted of 19 requirements (e.g., rod
diameter, air inlet diameter, air pressure). The values for
each requirement may vary with different applications. Thus,
this design problem could be viewed as a search problem in a
large space. Each requirement of this space reflected a
possible candidate for the answer to the design problem. Given

a goal (e.g. a specific application) and strategies for the

, 21
goal. achievement, a subset of ‘the 19 requirements with
specific values was able to be identified without exhaustive
search in the entire problem space.
The computational methods of conventional computer
programs were not engaging in the exploration of underlying
problem space to make possible hypotheses by use of explicit

knowledge and general exploration strategies.

e- m

The knowledge in a rule-based expert system was organized
in a way that separated the knowledge about the problem domain
from the system's other knowledge, such as general knowledge
about how to solve problems or knowledge about how to interact
with the user (Waterman, 1986). Therefore, an expert system
usually consisted of two elements: a knowledge base and an
inference engine (Figure 2).

The knowledge base, as shown in Figure 2, contained facts
(e.g. student class schedules, job priority), and rules/or
other representations. The rule (or other representation)
referred to the type of knowledge that used.those facts as the
basis for decision making (e.g. if the student worker did not
have<class right before or after the work shifts, then assign
the work shifts to the student worker) (Waterman, 1986:
Roberts, 1988).

The inference engine contained an interpreter that

decided how to apply the rules to infer new knowledge and a

22

EXPERT SYSTEM

 

KNOWLEDGE BASE
(Domain knowledge)

L j

l RULES I

 

 

 

tNTERI’RETER
LSCHEDULER

INFERENCE ENGINE
(General
problem-solving
knowledge)

 

 

 

 

Figure 2. The structure of knowledge base and inference

engine (Waterman, 1986).

scheduler that decided the order in which the rules (or other.
representations) should be applied. That is, the inference
engine used a predefined control strategy to find the enabled
rules or _other heuristics and decided which one to apply

(Boritz and Brown, 1986).

2.3,; Genenic Tasks; Eugen; Sysgene Beyonu Rules

The separation of knowledge about the problem domain from

that knowledge used in inference processes had resulted in the

23

field of expert systems being stuck at too low a level of
abstraction that obscured the essential nature of the tasks
that.current systems perform. For example, the fact that MYCIN
(a medical diagnosis expert system) engaged in some form of
classification problem solving was not readily visible at the
level of the rule representation used to make inferences
(Brown and Chandrasekaran, 1989).

The roots of this problem were two-fold. First, the
designer had to undertake a complex programming effort in
order to make the translation of the form in which knowledge
was needed for design task from that of diagnosis. Second,
seeking uniform mechanisms came at a cost: the architectures
that supported this uniformity did so by suppressing the
distinctions in control and inference between different kinds
of tasks (e.g. routine design task, diagnosis task; Brown and
Chandrasekaran, 1989). i

The available paradigms often forced us to fit the
problem to the programming language rather than to fashion the
programming language to reflect the structure of the problem.
Since structures and control regimes for one specific domain
(e.g. medical diagnosis) continued to be different from other
domains (e.g. labor scheduling), efforts had been made to
propose an alternative level of abstraction for expert system
building. This alternative level of abstraction was called
generic task. The generic task was to reflect a theory

specific for one type of problem solving (e.g. scheduling)

24
in the expert-system-building tool (Chandrasekaran, 1983;
Chandrasekaran, 1986: Brown, 1987).

DSPL was one of the efforts to reflect the theory of the
problem solving of routine design tasks. DSPL had been thought
of as a "task level" language. In this way, DSPL.had.been able
to provide exactly the knowledge types and control structures
necessary for the task: and the knowledge engineers were more
easily able to express the domain knowledge in that language,
which dramatically reduced the time required to develop a

system (Brown, 1987).

2.3.; Ingnoduction IQ QSEL
Design Structures and Plans Language (DSPL), developed

by Brown and Chandrasekaran (1989) , was a programming language
for designing expert systems that performed problem solving
on routine design tasks.

Routine design was defined as a type of task in which
standard methods of completing the task were fixed and well-
known. Also, the task had been done many times before, each
time with different but similar requirements, and would be
done in the future again and again. For example, the design
of an air-cylinder (AC) consisted of 19 requirements (e.g. rod
diameter, air inlet diameter, air pressure). With different
applications, the AC needed to be designed and manufactured
again each time according to a subset of the 19 requirements.

On the other hand, all requirements may be involved in the AC

25
design but the values for each requirement would vary with
different applications. Nevertheless, the method of designing
an AC was fixed and repeated each time.

DSPL provided two useful knowledge representation
facilities: agents (e.g. specialist, plan) and. a design
database. Domain knowledge was encoded both in the agent
structure (inference structure) and the design database. The
procedural know1edge (e.g. goals and strategies for solving
scheduling problem) was reflected in the agent structure:
while the declarative knowledge, used in agent structure (e.g.
job priority), was represented in the design database (Brown
and Chandrasekaran 1989).

Agents, such as specialist, plan, etc., were organized
into a hierarchy. This hierarchy acted as a reasoning
mechanism: the top levels of this hierarchy were specialists
representing the more general concepts (e.g., designing the
spring of AC); while the lower levels.dealt.with.more specific
instances of those concepts (e.g. showing how to design the
spring, checking the requirements for design, designing the
spring desired). Based on this hierarchy, conclusions were
able to be generated.

The control regime for the DSPL hierarchy was top-down.
The following was done recursively until a complete design was
worked out: a specialist corresponding to a component (e.g.
weekday scheduling) of the task (e.g. term scheduling) was

called; the specialist (e.g. weekday specialist) in turn

26
suggested further sub-specialists (e.g. lunch sub-specialist)
to call to set other details of the design. Failures were
passed up until appropriate changes were made by higher level
specialists, so that specialists who failed at the first
attempt may succeed on a re-try (Chandrasekaran, 1986).

DSPL provided a way of writing declarations of agents
(e.g. specialist), which allowed the programmer to represent
the knowledge easily. After all agents required were declared
and checked, DSPL allowed its underlying system to link them
by a top-down control regime and construct a hierarchy. Then
the problem solving could be invoked by requesting a design
from the top-most specialist. After a successful completion,
the design data-base would contain the results of the

completed design (Brown and Chandrasekaran 1989).

2 3.4 e v nt e 8 st ms
Basically, the advantages of expert systems could be
viewed from the following aspects:
1. Expert systems were able tO‘ mimic the reasoning
process of human beings.
2 . - Expert knowledge became distributable and permanent.
3. Expert systems could reduce the time:needed.to solve
the problem and reach a conclusion.
First, as in number one above, when human experts solved
a problem (e.g. decision-making problem), they did not do it

by solving sets of equations or performing other laborious

27
mathematical computations. Instead, they chose symbols to
represent the problem concepts and applied various strategies
and.heuristics to manipulate these concepts (Waterman, 1986).

Expert systems understood not just symbols, or
information, but the information and its interconnections.
Therefore, performance of expert systems was closer to that
of human beings.

Second, expert systems could facilitate the preservation
and dissemination of scarce expertise by encoding the relevant
experience of an expert and making it generally available as
a resource. In this way, they were able to help less expert
professionals (e.g. trainees) improve the quality of their
judgments, decisions and, possibly, their general problem-
solving skills across many locations (Lippert, 1987; Boritz
and Brown, 1986).

Third, an expert system. had a perfect ‘memory and
represented the knowledge in an explicit and intelligible
manner. Thus it could help compensate for limitations in human
information-combining abilities and minimize biased
interpretations and incorrect inferences to drastically reduce

the time needed for problem solving (Boritz and Brown, 1986) .

2,4 Builglng en Expent Systen

The evolution of an expert system had been divided into
five stages: demonstration prototype, research prototype,

field prototype, production model, and commercial system

28
(Figure 1: Waterman, 1986).

The expert system development could be viewed as five
highly interdependent.and.overlapping phases: identification,
conceptualization, formalization, implementation, and testing
and evaluation (Buchanan et al., 1983). The identification
phase involved identifying the type and scope of the problem,
expert system building tools, and the goals or objectives of
building an expert system. The conceptualization phase
involved.deciding concepts, relations, and control mechanisms
needed to describe problem solving in the domain. The
formalization phase involved expressing the key concepts and
relations in some formal way, suggested by the expert system
building language. These two phases, conceptualization and
formalization, coupled together, could also be thought of as
knowledge acquisition (Kidd, 1987: Waterman, 1986). The
implementation phase turned the formalized knowledge into a
working computer program. Finally, the testing and evaluation
phase involved evaluating the performance and utility of the
prototype program and revising it as necessary (Waterman,

1986).

2 1 h d f’ at'o a

The identification phase involved identifying the type
and scope of the problem, and expert system building tools.
As to the type of problems, the problem targeted for the

expert system development required the use of heuristics (e.g.

29

rules of thumb, strategies, etc.). Problems involved in the
use of heuristics were difficult to attack by conventional
approaches, but. may be amenable to expert system
methodologies. In addition, the problem targeted for expert
system development did not require knowledge from a very large
number of areas. If it did, the amount of knowledge needed for
the expert system would be probably beyond its acceptable
limits. Also, it was difficult to combine very heterogeneous
knowledge. Attempts to aggregate knowledge across a number of
areas were certain. to mask. the true structure of that
knowledge (Prerau, 1985; Waterman, 1986).

As to the problem designed for expert system development,
the problem must be neither too easy (e.g. taking a human
expert less than a few minutes) nor too difficult (e.g.
requiring more than a few'hours for an expert). If the problem
was too easy, the development of the system would not warrant
the effort: if too difficult, the amount of knowledge needed
may be beyond the state of the art in knowledge base size. In
other words, the problem targeted should be sufficiently
narrow and self-contained: the aim was not for a system that
was expert in an entire domain, but for a system that was an
expert in a limited task within the domain (Prerau, 1985:
Waterman, 1986).

Expert system building tools were programming languages
and support packages used to build the expert system. Expert

system building tools made it possible to develop an expert

30
'system in less time than would be required with the use of
traditional development languages (Gevarter, 1987). However,
selecting the right tool that made the development of an
expert system easy and time saving was difficult because most
tools had been developed to handle only one particular class
of problem (e.g. rule-based problem solving). On the other
hand, AI researches were not really sure what tool features
were required by specific classes of problems. Despite the
fact that there was no easy answer, some basic guidelines for
deciding what tool was appropriate for a specific problem task
were suggested by AI' researchers (Waterman, 1986; Lippert,
1987).
1. Would the tool have the features suggested by the
needs of the problem?
2. What level of nesting of rules and goals would be
allowed?
3. What kind of inference procedure had been included?
Which would be dominant in this tool?
4. Would the tool be reliable?
5. Would the tool have the features suggested by the

-needs of the application?

2.5,; Knowledge ncguisition

Knowledge acquisition was the process of transferring
knowledge from the expert to the computer. One view of

knowledge acquisition had been to consider the expert's head

31

as being filled with bits of knowledge and the problem of
knowledge acquisition as being one of "mining those jewels of
knowledge out of their heads one by one." In short, knowledge
acquisition involved acquiring the right kinds of knowledge
from the expert, mapping that knowledge into a coherent
organizational structure, and then translating the mapped
knowledge into a formal representation suggested by the expert
system building tool to replicate the expert's knowledge
(Naughton, 1989).

From this point of view, three major functions in
knowledge acquisition could be given (Breuker and Wielinga,
1987).

1. The elicitation of domain knowledge. Elicitation was

to "mine the knowledge out of experts heads."

2. The analysis of elicited knowledge. Analysis wasthe
transformation of knowledge into an interpretative
framework.

3. Knowledge representation. The organized knowledge
was translated into a formal representation
suggested by the expert system building tool.

2.5,2,l The elieitetion of denaln knewledge, The
interview was the most acceptable approach used in eliciting
domain knowledge. There were three types of interview that
had been used: ( 1) interview with "thinking-aloud" and

"cross-examination", (2) on-site observation, and (3)

32
intuitive interview.

Kuipers and Kassirer (1987) proposed an interview that
began with a "thinking-aloud" segment and ended with a "cross-
examination" technique. In this type of interview, the expert
was asked to explain out loud as much as possible what he/she
thought about as he/she solved a problem. Then the knowledge
engineer asked probing questions about the expert's knowledge
of particular topics. Thus, interview with ”thinking-aloud"
and "cross-examination" experiment was much more effective for
determining the limits of the knowledge represented (Kuipers
and Kassirer, 1987).

On-site observation relied on watching the expert solve
realistic problems within the domain and not to say or do
anything that might influence the expert's problem-solving
approach. In other words, the knowledge engineer observed the
expert solving real problems on the job rather than contrived
problems in a laboratory setting: the knowledge engineer did
not interfere but rather acted as a passive observer. This
approach gave the knowledge engineer some insight into the
complexity of the problem and into the type of interface
facility -needed by the user to interact with the finished
system in the field (Waterman, 1986).

For' the intuitive interview; the 'knowledge engineer
studied and interacted with both experts and the literature
of a field in order to become familiar with problem-solving

methods. Then the investigator developed a representation of

33
expertise which was checked against the opinion of other

experts (Waterman, 1986).

W The second

function of knowledge acquisition was analysis of elicited
knowledge. In this function, elicited data were translated
into an interpretative framework. Breuker and Wielinga (1987)
proposed an interpretation model that could be used for
communication between expert and knowledge engineer to check
consistency and completeness of elicited knowledge and, in
particular, to facilitate the mapping of those data into
structures.

There were five different levels of representation of
knowledge involved in the interpretation model of Breuker and
Wielinga (1987) . These levels were knowledge identification,
knowledge conceptualization, epistemological analysis, logical
analysis, and implementational analysis.

At the level of knowledge identification, individual
concepts were identified. To reduce the amount of data,
individual concepts were organized by identity (e.g. type of
knowledge, such as classification knowledge) . Later on, at the
level of knowledge conceptualization, concepts became
integrated, according to a number of relations between
concepts (e.g. is-a and part-of).

At the level of epistemological analysis, the analysis

uncovered structured properties of expertise by identifying

34
and examining the following five types of information: object,
knowledge source, model, structure, and strategy.

Objects were the input or conclusions (output) of
inference processes: a knowledge source was a processor that
inferred new objects from given objects; models were parts of
supporting inference knowledge: structures could be sequences,
resemblance groupings,or hierarchies along with some
relations: and a strategy was a plan invoking problem-solving
in the domain.

At the level of logical analysis, a framework expressing
the relations among knowledge was formed for use at the
implementational analysis level.

Finally, at the implementational analysis level, all
levels previously described were used to make up the mechanism

which then was used for facilitating knowledge representation.

2,5,2,2 Knowledge rennesentatlen. The third function
of knowledge acquisition was knowledge representation. In this
function, key concepts, sub-problems, and control features
were mapped into a more formal representation suggested by an
expert system building tool. An expert system building tool
may support one or more methods for representing knowledge.
From this view, methods for representing knowledge varied with

the expert system building tool chosen.

35
ua ' o s 5

Testing involved evaluating the performance and utility
of the expert system prototype and revising it as necessary
(Waterman, 1986). The performance referred to the accuracy of
embedded knowledge and any advice or conclusions that the
software provided: the utility was defined as whether the
software produced useful results, the credibility of its
results, its reliability, its efficiency and speed, its ease
of interaction, and the extent of its capabilities (Gaschnig
et al., 1983).

Evaluation, an important element in the development of
expert systems, could uncover problems with knowledge
representation (e.g. relations and missing concepts), unwieldy
control mechanisms, interface, or conclusions represented at
the wrong level of details. Such problems could suggest
possible improvements for’ the current developed system.
Therefore, evaluation enabled a feedback process to take place
whereby the comments served as a basis for iterative
refinements of the expert system (Liebowitz, 1986).

During the development of expert systems, several tests
in a laboratory environment were usually needed before the
system could be released for field testing. As the field
testing was completed, new complications may arise and take
additional time to correct. Thus, re-testing in a laboratory
environment would again be required, and then field testing

would be repeated (Waterman, 1986).

36

Two major‘ tasks were involved in evaluation: the
development of criteria for the evaluation of the test
problems and the determination of evaluation approaches
(Harrison, 1989; Miller et al., 1985). These two tasks are

discussed below (Sections 2.4.3.1 and 2.4.3.2).

2,5,2.l Establlehlng onlgezie, A review of the

literature indicated that various criteria had been used to
evaluate the quality of expert systems (Liebowitz, 1986) .
Boehm. et al. (1978) studied ‘the development of quality _
software and grouped characteristics of quality software into
seven categories: portability, reliability, efficiency , human
engineering, testability, understandability, and
modifiability. Table 3 summaries the criteria in each category
(Liebowitz, 1986).

Gaschnig et a1. (1983) also identified evaluation
characteristics of expert systems. The four characteristics
identified by these authors included quality of the system
decisions and advice, correctness of the reasoning techniques
used, quality of the human-computer interaction (both its
content ,and the mechanical issues involved), system's
efficiency, and cost effectiveness.

Additionally, Miller et al. (1985) proposed four major
sub-tasks for use in the evaluation of a commercial expert
system: (1) evaluating the accuracy of the system, (2)

evaluating the system sizing and the performance of hardware

.Amwmav Nuflzonmflq “Awhmav .Hm um Ezmom a

 

.mcofluocSM Hmcoflumusmeoo ucwcomeou

 

 

 

37

no mucmEonflsomu ommuoum pump CH mcoflmcnoxm monopoeeooom >Hflmmm OH uxuflaflnnucmemsm mueaflnoemaooz
.mpoo ecu mcflpmou we pocumomflp
xaamom mum mucmewunum newcomeoo new no mmozu new cofluocsw mug ” apfiawnammq
.ucmmmum ma coHOMEOOHCH m>wmmmoxm o: " mmmcomflocou
.msumum new .mucocooeoo
.muoouso .musocq .mucemuumcoo .mCOeOQEsmmm .mo>wuomnno mow
modeumumo on umpmou m MOM coflum5u0mca nmsocm mcamucoo Ow “mmmcm>wuoauommpumaww muaaflnmocmumumoc:
.mucmcooeoo mow mo mom m>wuomamm one mmumuflaflflocm OH u xuwawowmmmooe
.mmousommu mo mumm3 unocuwz mmomusm mu“ maaflwasm OH H wocmfloawwm woe>mo woodwoflmum
.muumo ucwpcwompcfl
mua mo coHOMNflcmmuo mo cumuumo ouficflmmo o mommmmmoo be " mmwcpououosoum xuflaaomummfi

 

.stmms new mudeEHmmm on xmmo mum ucmucoo new BOOM wmoc3
Wusouso mmow>0uo new muses“ mo COHOmonaomom ecu monumuflaaomu OH

u mmocm>flumoHcsEEOU

chOOOCamcmlcmesm

 

.mco acouuso new core umcuo

 

 

 

 

mcoHumusmewcoo mum3pumr nouseeoo co nousomxm on coo mum3uuom uucwpcwmmpCHlmoH>mo mueaaenouuom
.mucmfimuasoou on» on flamenco» ma ucmucoo mua new .333
55a: xmoHonExm new £83:ng 50330: 503.5 mcwmucoo be " xocmumfimcou
.mms umpcmuca uwmnu ammeumm on mmwoouo xaucmwoflwmsm one mucouso mufi u zomusoog
.oooon>mp
xaasw me woman was no some new pronoun mum momma mumzuw0m emu «0 Han "mmmcmumameou. xuflaaomflaom
mUhREMflHQSSEU uflQHEEU

 

«oum3uu0m xuwamso mo mOHumHnmuOmnnco 019 .m manna

 

38

and software to find the lowest cost of implementation, (3)
evaluating the quality of human/computer interaction, and (4)
economic evaluation (Does the action carried out by the system
justify the costs of hardware, software, training and
maintenance of developed system).

Because initial goals and requirements of expert systems
differed, criteria also varied with each expert system

developed.

va u a A common approach used
for evaluation of expert systems was the judgment of experts.
There were two ways to use such expert judges. The first way
was to use judges (experts) by giving them the output from the
system and asking them to evaluate the results using pre-
determine criteria. The second way was to use blind
evaluation. That is, both experts and the developed system.
were asked to solve some test problems. The results may be put
into a standard format and the judges (experts) were asked to
evaluate the results without knowing whether the results were
those of an expert or from the computer system
(Chandrasekaran, 1983; Harrison, 1989).

Gaschnig (1982) described several evaluations of the
Prospector System (an expert system used in medical
diagnosis). These evaluations relied on comparisons of the
evaluations by Prospector and by human experts. The data from

each judge were gathered using a questionnaire that required

39
numeric answers reflecting the degree of agreement or
disagreement on a scale of -5 to +5 with regard to each
question given. The results were converted into an overall
score that reflected the goodness of fit. Then, the scores
from the expert were compared with the Prospector scores
(Harrison, 1989).

There had been no generally accepted method for
evaluating expert systems, although some work in this
direction had begun (Harrison, 1989; Liebowitz, 1986) . In lieu
of generally accepted methods, the three principles of
evaluation, proposed by Gaschnig et al. (1983), could be used
as a guide for evaluating expert systems.

1. Complex objects or processes could not be evaluated

by a single criterion or number.

2 . The larger the number of distinct criteria evaluated
or measurements taken, the more information would
be available on which to base an overall evalaiation.

3. People would disagree about the relative
significance of various criteria according to their
respective interests.

The same problems of reliability and validity that made
behavioral and cognitive evaluations of a human being
difficult also made the evaluation of expert systems difficult
(Harrison, 1989) . Additionally, when expert systems were

applied to the domains in which it was not easy to ascertain

40
correct answers or in which there was disagreement about what
the correct answer was, the evaluation would become difficult

to perform (Miller et al., 1985).

Chapter III

HATERIALS AND METHODS

The purposes of this study were to (1) obtain objective
time data (e.g. the amount of time used in term scheduling)
on. manual scheduling' of‘ part-time student ‘workers in a
university residence hall foodservice operation, (2) construct
an expert system.demonstration prototype for scheduling part-
time student workers in a. university residence hall
foodservice operation to demonstrate feasibility of use of
expert systems in foodservice management, (3) develop a method
for evaluation of the expert system, and (4) evaluate the
demonstration prototype of the expert system in a laboratory
environment.

Discussion of methods used in this study was divided into
two parts: (1) collection of time data on manual scheduling
of part-time student workers and ( 2) developing a
demonstration prototype of an expert system for scheduling
part-time student workers in a residence hall foodservice
operation at Michigan State University (MSU). The first part
was conducted.by the researcher'during Spring Term 1989, Fall
Term 1989, and Winter Term 1990. The second part was conducted
using a team approach: the researcher and a knowledge engineer
worked together to lead to complete the demonstration

prototype.

41

' 42
3.1 Collection of Time Data on Manual Scheduling of Part-time

Student Workers

The main objective of collection of time data on manual
scheduling of part-time student workers was to determine the
amount of supervisor time needed for scheduling student
workers each term (4 terms per year) at Michigan State
University (MSU) and then use these data to evaluate the
completed expert system. Based on the study of Block et al.
(1985) and Olsen and Meyer (1987), collection of time data
on manual scheduling of part-time student workers in the
residence hall foodservice operation at MSU was divided into
four steps: (1) developing the worksheet for collection of
time data, (2) determining the sampling frame, (3) conducting

the collection of time data, and (4) analyzing the data.

3.1.1 Developing the Worhsheet fog Collection ef Time Data

The purpose of the worksheet was to enable personnel
supervisors or their’ delegates to record time spent in
scheduling part-time student workers.

As discussed in section 2.2.1, work measurement started
with identifying work function activities and then classifying
them into elements that were smallest units and able to
facilitate time measurement (Block et al, 1985). Therefore,
to develop the worksheet for collection of time data on manual

scheduling of part-time student workers, the food manager and

43

personnel supervisor at Brody Complex, MSU, were contacted
and given brief descriptions of the purpose and methods of
this study. A follow-up meeting with the food manager and
personnel supervisor at Brody Complex was then scheduled by
the researcher to identify scheduling functions (term, final,
and volunteer scheduling), elements of each scheduling
function, and descriptions of the elements. A rough list of
scheduling functions, elements, and descriptions was then
formed and checked by the personnel supervisor at Brody
Complex and other two experts, personnel supervisors at Mason-
Abbot and Shaw Hall, to make sure each element was distinct,
describable, and measurable.

Finally, the worksheet (Table 4) and coding guides (Table
5, 6, and 7) were formulated for collection of time data on
manual scheduling of part-time student workers. Each
scheduling function was assigned a two-digit code number, as
shown in Table 4. As indicated in Tables 5, 6, and 7, each
element was assigned a four-digit code number: the first two
digits indicated the scheduling function and. the last two
digits indicated the element within the specific scheduling

function.

. te ' the Sam 'n am
Three residence halls were chosen among all of the
residence halls at Michigan State University (MSU) , East

Lansing, MI, based on the number of students staying in each

44

Table 4. WOrksheet for collection of time data on manual
scheduling of part-time student workers

SUPERVISOR TIME REQUIRED TO SCHEDULE STUDENT WORKERS

 

 

 

Hall Term Year

Code Number and Supervisor Levels Scheduling Function
I Full-Time Personnel Supervisor 01 Term Scheduling
II Part—Time Student Supervisor 02 Final Scheduling

III Part-Time Student Worker as a 03 Volunteer Scheduling

Dishroom Supervisor
IV Part-Time Student worker as a
Student Secretary

 

Date Supervisor Time Scheduling Element Description!
(code) Start/End Function (code) (code)
(code)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Notes:

45

Table 5. Code Number and Elements for Term Scheduling

 

Code Number

Element

Description

 

0101

Organize to
Ekmehfle

 

of student workers and so
them into pre-define
groups (e.g. lunch,
breakfast, and dinner
groups; returning and non-1
returning groups).

Check the job type andk
number of work shifts: add,
delete, or remain the same.
Check the in/out time of
each work shift.

Other (please specify).

Go through all application?

 

0102

Schedule

Match student workers wit
available work shifts b
considering their past
experience and performance,
job, work time, and work
day preferences, number of
work hours desired, in/out
time of available work
shifts , alternate weekend,
and class schedules.
Other (please specify).

 

0103

Check Schedule

Make sure critical jobs are
full.
Count the number of open
and filled work shifts.
Look for student workers
who can*work.extra to fill
open work shifts.

Other (please specify)

 

 

0104

 

Copy Schedule

 

Copy the schedule for use
of student worker.
Copy the schedule for other
purposes.
Prepare and send letters
to student workers.

Other (please specify).

 

 

Table 5 (cont'd.).

46

_‘

 

Supervisors file completedl

 

 

 

 

0105 File Schedule 1.
student schedules.
2. Other (please specify).
0106 Reschedule 1. 'Re-do the scheduling when
student workers are unable
to complete their work
shifts as scheduled.due to
illness, bad performance,
drops and adds of classes,
or other conflicts.
2. Other (please specify).
0107 Delegation 1. Supervisors delegate thJ
scheduling to other people.
2. Other (please specify).
0108 Other (please specify)

 

 

 

 

O

47

Table 6. Code Number and Elements for Final Scheduling

 

Code Number

Element

Description

 

0201

Organize to
Sdmfluhe

Collect the final exa

schedule sheet filled on:
by student workers.

Check. the job type an
number of work shifts: add,
delete, or remain the same.
Check the in/out time of
each work shift.

Other (please specify).

 

0202

Schedule

Match student workers with)
available work shifts bm
considering their present
position, performance,
job, work time, and work
day preferences, in/out
time of available work
shifts, and exam schedules.
Other (please specify).

 

0203

Check Schedule

Make sure critical jobs arel
full.
Check if there are three
work shifts assigned to
each student worker.
Count the number of open
and filled work shifts.
Look for student workers
who can work extra to fill
open work shifts.

Other (please specify)

 

 

0204

 

Copy Schedule

 

Copy the schedule for use
of student worker.
Copy the schedule for other
purposes.
Prepare and send letters
to student workers.

Other (please specify).

 

 

'nflfle£5(omm&d.).

48

J

 

Supervisors file completetJ

 

 

 

 

0205 File Schedule 1.
student schedules.
2. Other (please specify).
0206 Reschedule 1. Re-do the scheduling when
student workers are unable:
to complete their worm
shifts as scheduled due to
illness, bad performance,
or other conflicts.
2. Other (please specify).
0207 Delegation l. Supervisors delegate the:
scheduling to other people.
2. Other (please specify).
0208 Other (please specify)

 

 

 

 

49

Table 7. Code Number and Elements for Volunteer Scheduling

 

Code Number

Element

- Description

 

 

 

 

 

 

 

 

0301 Organize to 1. Determine the job type and
number of work shifts for
SChedUle holiday or special dinner.

2. Check the in/out time of
each work shift.

3. Post the master schedule
determined on board and ask
student workers to sign up.

4. Other (please spedify).

0302 Schedule 1. Supervisors adjust and
complete the master
schedule.

2. Other (please specify).

0303 Check Schedule 1. Make sure critical jobs are
full. »

2. Count the number of open
and filled work shifts.

3 . Look for student workers
who can work extra to fill
open work shifts.

4. Other (please specify)

0304 Copy Schedule 1. Copy the schedule for use
of student worker.

2 . Copy the schedule for other
purposes.

3 . Prepare and send letters
to student workers.

4. Other (please specify).

0305 File Schedule 1. Supervisors file completed
student schedules.

2. Other (please specify).

 

 

Table 7 (oont'd.).

50

 

 

 

 

«0306 Reschedule 1. Re-do the scheduling when
student workers are unable
to complete their worm
shifts as scheduled due to
illness or other conflicts.

2. Other (please specify).
0307 Delegation 1. Supervisors delegate the
scheduling to other people.
2. Other (please specify).
0308 Other (please specify)

 

 

 

 

51
hall. The jpremise ‘was that. scheduling' part-time student
workers in residence halls of various sizes could experience
different problems. The three residence halls chosen were:
Brody Complex, the largest with 2640 students and hiring an
average of 288.33 student workers (N=3: Spring 1989, Fall
1989, and Winter 1990): Shaw Hall, middle with 920 students
and hiring an average of 125 student workers (N=3; Spring
1989, Fall 1989, and Winter 1990): and Mason-Abbot Hall, the
smallest with 750 students and hiring an average of 79.33
student workers (N=3: Spring 1989, Fall 1989, and Winter

1990).

3 . o C

Three separate meetings with personnel supervisors in
each of three halls (one meeting/hall) were held to explain
how the worksheet was to be used. First, scheduling work
functions and elements in each scheduling work function were
explained.by the researcher; Then, personnel supervisors were
shown how to use the coding guides with 24 elements (Tables
5, 6, and 7). Finally, a demonstration of how to fill out the
worksheet (e.g. recording date, supervisor levels, the time
of starting and the time of stopping scheduling) was given.
In summary, personnel supervisors were asked to record the
time of starting and ending scheduling when they performed a

specific element of each of three scheduling functions.

- 52
Weekly visits with personnel supervisors at the three
residence halls were set up to collect the worksheets (Table
4). The collection of time data was completed during Spring
Term 1989, Fall Term 1989, and Winter Term 1990 to obtain an
estimate of the amount of time spent in scheduling student
workers for one school year (most residence halls at MSU were

closed to students during Summer Term).

31114.83a1rzing_the_nata
Summarized data to be obtained from the worksheets were
the amount of time spent in a specific element of one
scheduling function per term. Then, the amount of time needed
for each scheduling function.per term (hours), the total time
needed for scheduling student workers per term (hours), and
the mean time needed for scheduling one student worker per

term (min) were calculated.

3.2 Developing a Demonstration Prototype of an Expert System
for Scheduling Part-Time Student Workers in a Residence Hall

Foodservice Operation at Michigan State University (MSU)

e ' o l Sc ed
- ' u e t Wo s
The residence hall foodservice at Michigan State
University employed a large number of student workers on a

part-time basis. Scheduling these student workers was

53

characterized by the following features.

1.

The scheduling of part-time student workers in
residence hall foodservice operations (SPRF) had
student constraints and job constraints. Student
constraints included class and activity schedules,
job preference (including position, time, and day
preference), experience, number of work hours
desired per week, and alternate weekend desired
(weekend A or weekend 8). Job constraints included
job priority (critical jobs were on the top
priority), in/out time of work shifts, and meal

priority (lunch was scheduled first, then breakfast,

” and finally dinner).

SPRF dealt.with a large number of students and work
shifts. Because of class conflicts, most of the
student workers could not work continuously for 12
hours. Therefore, each job (12 hours/job) must be
broken into three shifts (breakfast, lunch, dinner)
during the day. Also, on the average, student
workers desired 12-15 hours per week. Thus, to fill
all work shifts, the supervisor must employ a large
number of student workers.

A master schedule was constructed each term. Each
term, new student workers must be integrated into
the scheduling; new class schedules of retained

student workers must also be accommodated.

54

4. Since final exam schedules were different from
regular class schedules, all student workers needed
to be rescheduled for the final week each term.

5. Volunteer scheduling was needed for holidays (e.g.
Mother's day, Easter) and for work periods between
terms when student workers would not normally be
available to work. The work shifts needed to be fill
weremdetermined.by'personnel supervisors and.posted
on a bulletin board. Student workers were asked to
sign up if they were available at the time. Then
personnel supervisors adjusted and completed the
schedule.

Based on the features described above, much time would
seem to be needed to implement the sophisticated matching
between the large quantity of student workers and the large
number of work shifts. Further, scheduling part-time student.
workers seemed to be a problem of routine design. Updating
student.constraints (e.g., class schedules, adding new*student
workers) and.job‘constraints (e.g., changing in/out times, job
priorities, etc.), rather than major changes in reasoning or
decision-making processes, was all that was required each
term. Therefore, personnel supervisor time could be saved if
this type of routine design could be performed by a computer

program.

A _.

 

3.. 'o e C oC owre

When personnel supervisors did SPRF (an abbreviation of
"scheduling of part-time student workers in the university
residence hall foodservice operation) , they usually broke this
problem into a set of tasks (e.g. decomposing term scheduling
into weekday and weekend scheduling), worked on each task in
a predetermined order (e.g. scheduling weekday first, then
weekend), and then combined the schedules generated from each
task into the master schedule (e.g. combining weekday and
weekend scheduling to form the term schedule) (DHFS, 1988).

SPRF could be thought of as a problem solving process
that tried to satisfy multiple constraints resulting in a goal
state. For instance, one of the goal states was that there was
no conflict between student workers' class schedules and work
shifts scheduled. To meet this goal, personnel supervisors
needed to check students' class schedules and in/out time of
work shifts and saw if they had a class during, right before,
or right after the time of the shift being considered.

The scheduling process may be repeated several times
because all constraints could not be satisfied on the first
attempt. The process of matching student workers and available
jobs may be repeated by only considering one or a small subset
of constraints instead of all constraints at one time. For
example, one student worker had experience as a exit host

(HE), the lowest priority job. According to the experience

56

constraint (scheduling student workers to the jobs they had
done before), this student should be scheduled to the HE. At
the same time, the job priority constraint that student
workers should be scheduled into critical jobs first must also
be considered. In such a situation, this student worker could
not meet those two constraints at the first attempt since HE
was the lowest priority job. Therefore, only one constraint
could be considered in the mind of the personnel supervisor
at that time, instead of both constraints (experience and job
priority). Usually the personnel supervisor would schedule
this student worker into one of critical jobs rather than the
job (HE) he/she had previous experience on, because filling
all critical jobs was more important than placing student
workers in jobs for which they had previous experience.

Therefore, to meet the needs of the scheduling task, a
programming language should have the following features: (1)
the ability to decompose a problem (scheduling) into tasks
(e.g. term scheduling) and then combine them into a complete
schedule, (2) the ability to consider'multiple.constraints in
an effective manner (be able to reduce the time for
scheduling), and (3) the ability to mimic personnel
supervisors' (expert) reasoning processes, such as making an
alternate plan if the original one could not satisfy all
constraints.

Conventional software, to date, has not been equipped

with. the features described above. The strong’ point. of

- 57

conventional software has been the ability to manipulate data.
That is, it could perform complex computations very fast: it
could use highly complex algorithms to sort through
information (words, numbers, data, or patterns) in thousands
of database records and then read them to determine the final
answer. Since the SPRF problem was not amenable to an
algorithmic solution, conventional programming could be used
to solve this problem, but the programming process would be
complicated and time-consuming (Rauch-Hindin, 1988).

The primary purpose of expert systems was to perform
tasks that only experts in the given domain could do. To
achieve this purpose, expert systems were equipped with the
ability to mimic the reasoning process of human experts. For
example, information processing in expert systems was
performed in a way that implied reasoning, inferring, and
meaning as done by human experts. Therefore, an expert system
may not only understand the concepts of ”car," "start," and
"gas," but also understand the relationship among them. Then
it could infer "if a car will not start, it may be out of gas"
(Rauch-Hindin, 1988). In addition, they represented
information in a structured way that showed the relationships
among each piece of information, which could facilitate the
decomposition of a problem and make data processing more

efficient (e.g., reducing the time spent in processing

unrelated information; Rauch-Hindin, 1988).

 

58
Based on the above, expert systems seemed to be more

qualified for computerized SPRF than conventional software.

1° - t.s__' o, -,f. -. ,. -:;F POu:_!

Based on the characteristics of the SPRF problem (Section
3.2.1) , a set of four features of expert system building tools
for the SPRF domain were determined (Section 3.2.2). These
features were problem decomposition, structured organization
of tasks, constraint satisfaction in an effective manner, and
failure handling (e.g. proposing alternate plans that would
be proposed if the original plan was not satisfied; DHFS,
.1988; Brown and Chandrasekaran, 1989).

As discussed in Section 2.3.3, DSPL.provided a hierarchy
that could facilitate problem‘decomposition. In addition, the
hierarchy of DSPL‘was organized by a top-down control regime.
Thus, all tasks. could be linked together -by a desired
sequence, which in turn would be able to reduce the search
space and increase the speed of problem solving. Additionally,
DSPL.provided the functions of failure handling and redesign.
That is, if the problem could not be solved on the first
attempt, the process of problem solving would be undertaken
again by adjusting the original plan used in the first
attempt.

Thus, DSPL was chosen as the system-building tool for the

development of a SPRF expert system demonstration prototype.

59
3. 4 e v te
W

The process of building an expert system usually has
involved interaction between the expert-system builder,
knowledge engineer, and one or more human experts in some
problem area. That is, the knowledge engineer has transferred
knowledge from the human expert to a computer program and made
the computer program solve problems in much the same manner
as did the human expert.

However, an intermediary between the knowledge engineer
and human expert (personnel supervisor) was needed by the
present study. The human expert (personnel supervisor) was not
available to work with the knowledge engineer for the entire
time needed for expert system development for two reasons.
First, the personnel supervisor was an entry-level position;
people in this position were often promoted to higher
positions after they have been in the position for approximate
6 months. Second, the turnover rate was high in university
foodservice even at the supervisor level. Therefore, the
researcher acted as an intermediary in this study.

Responsibilities of the researcher included gaining
expertise on scheduling part-time student workers,
communicating expertise gained to the knowledge engineer, and
testing and evaluation of the SPRF expert system.
Responsibilities of the knowledge engineer included knowledge

representation and implementation of the SPRF expert system.

60

Based on the discussion in Sections 3.2.1, 3.2.2, and
3.2.3, the objectives for the development of a SPRF expert
system.demonstration prototype in DSPL*were: (1) to construct
a schedule of the same quality as the one constructed by the
personnel supervisor at Brody Complex currently handling the
task, ( 2) to greatly reduce time needed for scheduling student
workers each term, and (3) to allow for easy modification of
the master schedule so that the new schedule could be
constructed easily each term by only updating student
constraints and job constraints (e.g. student class schedules,
job priority, job type).

The process of developing a SPRF expert system
demonstration prototype in DSPL (Figure 3), based on the study
of Buchanan et a1. (Section 2.4) , was divided into four
phases: (1) identifying the scope of the SPRF problem, (2)
knowledge acquisition, (3) implementation of the SPRF expert
system, and (4) testing and evaluation of the SPRF expert
system. The knowledge acquisition phase was to gain expertise
on scheduling part-time student workers as done by the
personnel supervisor at Brody Complex, MSU. The implementation
phase was to turn the formalized knowledge into a working
computer program. The testing and evaluation phase was to
evaluate the SPRF expert system and revise it as necessary.
Recycling through the various steps in the knowledge
acquisition phase was needed to validate knowledge elicited

from the personnel supervisor at Brody Complex and design

Identifying the Scope of

the SPRF Problem

 

 
  
     
     
   
   
 
  
   
 
  
    

  

Knowledge Acquisition

 

 

' ’| Interviewing the Expert

0

Analyzing Knowledge
Elicited from Interview

W

Representing Knowledge

 

 

 

 

 

  

Implementation of the
SPRF Expert System Prototype

  

Testing and Evaluation of the
SPRF Expert System Prototype

   
    

  

Fully Developed SPRF
Expert System Prototype

Figure 3. The process of developing a SPRF expert system

demonstration prototype in DSPL.

. 62
concepts. Also, recycling, through the knowledge acquisition
phase, implementation phase, and testing and evaluation
phases, was needed to reformulate the concepts, refine
strategies used in problem solving, and revise the control
flow of problem solving until the objectives for the
development of the SPRF expert system (discussed in the
beginning of Section 3.2.4) could be achieved. In short,
results obtained from the testing and evaluation phase would
contribute to the improvements of the current knowledge
acquisition and implementation phases. After improvements were
implemented in the current knowledge acquisition and
implementation phases, the testing and evaluation phase would
be repeated to validate changes made and evaluate the

performance of the currently improved SPRF expert system.

de ' o o t R b

As may be recalled from section 3.2.2, expert systems
seemed to be more appropriate for computerized SPRF than
conventional software. The problem targeted for expert system
development should be of a manageable size that made the
developed knowledge base interesting and also warranted _ the
effort of developing the SPRF expert system (Section 2.4.1).
Finally, the problem targeted for the SPRF demonstration
prototype development should not require knowledge from a very

large number of areas (Section 2.4.1).

63

The SPRF problem at MSU, taken as a whole, seemed not to
be appropriate for expert system development for two reasons.
First, the SPRF problem at MSU included both cafeteria and
kitchen areas. Because those two areas were different in job
types, in/out time of work shifts, etc., working on an expert
system design for both areas at the same time would have made
the problem scope unmanageable in size. Second, the method of
performing SPRF varied slightly with personnel supervisors in
different residence halls. Attempts to aggregate knowledge
across different residence halls would have been certain to
mask the true structure of scheduling methods.

To keep the scope from being beyond the acceptable limits
of expert systems but still sufficiently broad to ensure that
the software had practical application, scheduling part-time
student workers for one area (cafeteria area) at one Hall
(Brody Complex) at a time was considered for initial system
development.

From this point of view, scheduling part-time student
workers at the cafeteria area of Brody Complex was chosen as
the domain for developing the SPRF expert system. The reason
was that Brody Complex was the largest hall at Michigan State
University and hired a greatest number of student workers (an
average of 288.33 i 63.55 student workers; N=3, Spring 1989,
Fall 1989, and.Winter 1990) of all MSU residence halls to fill
approximate 1026 work shifts for the weekly schedule. The

large number of jobs and student workers involved would make

64
the SPRF expert system development more interesting and
warrant the effort. Also, if this system was successful, it

could be modified for use in smaller residence halls at MSU.

.2 4. ow ed c s 'o

The major purpose of knowledge acquisition in this study
was to gain expertise on scheduling part-time student workers
and then to use that knowledge as a basis to construct an
expert system demonstration prototype. Since one of the
objectives for the SPRF expert system development is to
construct a schedUle of the same quality as the, one
constructed by the personnel supervisor at Brody Complex, the
SPRF expert system was developed to mimic the problem—solving
process of the expert scheduler. Also, as discussed in section
3.2.4.1, the scheduling in the cafeteria area of Brody Complex
was chosen as the domain for developing the SPRF expert system
demonstration prototype. Therefore, the expert scheduler (the
personnel supervisor at Brody Complex) was used as the
resource for gaining expertise on scheduling part-time student
workers.

The knowledge acquisition process of this study was
divided into three steps: (1) interviewing the expert (the
personnel supervisor at Brody Complex), (2) analyzing
knowledge elicited from interview, and (3) representing

knowledge.

65

Interviewing the expert, was undertaken to acquire
knowledge about scheduling part-time student workers through
direct interaction with the personnel supervisor at Brody
Complex. An intuitive interview technique (discussed in
Section 2.4.2.1) was used in this phase. The second phase,
analyzing’ knowledge elicited, was conducted. to 'map that
knowledge into a coherent organizational structure. Finally,
the mapped knowledge was translated into a formal
representation suggested by the expert system building tool
(DSPL) and refined until it replicated important and valuable
parts of the knowledge of the personnel supervisor at Brody

Complex.

4. I e ' 'n e t. The, approach,
combining intuitive interview and on-site observation_methods
(discussed in 2.4.2.1), was used in this.phase. The researcher’
talked with the personnel supervisor at Brody Complex to
become familiar with the method used to schedule part-time
student workers. During this interaction, the personnel
supervisor introspected while solving a problem in front of
the researcher. The researcher jumped in whenever it seemed
appropriate, asking relevant questions to verify his basic
scheduling method. The personnel supervisor was also asked to
provide a step by step list of his scheduling process. At
three follow-up meetings (one hour/meeting), the researcher

and knowledge engineer questioned the personnel supervisor

66
about his scheduling procedure to gain more detail about the

method for scheduling part-time student workers.

3 .4. w c' ' te iew.

The purpose of this phase was to map the obtained
knowledge from the personnel supervisor into an interpretative
framework. Four tasks, derived from the study of Breuker and
Wielinga (1987), were involved: (1) identifying knowledge, (2)
conceptualizing knowledge, (3) constructing the reasoning
process of scheduling, and (4) determining the structure of
the scheduling process.

a. Identifying knowledge

Knowledge identification was mainly an information
management task. A large amount of knowledge was gathered from
the interview process, but the relative importance of each
piece of knowledge was yet to be determined. Identifying
knowledge provided an important reduction in the amount of
knowledge through organizing it by knowledge type, such as
classification, prioritization, and design. Classification
referred to knowledge used to break a group of jobs or student
workers into subgroups (e.g. knowledge for sorting student
workers into lunch, breakfast, and dinner). Prioritization
referred to knowledge used to rank jobs and student workers
(e.g. grouping jobs according to job priority). Design
referred to knowledge used implicitly by the scheduler in the

scheduling process, such as decision-making strategies (e.g.

. 67

scheduling experienced student workers first; filling in
critical jobs first).

b. Conceptualizing knowledge

The purpose of knowledge conceptualization was to
integrate individual concepts into a conceptual framework
representing the logical combination of tasks. For example,
breakfast, lunch, and dinner scheduling (individual concepts)
were integrated into weekday and weekend scheduling; weekday
and weekend scheduling (individual concepts) were integrated
into term scheduling. The result was that the concept of term
scheduling was finally constructed.

c. Constructing the reasoning process of scheduling

The construction of a reasoning process was to organize
reasoning strategies used by the personnel supervisor in a.
structured manner. For example, lunches were the hardest to
staff so they were scheduled first, followed by breakfast and
dinner. Weekend scheduling was done last because it was easy
to staff.

d. Determining the structure of the scheduling process

Structural determination of the scheduling process was
done to determine the scheduling structure by mapping the
reasoning process into a conceptual framework, This structure
facilitated interpreting the data and getting a fine-grained

description of the architecture of the SPRF domain.

68
s ow The knowledge
representation. was developed. by the knowledge engineer.
Appendix 8, written by the knowledge engineer, shows how the
knowledge was presented in the way suggested by DSPL.

In summary, DSPL provided two useful knowledge
representation facilities, agents (e.g. specialist, plan) and
a design database. Domain knowledge was encoded both in the
agent structure (inference structure) and the design database.
The procedural knowledge (e.g. scheduling method) was
reflected in the agent structure; while the declarative
knowledge, used in agent structure (e.g. job priority, in/out
time of work shifts), was represented in the design database.
The design database also provided the structures for storing
the actual schedules that needed to be filled. Agents (e.g.
specialist, plan) were organized into hierarchies in DSPL to
break down the scheduling problem.

Major functions had a specialist associated with them
along with their sub-specialists. The control regime for the
DSPL hierarchy was top-down. The following was done
recursively until a complete design ‘was worked out: a
specialist corresponding to a component (e.g. weekday
scheduling) of the task (e.g. term scheduling) was called; the
specialist (e.g. week specialist) in turn suggested further
sub-specialists (e.g. lunch sub-specialist) to call to set

other details of the design.

 

lc .1

69

3 - - -stino .,. -v.” .t,-; or ,- S’i, _ao-, stem

To evaluate the SPRF expert system, objectives with
criteria plus two evaluation approaches were developed, but
were not used in the present study due to time constraints.
Since expert systems needed to be refined and tested in a
laboratory environment before they could be released for field
testing, two evaluation approaches were preposed: laboratory

testing and field testing (user environment).

3. . .3 va ua 'te ‘ . As indicated in the
beginning of Section 3.2.4.2, the SPRF expert system was
developed to mimic the problem-solving process of the expert
scheduler. Therefore, criteria used to evaluate manual
scheduling of part-time student workers were also used to
evaluate the SPRF expert system. Additional criteria used for
evaluating expert systems were derived from the works of Boehm
et a1. (1978), Gaschnig et al. (1983), and Waterman (1986).
These criteria were to evaluated expert systems in terms of
their completeness, consistency, accuracy, ease of use, and
the ability to update.

Finally, criteria for evaluating the SPRF expert system
were formulated by combining criteria for evaluating manual
scheduling' of part-time student. workers and SPRF' expert

systems (Table 8).

 

 

 

 

 

 

 

 

.mumxuo3
ucmccum an omaaflm mum muwficm xuoz comma Hag m.m
.umumm
no muOMon mamumflnoEEH mommmHo m>mn >mcu noses oEHu comm mumxuos
new mavenm men cm>flm was one muoxuo3 ucmcsum ~.m ucmmsum mEHucuqu mafiascmcom mo
.muwxuo3 acopsum an poaaflu mum mnofl mauoufluo Add H.m mucoswuflsmmu on» Dome OD mans on 09 .m
.eofisemnom was
new mafia xuo3 men can cm>flm mumwcm mo women:
0:» cues pneumaumm mum mumxu03 ucmnsum mo wow m.v A
onwcmflmmm mumxuog UCQUSUW
non mnu mcfloo mo manmmmo mum mumxuoz aconsum H.v mo cofiuomwmflumm we» oNeEquE 09. .v
.Ammmooum mcaasuonom on» no new as muwfinm mumxuOB uconsum manuaam>m
ammo mo wanes: ummoav nodawu mum mumacm Had H.m om: xaucwHmewm ou mans on QB .m
N .umufim mumxuos acousum 8053098 32.6028 TN
.pocmflmmm
when one macaw: on mane mum muoxuoz bemosum m.~
.uoquB
ucmnsum mom mmmxu non mo menace mcfiNflEHCHz ~.N
.mmmoosm nuH3 poEuOMuwm
xamsofi>mum m>ms >mcu noflsz ca moon once mumxuos muoxuoz unopsum mcflcflmuu new
ucmpsum mascocom ou xuu .manfimmom mm cmumo mg H.N mafia momfl>ummsm may oNHEHCHE 08 .m
.mcwasnocom ummucsao> new .Hmcflm .Eumu EuOmumm
o... 083 “533096 mo bemoan .5555 one mm: HA was mcflasvwnom undead: 8. .H
iamumxm uummxw mama new maddsconom Hmncmz_.H
EH6 EH8
.coflumummo m0e>umm©00u Ham: mocmcflmou xufimum>flcs we» ca mumxuos ucmcsum weeuuuumm mo
mceqacozom Emumxm uummxm mmmm can HmscmE Hmeflumo mo c0«ums~m>m on» now mfiumufluu .m manna

m no a moms

 

71

 

 

.mcfiasnmcomtmu cam amasumcom
xmom new .>QOO .xmmcm “mafiasnmcmm vcmxmmz cam
xmpxmm3 imceflsmmcom ummucsao> cam .Hmcflm

4 Emu “mmmnmumm mascmcom Vmumamfioo .mmmnmumn
DOM .mmmnmumc ucmmsum mcamcmnm umHm>HuomHmm
chAuocsm mcHBOHHOM mSUAEHOMumm on manm mm
.mumc mom can sump ucmnsum mo usage
was monogamomm on mean we zones; smammun one

.mm: mmpcmucw muw >umwumm 09 mmwo

umum xaucmHOHuwsm we mascmnom cmumameoo mas
.mumanpmcom uummxm mo mommEGOMDmQ

mcu mo manna manmsam> use uncuuomEH mumoflammu

00 Emumxm ms» msoaam mmmoonm umnfimnem msB

.meump mo Hm>mH pomuuoo mnu um nmucmmmum

mam nmumcuo cam umnwcmmuo >~mumflumoummm mum
ummwauum cmammn men mcwuoum Haw musuocuum may
Em Ammwmmumuum mchomlcmHnoum .m.mv mmpmdficx
.Amumxu03 ucmusum

so confine mum anon Hmoauuuo Han “mommmao unmesum
new mmcmflmmm muwezm cmm3umn uoflawcoo 0c .m.mv
mumqus ucmcsum going—mm 05.3828 «0 3cm...
umuflsvmu mnu mummE mxmzam masumnom nmumameoo mse
.Am manna ca mmucoflccav mascmnomnmu new .masnmzmm
mad“ .xmoo .xomno .masmmmmm .masnmnom on mmacmmuo
"muomfl>ummsm amccomumm an moon mcwasnmcom

CH mucmemam non mew mo Ham Esomumm ou mans mm
.Emumoum umusmsoo menu oucw omumuom

luooca mum mcwasmmnom ca um>aoscH Amcfiasumcmm
ems: cmumcwmcoo mum3 mocmummmum mom can mocmfiummxm
m50H>mum .m.mv mmfimmumuum mcfl>aomusmanoum HmoHufluU

N.h

m.m

m.m

¢.o

m.m

N.m

H.w

mm: o» >mmm mam wapcmflum umm: me 99 .n

.mHMusomm 0cm .ucmumflmcoo .mumam
LEGO we Emumxm uummxm mmmm mnu ca
cmcpmnem mmcm~3ocx map mucm mme 09

.umumsm unomxm mmmm .HH

.m

 

 

 

m>HBUQHmO

 

 

m «o N mmmm

Acouscnucoov

.m manmb

 

.coflumummo mogummnOOM 3m: mocmgmmu >33ng m5 cw mumxuoz ucmvsum 93......»qu mcflaspmnum "mmmm ..

 

. «mongowumm ummm ucmcsum .m.mv mumumEmumm
mace 3m: maflmmom magma. MOM musuoaum
when an suflaflnaxmam onu oefi>oum on mama mm m.m
.usmcw 3m: :0 Ummmn
mmmcmcu uooamou on banana one segues on menu mm m.m
.umms m5
sn ssmmmo mucflmuumcoo umnuo cam .mpuanm no mass
usoxcm .sufiuoaum non .mmsu non omcmco 0» menu on H.m

.ucmmmum ma COHDQEOME
m>Hmmmoxm 0: “mm: 9.8 comm 0» >93 mum Bauso 3H mé
.xmoHocfiEumu :30 m.umms m5 5
mmmmmmme nouum Hummus 6cm ummao umHQmMU B mans mm mK
.pmcmmc mm min
3mm ucmuSUm .mmasnmcom vmumamcbo .m.mv muauso
mo ucmucoo mam Show me... mSuEmumc ou mans mm 5.5
.mEHu mEmm m5 um pmuomuuoo me 0... 303m cam hamumo
Lump—EH mm3m5 mace momuuomsh ummump o» mans mm mfi
. $83:ng ES
mmes m5 ca cmmmmumxm cam pmcgoum we acme umm: < m4.
. umms m5 Cu Humvee/0mm we
umfi c0305.“ comm uOm cm>flm we :oflumwuommp mmflun m e4.
SQ; pmusuogum m 5
umuwcmmuo mum umm: m5. uOu ©m©fi>oum mcoflumcsm H2 mK

72

.saammo common: on on mane on as .m

.mm: 3 ammm mam 32min umms mm 09 .n

 

(Hg

 

 

 

Eng

 

m MO m mmmm

inauscuucouc .w mange

CC

 

of

0t
a1:
ev

Ta

to

Bag

73
. 3. ' The two evaluation
approaches for the SPRF expert system were: laboratory
evaluation and field evaluation.

A check-list approach was developed for the laboratory
environment evaluation. The researcher examined test results
according to the questions indicated on the check-list. The
check-list (Table 9) used in this study was based on Table 8
(Criteria 1.1, 3.1, 5.17 5.3, 6.1, 6.4, 6.5, and 7.1-7.9).

For field testing, the suggested evaluation approach
consisted of three parts: (1) evaluating the system in terms
of its completeness, consistency, and accuracy (criteria 5.1-
5.3: 6.1-6.6 in Table 8), efficiency (criteria 3.1 in Table
8), ease of use (satisfaction of personnel supervisors and
other users: criteria 2.1-2.4: 7.1—7.9 in Table 8), and
ability to update (criteria 8.1-8.3 in Table 8), (2)
evaluating the system in terms of its speed (Criteria 1.1 in
Table 8), and (3) evaluating the satisfaction of student
workers with schedules given (criteria 4.1-4.2 in Table 8).

For the first part, the method would involve responses
from a group of experts (judges) who rate the SPRF expert
system using a questionnaire. Personnel supervisors of MSU who
havemdone scheduling for at least one year will act as judges.
They will be instructed in the use of the SPRF expert system.
After becoming familiar with this system, they will be asked
to work on the same scheduling problem by using this system.

Based on their expertise in manual scheduling, they will then

74

Table 9. A check-list for evaluation of the SPRF‘ expert
system demonstration prototype in a laboratory environmentb

 

Question

 

1. Are problem-solving strategies that are
incorporated into the SPRF expert system
demonstration prototype complete, accurate
and consistent ?

2. Does the SPRF expert system demonstration
prototype solve problems in the natural way
the expert scheduler prefers ?

3. Are the results of scheduling organized
and presented at the right level of
details that satisfy the schedule's
intended use ?

4. Does the SPRF expert system demonstration
prototype help the user in some significant
way ?

a. Greatly reduce the scheduling time

b. Efficient utilization of available
student workers

5. Is the interface friendly and clear for
different groups of users (e.g. personnel
supervisor, student supervisor) ?

 

 

 

: SPRF: Scheduling pert-tine student workers in the university residence hall foodservice operation.
Boehm et el.(1978): Uatennen (1986).

75
evaluate the SPRF expert system. Finally, the responses to the
questionnaire from. each 'test site (each judge) will be
gathered and analyzed statistically.

For the second part of the evaluation approach suggested
for field testing, the worksheet used in collection of time
data on manual scheduling of part-time student workers will
be also used to collect time data on scheduling by the SPRF
expert system. However, to make the worksheet specific for
scheduling by the SPRF expert system, the coding at the
worksheet for manual scheduling (Table 4) and coding guides
(Table 5, 6, 7) should be modified. That is, each element in
the scheduling function will be able to represent scheduling
by the SPRF expert system rather than manual scheduling.

For the third part of the evaluation approach suggested
for field testing, a questionnaire to student workers will be
used. as a survey instrument: students ‘who 'work in the
cafeteria area of residence halls that use the SPRF expert
system will act as the sampling frame. Those students will be
asked to fill out the questionnaire and the responses from
each student will be compiled and analyzed statistically. The
development of this questionnaire will be based on related

criteria 4.1-4.2 in Table 8.

Chapter IV

RESULTS AND DISCUSSION

The purposes of this study were to (1) obtain objective
time data (e.g. the amount of time used in term scheduling)
on manual scheduling of part-time student workers in one
university residence hall foodservice operation, (2) construct
an expert system.demonstration.prototype for scheduling part-
time student workers in one university residence hall
foodservice operation to demonstrate feasibility of use of
expert systems in foodservice management, (3) develop a method
for evaluation of the expert system, and (4) evaluate the
demonstration prototype of the expert system in a laboratory
environment.

Discussion of results below was divided into three parts:
(1) manual method to schedule part-time student workers, (2)
collection of time data on manual scheduling of part-time
student workers, and ( 3) the development of a SPRF ' expert
system demonstration prototype.

Information on the manual method to schedule part-time
student workers was obtained through formal meetings with
three personnel supervisors (discussed in Section 3.1.1) and
weekly visits with three personnel supervisors for the
collection of time data on manual scheduling of part-time
student workers (discussed in Section 3.1.3) . These tasks were

done by the researcher.

76

77

The objective of collection of time data on manual
scheduling of part-time student workers was to determine the
amount of time needed for scheduling student workers per term
and then use the information as a basis to evaluate the SPRF
expert system demonstration prototype. This collection of time
data on manual scheduling of part-time student workers was
done by the researcher. Data obtained from the worksheets for
collection of time data were compiled and analyzed. The amount
of time needed for each scheduling function (term, final, and
volunteer scheduling) per term (hours), total time needed for
scheduling student workers per term (hours), and mean time
needed for scheduling one student worker per term (min) were
calculated for each of three residence halls (Tables 10, 11,
12). Findings and implications suggested are discussed in
section 4.2 below.

The development of a SPRF expert system demonstration
prototype was done using a team approach. The researcher, an
intermediary between the knowledge engineer and.human expert,
gained expertise on scheduling part-time student workers from
the human expert, communicated expertise gained to the
knowledge engineer, and then tested and evaluated SPRF expert
system demonstration prototype. The knowledge engineer
represented knowledge (expertiSe) in the formal way suggested
by DSPL, the expert system building tool, and implemented
formalized knowledge on a working computer program to make it

solve problems in much the same manner as the human expert.

78

Table 10. Number of students and times used to manually
schedule cafeteria part-time student workers at Mason-Abbot
Hall, Michigan State University.

 

 

 

Tern Spring Fell Uinter Mean of three
lten 1989 1989 1990 terms
tuner of Students 75 88 75 79.33 3 6.13
Time Spent in Term 16.65 78.8 17.5 37.65 I 29.10

Scheduling (hrs)

 

Time Spent in Final 7.42 13.25 4.91 8.53 3. 3.49
Scheduling (hrs)

 

line Spent in
Volmteer 1.00 0.00 0.00 0.33 z 0.47
Scheduling (hrs)

 

total Time Needed 25.07 92.05 22.41 46.51 ; 32.22
Per Tern (hrs)

 

 

lean Time Needed 20.06 62.76 17.93 35.12.
Per Student (min)

 

 

 

' (Mean Time Needed Per Yerm (mu/Mean Huber of Students Per Term)
* 60 I 35.12 (min)

79

Table 11. Number of students and times used to manually
schedule cafeteria part-time student workers at Shaw Hall,
Michigan State University.

 

 

 

Term Spring Fell winter Mean of three
Item 1989 1989 1990 Term
luster of Students 120 135 120 125.00 .3 7.07
Time Spent in Tern 49.06 161.50 100.00 103.52 3 45.97

Sehedfling (hrs)

 

tia Spent in Final 12.67 34.5 18.00 21.72 g 9.29
Sdtedning (hrs)

 

tin Spent in
Volmteer 14.25 0.00 2.00 5.42 z 6.30
Schehling (hrs)

 

Total Time Needed 75.98 196.00 120.00 130.66 3; 49.57

 

Per term (hrs)
Mean Time Needed 37.99 87.11 60.00 62.72ll

Per Student (mi n)

 

 

 

 

‘ (lean Time Needed Per Term (hrs)/Nean Number of Students Per Tern)
' 60 a 62.72 (min)

80

Table 12. Number of students and times used to manually
schedule cafeteria part-time student workers at Brody
Complex, Michigan State University.

 

 

 

Tern Spring Fell Ninter Mean of Three

ltea 1989 1989 1990 Terms
I”? of Students 235 200 250 288.33 3 63.55
line Spent in Ter- 46.92 40.30 78.49 55.24 1 16.66

~ Scheduling (hrs)

 

Ties Spent in Final 9.52 8.75 42.50 20.26 3 15.73
Scheduing (hrs)

 

Time Spent in
Volmteer 13.72 0.00 0.00 4.57 3 6.47
Sehechllng (hrs)

 

total Tine Needed 70.16 49.05 120.99 80.07 g 30.19
Per Tera (hrs)

 

Neon 1m Needed 17.91 14.72 29.07. 16.66“
Per Student (min)

 

 

 

 

' (Mean Time Needed Per Tern (hrs)l1lean Nuber of Students Per Tern)
1' 60 8 16.66 (min)

81

The SPRF expert system demonstration prototype has
completed weekday scheduling (vs. weekend, final, and
volunteer scheduling) and.heen evaluated in a laboratory
environment. This suggested that expert systems
technology could effectively be applied to the scheduling
problem, and that the system development would be
beneficial for the university foodservice operation.
Based on the results of testing and evaluation,
suggestions for improving the SPRF expert system
demonstration prototype are given in Section 4. 3. In
addition, criteria and approaches for evaluating the SPRF
expert system demonstration prototype were also developed

(and were presented in Sections 3.2.4.3.1 and 3.2.4.3.2.

4. 9:1, ._ 9’ -1". o .. 1'1. ‘ 'te-T1: .t_-.-1 W0 =

The basic manual method, used to schedule part-time
student workers by personnel supervisors at Brody
Complex, Shaw Hall, and Mason-Abbot Hall, consisted of
term scheduling, final scheduling, and volunteer
scheduling. The term scheduling procedure below was
obtained from written (DHFS, 1988) and oral information
provided by the personnel supervisors at Brody Complex,
Shaw Hall, and Mason-Abbot Hall, which was similar in
part to the written method of MSU's Department of Housing

and Food Service (DHFS, 1988).

. 82
Briefly look through.all student applications,
make mental notes of those with lunch work
shifts open, check for returning and
outstanding student workers, and check specific
jobs that student workers want. The personnel
supervisor at Brody Complex sorted all
applications into lunch, breakfast, and dinner
groups at the same time (Personnel supervisors
at Shaw and Mason-Abbot Hall did not do this
task).
Take all of student applications and begin
filling in jobs. Student workers who turned in
their applications first were scheduled first.
Lunch work shifts were scheduled first, then
breakfast, dinner, and weekend..Also, critical
jobs were filled first.
Schedule student workers.
a. Consider previous positions or experience

held by student workers.,

b. Consider job preference of student
workers.
c. Consider the number of work hours desired

by student workers.
d. Consider work time and day preferred by

student workers.

83
Consider class schedules of student
workers.
Give student workers the work time and day
they prefer unless they have lunch work

shifts open but not mark preferred.

. Give student workers the jobs they prefer.

Otherwise, try to fill in critical jobs
first. The personnel supervisor at Brody
Hall filled critical jobs first and did
not consider student worker preference
(the personnel supervisors at Shaw and
Mason-Abbot Hall did do this).

If the student worker only had certain
time periods open that were early filled
by other student workers, then supervisor
went back through the completed schedules

and made some changes.

If there were still open work shifts during the

week, split any remaining work shift as a last

resort so that two student workers shared one

work shift.

The personnel supervisor filled breakfast work

shifts with student workers who had no class

at 10:20 am (personnel supervisors at Shaw and

Mason-Abbot Hall did not do this).

Double check the completed schedule.

1.

5.

84

final scheduling contained the following steps:

Fill in the last.meal served first and work
backwards (e.g. Friday Dinner, Friday Lunch,
Friday Breakfast, then Thursday Dinner, etc.)
because the last meal during final week is the
hardest to fill.

If possible, schedule student workers to the
jobs they have done during the term (The
personnel supervisor at Brody Hall did not
follow this rule). At the same time, consider
student worker job preferences (The personnel
supervisor at Brody Hall did not do this).
Give a similar amount of work shifts to each
student worker.

Fill in each meal completely before going on
to the next meal.

Double check the completed schedule.

The volunteer scheduling was needed for holidays

(e.gu Mother's Day, Easter) and:for work.periods between

terms when student workers would not normally be

available to work. The work shifts needed were determined

by personnel supervisors and posted on the bulletin

board. Student.workers were asked to sign.up if they were

available at the time. Then personnel supervisors

adjusted and completed the schedule.

85
4 c o n
ef_Eart:Iime_§tudent_flerkers

As indicated in Table 13, the amount of supervisor time
needed for scheduling student workers varied among terms
(Spring Term 1989, Fall Term 1989, and.Winter Term 1990) in
each of three residence halls. For example, the personnel
supervisor at Mason-Abbot Hall spent 25.07 hours scheduling
75 student workers during Spring Term 1989, 92.05 hours
scheduling 88 student workers during Fall Term 1989, and 22.41
hours scheduling 75 student workers during Winter Term 1990.
Also at Mason-Abbot Hall, mean times needed for scheduling one
.student worker Spring Term 1989, Fall Term 1989, and Winter
Term 1990 ranged from 17.93 to 62.76 minutes whereas times
required by supervisors at Shaw Hall ranged from 37.99 to
87.11 minutes, and at Brody complex from 14.72 to 29.04
minutes.

The amount of supervisor time needed for scheduling
student workers did not appear to be related to the number of
student workers scheduled. Time required for scheduling did
not increase directly with number of student workers
scheduled. For example, schedulers at Mason-Abbot Hall, with
an average of 79.33 i 6.13 student workers (N=3: Spring 1989,
Fall 1989, and Winter 1990), spent an average of 46.51 1 32.22
hours in scheduling student workers per term: Shaw Hall with
an average of 125 i 7.07 student workers spent an average of

130.66 i 49.57 hours: and Brody Complex with an average of

 

 

86

 

 

 

 

 

 

 

 

 

365 63:02 6838 066
ewe: 81.8 Mdmm Emission 68 8862 85 89,. 952.
3.8 .... 8.8 H 8.02 8.8 H 3.8 88: 95805.6. 5 688 9.5 H38. moan.
8.8 + 8.88 + 8;...2 2.... + 8.8 39:03 8088 no 5952 do
BE. 69.83 6538 pom
3.8 8.8 8.2 8:828 you 8682 85 :82
8.82 8.8a 3.8 88: 8380.8 5 0:08. 050 H38. 83
8m 03 ms 89.82 0888 no umnfiz .065.
Aria: meug ucmfisum Hg
NS: 3.8 8.8 mcngnosom .38 86062 2:» :82
8.8 863 8.8 $9: 83829.... 5 08% 0.5. ~38. 82
Q8 9: 8 38.8: E088 mo 68.52 :8
:95 .8639: ucmosum mom
8.2 8.8 8.8 8:828 no.0. 666062 85 :82 82
3.8 8.8 8.8 $65 8:828 5 6:68 25 :32.
mmm oma mm mumxuoz ucmosum mo umnEBz mcHumm
x0888 68mm 88 6821.89.
HHMI monomm 609

 

 

inflame/«c: macaw 5.63:3: imam—sou xooum one Jam: 32m Jam: “5821:8302
up mumxuo: ucmosum owumumumm masumsmm advices on ammo omen» one mucmosum mo umncsz

.2 fig

‘ 87
288.33 1 63.55 student workers spent an average of 80.07 i
30.19 hours. -

The results were discussed with personnel supervisors
(schedulers) in three residence halls to determine possible
factors accounting for“the‘variability'of'results. The factors
are listed immediately' below‘ and are then. discussed in
Sections 4.2.1 to 4.2.6.

When the personnel supervisors were in training for their
job, those personnel supervisors had only been given brief,
basic oral instruction on scheduling (e.g. when do the term
and final scheduling need to be started: the requirements of
scheduling, such as scheduling critical job first) and then
required to practice scheduling on a given sample: no written
material that showed how to schedule student workers step by
step had been provided to them. These three personnel
supervisors learned scheduling on the job. Therefore,
scheduling' methods used. by each of the three jpersonnel
supervisor were not expected to be the same. With different
methods used, the time spent in scheduling could be expected
to be different.

In addition, the quantity of returning vs. new student
workers varied. Also, human error effecting the precision of
time data collection on manual scheduling of part-time student
workers was also a possible factor accounting for the

variability of results among halls.

88
4 e ' t d ' e

The manual method of scheduling used at Brody Complex was
not entirely suitable for Shaw or Mason-Abbot Hall because of
the smaller quantity of student workers. There were only a
limited number of student workers at Shaw and Mason-Abbot
Hall: the average number of student workers per term was 125
and 79.33 respectively vs. 228 at Brody Complex (Table 13).
With such a limited number of student workers, personnel
supervisors at Shaw and Mason-Abbot Hall briefly looked
through all the applications, made mental notes of those
student.workers with lunch.open, and checked for returning and
outstanding student workers, instead of physically breaking
all student job applications into lunch” breakfast, and dinner
piles. Also, personnel supervisors at Shaw and Mason-Abbot
Hall were better able to be more concerned with the needs of
student workers (e.g. job preference, work“ time and day
preferences) when scheduling a smaller number of student
workers.

Thus, not only the number of student workers employed,
but also the scheduling method used by personnel supervisors
in different residence halls at MSU accounted for the
variation in the amount of time needed for scheduling student

workers.

89

4 . g-gu,7 . .- .._ - f-_ AmOe- Sc ‘1 - ~ ' One Hall

Student supervisors, assigned by the personnel supervisor
to do some scheduling, may not have had adequate training and
practice to perform the scheduling task using the preferred
method with "normal" amount of time. Thus, scheduler was also
a.factor effecting the amount of time needed. For example, the
student supervisor at Brody Complex, Winter Term 1990,
considered the job preference of student workers when
performing final scheduling. Thus, rather than looking at the
individual final exam student employment schedule sheet
(Figure 4), he also referred to the individual application for
student foodservice employment sheet (Figure 5) which
indicated the job preference of student workers. In this way,
more time*was needed for assigning to student workers the jobs

which they preferred.

4 . The uan t f etu n' vs New u e Wo ke's

The quantity of returning student workers was a factor
which, in part, accounted for the time needed for scheduling
at Brody Complex, Show, and.Mason-Abbot Hall. For example, in
Spring Term 1989, Mason-Abbot Hall had a greater number of
returning student workers: nine out of 10 student workers were
returning student workers.in.Spring'Term 1989, while three out
of 10 student workers were returning student workers in Fall
Term 1989. As indicated in Table 10, the mean time needed to

schedule one student worker was 20.06 minutes in Spring Term

 

90

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 4. Final exam student employment schedule sheet
urcmmu stars umveasm
omsrow or uousmo me soon stances
FINAL mu sruosnr sanctum scueouu
flichlgon State University is an Atilrrnatlse Mon/lust Opportunity institution
trans ' sruoewr no. oats
(uan (First) (111)
caurus anonsss Phone
ta 1’ w Th r
ms . 8:48 ‘7 exam scue'ouu
10.82 MMXMWWIOM'UYWOMS.
12- “. res - All student employees are moheduiod tor osarn week. lrom Friday
° dinner oatoroms through Friday oi exams. As a condition oi
5 WI m .7. required '0 M 09 10 818 "1.318 during "118 W.
a C
YMMW' may 0. changed I. my “I“. by M ”"10"!“
5:4! . 7:48 supervisor and you may be required to wed: in any area.
a - 1o
wont: somatic:
Indicate wort: preiorenoa with a small “a” at the W1. meal.

soon ratoav safilnoav suuoav uonoav meson weowesoav mussoav rnloav

snurrusr 7//

portrait

 

 

 

 

 

 

 

 

 

 

 

At the beginning oi next term. work will be on a voluntary sign-up basis. first some/first serve ior job preterencos. We reserve
the right to change your wont asslgnrnent. The sign-up will be posted during liners week.

Please help us out and sign up tor shiits. Also. please till outs schedule term tor next term and hand it in oeiore leaving campus
tor the areal. We do not guarantee any job term to term.

Day and time leaving campus

Day arriving next term

 

 

snow "118 UN! l8 '0' 03m US! ONLY

 

 

 

 

 

 

 

FIRM Exam WM Schedule
um: ‘00.!”
Final exam weal: begins Friday at dinner. Please be prompt when punching 1n tor work.
3".” FHID‘V S‘WND‘Y—J - SUND’AY MONM' TUESDAY WEDN‘SDAY 7"" “30" FRID‘Y
CNUKFAST %
/
Lunch ////,
prunes l

 

 

 

 

 

 

 

 

 

 

91

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 5 . Application for student foodservice employment
sheet.
MICHIGAN STATE umvensrrv rsnta m m ...
Division oi Housing and Food Services 7 w a. 80 our more.
amcanou FOR sruoem FOOD sermcs sanctum it...
‘ H W TC N a.“
”mmanmwmwm
mum-.5 11 T mun. Motor c1... m.»
we“. SoeSoc. l Tm
”was“. A Md” I I
lmhavoworaodatMSUhatoro.whern entail—Whirl

A ”at: al classes and activities with on X
m 8580'” 10511:” ",1“ ”1, 1m 3w 8:“ m” 7-.

-B——lndicateworir preteroncaewithsmaiiXsotapproprinarnaalainbosbeiow. DO NOTWRITIANYTHING 81.88.

 

 

 

 

mm:mmmmmummwwwswm __ EqualOppornmity
Grun— Wort-Suryl’rogrsm_ Other

WWWlagnmmmJammmmuwwm?

 

 

 

deaahwnwentedeachwealt MMWmmmmA_—. 8...... Both .—

newsman-mm

 

  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

WORK SCHEDULE Weekend
are" H 1' VI 111 F 8! In
em
1 Listen
{
5 Dim-
Narna Campus Address Elioctive ._ Weekend
Shirt M 1' w n. F St Sn
”lost
lunch
Dimer

 

 

 

 

 

 

 

 

 

92

1989, while 62.76 minutes were required in Fall Term 1989.

Because those returning students were familiar with jobs
and shifts possible for them to work, they wrote down the
shifts they were able to work on individual application sheets
in Section C (Figure 5). Normally only supervisors filled out
this section. However, returning student workers, who knew
their supervisor well, completed this task for him whereas new
student workers did not complete Section C (Figure 5). Thus,
when looking at their applications, the personnel supervisor
at Mason-Abbot Hall gave returning student workers the shifts
they requested if those shifts were on the list of top job
priorities. Since the personnel supervisor at Mason-Abbot Hall
did not need to go through student worker class schedules, job
preference, number of work hours desired, etc. as they had to
do for new student workers, much supervisor time was saved.

On the other hand, in Fall Term 1989, Mason-Abbot Hall
employed many new student workers (75 percent of 88 student
workers were new student workers). As they turned in
application sheets, the personnel supervisor at Mason-Abbot
Hall scheduled them right away. Since they turned in
application sheets with only Sections A and B (Figure 5)
completed individually, the personnel supervisor at Mason-
Abbot Hall was required to perform scheduling in Section C
(Figure 5) and, at the same time, communicate with new student
workers individually. Therefore, more supervisor time was

spent in this way.

' 93

4. u'a o ' m a a

In this study, schedulers were asked to record the amount
of time they spent in scheduling on the worksheet (Table 4).
However, schedulers did not always remember to record the time
spent in scheduling immediately after each scheduling session.
When they tried to go back at a later time and recall how much
time they had spent, schedulers could only estimate the time.
Also, schedulers forgot to consider the amount of time taken
by interruptions (e.g. answering the phone, asked to meet’ some
people or help something) during scheduling.

Because the researcher was unable to be with schedulers
100% of the time, scheduler accuracy could not be evaluated.
Based on weekly interaction with schedulers (to collect the
worksheet) in three residence halls, the researcher estimated
that such a human error (e.g. not always remember to record
the time spent in scheduling immediately after each scheduling

session) occurred 40% of the time.

4. .5 0 er 0 3

Personnel supervisors in the three residence halls
usually delegated part of scheduling task to student
supervisors. Time needed for communicating with student
supervisors, frequency of delegation, and the productivity of
student supervisors also accounted for differences in the

amount of time needed for scheduling student workers per term.

94
4. .
Table 14 summarizes the possible reasons accounting for
differences in the amount of time needed for scheduling

student workers per term at three residence halls.

4 e 0 st
emo 'o o t e

The development of . a SPRF expert system demonstration
prototype was done using a team approach. The researcher, an
intermediary between the knowledge engineer and.human expert,
gained expertise on scheduling part-time student workers from
the human expert, communicated expertise gained to the
knowledge‘ engineer, and then tested and evaluated the
demonstration prototype. The knowledge engineer represented
knowledge (expertise) in the formal way suggested by DSPL (the
expert system. building' tool) and implemented. formalized
knowledge on a working computer program to make it solve
problems in much the same manner as the human expert.

The SPRF expert system. demonstration jprototype has
completed weekday scheduling (vs. weekend, final, and
[volunteer scheduling) and been evaluated in a laboratory
environment. This suggested that expert systems technology
could effectively be applied to the scheduling problem, and
that the system development would be beneficial for the
university foodservice operation. The results of evaluation

are discussed in Section 4.3.2.

 

95

.AommH umunHZ
5. mo one. mmmH HHmh nH wo

mHHn3 . mnHHsmmnom ummuHBHqS

3 633089... gnzuocum

nH unmmm mEHu H33 «0 womv

mafia—Genomes»

lgHgnHunmmmflnn-S.

. Amumxug
unmcsum manuBmu mums

. AommH umunHz

«H 34 end mmmH HHmm nH mo
mHHn3 .mnHHnumnmm ummu§H0>

on couanuuum mnH Hspmnom

5 ocean man. H33 «0 8:

gummy

lgHgnHunmmumn—Hufil.

. Amumxuoz
unmnsum manusumu muma

.mmummmHmo umnuo

end 38 Hammsm unmosum
3H3 mnHumoHnse-BO no.“
pmommn was mEHu on GHmm
...an mnHHnomnmm @958qu
uOmHPummsm HmnnOmumm men.

803306 oz .

.mumH...omnom mo 95...
nm>mm one mHsomnmmiuHmm
0... man mums 0:3 mumqu:
unmcsum manusnumu mum:

 

 

 

 

 

mumxuos acmosum mmm up Home 3333 ucmpsum 8H no 88 39.33 ucmgum me «0 mom 83
sums
puma—HQ: ago g .H puma—H5 ugh mag—5.0m .H puma—Hos £85m SHE—Sum”— .H mnHunm
Ems
x0380 snoum do: 32m :8 nonneacomm: 2

 

m no H 086
up made 95.... mo nOHuomHHoo mnu ERG nmumuflnmm muHsmmu mo muHHHanumt, m5 uOu mnHunsooom muoummm

.mnnmu mmunu mnHusc mHHmn mmnmonmu mmunu
.3. mHnme

 

 

36

.mEHu mnu mo woo omumEHumw
cm nouusooo uH “mCHHsnmnum
Macaw 2H ucmmm QEHu may
Uuoowu ou uomu0w mumHsuonom
mad»

mchyoomu go“ gonna anal: .

.muomH>umm:m

ucmnsum mmunu new MomH>
tummsm «EH» HHsm 3w: m >2
um>o :mxmu mm: mCHHsuocom

uonuucum_3uz_.

.Amumxuoz
ucmusum mCchsumu mum3
mumxu03 ucmnaum com mo ammv

uuuxuoa.u:uu:uu mcHauauuz .

.muomH>umm:m ucmusum o3u an
um>o cmxmu mmz mcHHsnonom

uufinuucum.3mz .

.Amumxuoz
ucmusum mCchsumu mums
mumxuo3 ucmcsum mMH «0 momv

«Hagan: ucauauu mchuauma .

.mumxuo3 unansum «0 mean
long nuH3 Hmmc new mUHmmo
5 >96 3 88m: 2 25
may wonuoowu uomH>ume5m
ucmvsum m5 . ucmmfium comm
mcHHsnmnomlmu :H ucmmm mEHu
ms» mCHuuoowu can» umnumm
mad»

mcHunoomn gnu gonna anal: .

.ommH umucHz :H

wmm van mme HHmm :H wvH
mHng .mcHHsnonom HMCHM 0»
nouanuuum mcflflsoonom :H
ucwmm QEHu Hmuou mo won
mcHasuonum

Hagan cH gamma uawu_ax¢..

.mumxuoz
usmnsum mchusuwu mum:
mumxuoz ucmnaum mm «0 amN

numxg03.ucuasuu mchusuua .

.AcmHm
Vumom .m.mv mumxuoz ucmo
Isum 3m: qu xooH Cu moms

 

 

mumB muanum 60mmmuocH mmmH

sums

muuxaos accusum uo uuonm . mumxgozmucuuaum yo yucca . muugaoa unuuaum yo uuonm . Mama
sums

meQEoU anum

Hana zanm

 

 

HHmm uonndlcommz

 

Ham:

 

m no N mwmm

ADOUSCHMCOOV .vH mHnme

 

 

 

 

.mumxug
ucmmxfim mchusumu Mums
mumxuos acmusum omm «0 mom
mug—“53 admin—m gum .m
.mmmH 3mm 6.5» 05 «0 wow Bug:
CH m3 and $3 mcHumm :H 1.3%. cm 62.3000 3. umcH .mumxug
w: 333 5536028 Hach IHscmnom CH ucwmm Q59 05 usmnnfim mchusumu 803
m“ :H gamma «EH» Hmuou mo “mm oncomu on nomuom mumflsnmnom mumxuo3 acmnsum mp mo mom
maHanuuaum mean
HmcHucHqu—US.N §5u§§.~ gugu§m§§§&
. Amonflummmum non
..Emnsum .3398 36 w:
wmswomn 33 3mm new mmmH
mcHumm 5 00m: umfi Eouw
ucmumflHn mmz uOmHPummsm .uOmHPummsm .uOmHEmmsm
ucmcaum m5 3 flow: 05 HoncOmuwQ 3m: m an “6.5 HmEOmuwm 3m: 0 >9 meow
-Hauocom HmcHu «0 oozuue was :mxmu mm: mcHHsnmnom one mm: mcHaaoonom Hmch was ommH
. Emma.
@938 gnu—Um #5333 .H Mug 3&2 .H 33% .62 .H umucHz
. . sums
xmaggpu xuoum Ham: swam Ham: uonn<-c0mmz damn

 

 

 

 

 

m mo m mmmm AnwuscHucoov .vH magma

' 98
_a_ s o‘tuz- "' 49“ S s ‘u neuon: : '1 P oto -e

The interface and scheduling strategies implemented in
the SPRF expert system demonstration prototype were developed
by the knowledge engineer. Appendix C, written by the.
knowledge engineer, is a detailed description of the interface
and scheduling method used by the SPRF expert system
demonstration prototype.

In summary, Figure 6 displays the basic scheduling
structure implemented in the SPRF expert system demonstration
prototype. This scheduling structure consisted of the top
specialist and its sub-specialists. The Term Scheduler was
responsible for all of the scheduling. The plan it used called
the Partition Students specialist, then the week Scheduler
specialist, and finally the‘leekend Scheduler specialist. The
Partition Students specialist loaded in student workers from
the file and placed them into lunch, breakfast, and dinner
groups by examining their class schedules to determine into’
which meal during the week student workers could be scheduled.

The Week Scheduler specialist was to schedule lunch, then
breakfast, and finally'dinner. The plan used by the Lunch sub-
specialist was to call the Priority specialist until all of
student workers in lunch group had been scheduled. The
Priority specialist's plan was to set up a group of jobs and
student workers needed to fill the jobs, and then to call the
sub-specialist Group. The sub-specialist Group was responsible

for scheduling a group student workers into a group of jobs.

99

Figure 6. The scheduling structure in
plemented in the
SPRF expert system demonstration prototype (Appendix C).

 

FSA Term
Scheduler

 

 

 

 

1

Partition Students Week Scheduler Weekend Scheduler J

 

 

 

 

 

 

 

 

     
 
    

 

 

 

 

Dinner

 

Breakfast

 

 

 

 

{

 

 

Priority

 

 

 

Group
‘L____,,

 

Laboratory testing of the SPRF expert system
demonstration prototype consisted of scheduling the 213 actual
student workers of Brody Complex, Winter Term 1990. This test
was evaluated using a predetermined check-list (Table 9 in
Section 3.2.4.3.2). After the answer to each question,
suggestions were given for further improvements of the SPRF
expert system demonstration prototype. Results of the
evaluation were summarized in Table 15.

For the first question in the check-list "Are problem-
solving strategies that are incorporated into the SPRF expert
system demonstration prototype complete, accurate, and
consistent 7", the answer was no. The problem-solving
strategies incorporated in the SPRF expert system
demonstration prototype were not complete for two reasons:

1. Rather than just considered student worker class
schedules as the SPRF expert system demonstration
prototype did, the personnel supervisor also
considered student worker job preference, work time
preference, and work day preference.

2. The personnel supervisor tried to give each student
worker the number of shifts he/she desired rather
than to schedule each student workers into three
shifts per week as the SPRF expert system

demonstration prototype did.

101

Table 15. The results of evaluation of the SPRFa expert
system demonstration prototype in a laboratory environment

 

Question Evaluation

 

1. Are problem-solving strategies that are No
incorporated into the SPRF expert system
demonstration prototype cowlete, accurate,
and consistent ?

2. Does the SPRF expert systu demonstration No
prototype solve problems in the natural way
the expert scheduler prefers 7

3. Are the results of scheduling organized Yes
and presumed at the right level of
details that satisfy the schemle's
intended use ?

s. Does the SPRF expert system demonstration
prototype help the user in some simificant

way 7

a. Greatly reduce the scheduling time *

b. Efficient utilization of available *
student workers

5. is the interface friendly and clear for "

different grows of users (e.g. personnel
slpervisor, student smervisor) 7

 

 

 

 

. SPRF 8 Scheduling part-time student workers in the university residence hall foodservice
b operation.

Boehm et al. (1978); waterman (1986).
* There is no answer available yet.

102

To satisfy student workers but at the same time make sure
their needs were fulfilled, personnel supervisors usually gave
student workers the work time and day they preferred unless
they had lunch work shifts open but not marked preferred.

The SPRF expert system demonstration prototype did not
consider student workers' preferences of job, time and day at
all because these considerations could make the SPRF expert
system demonstration prototype too large. Thus, it could be
possible that student workers would ask to be rescheduled if
they were not happy with the work time and day assigned. With
this demonstration prototype, human judgement (from human
scheduler) would still be part of the final solution
(rescheduling student workers to meet their needs).

deHass (1983) pointed out that computer programs to
handle a scheduling problem on a large scale were relatively
simple, but could be expensive to use due to the large number
of human judgments going into the solution. The SPRF expert
system demonstration prototype could result in excessive
costs. However, considering student workers' preferences of
job, time and day could be done by further programming efforts
in the stage of research prototype (Figure 1).

The SPRF expert system demonstration prototype attempted
to schedule each student worker into 3 shifts for weekday
scheduling and 2 shifts for weekend scheduling instead of
considering the number of work hours desired by student

‘workers. The quantity, three and two shifts, were the average

‘ . 103

number of work shifts assigned to each student worker for
weekday and weekend scheduling respectively each term.
However, some student workers could not work.more work shifts
than they had indicated on their applications because they
may have wanted to spend more time in studying and/or other
employment activities; in the meantime, other student workers
who wanted to obtain more work shifts to earn extra money may
not get the extra work shifts. The method used by the SPRF
expert system demonstration prototype may have been able to
speed the scheduling process due to less complexity in
decision-making process, but would require extra time of
personnel supervisors to reschedule student‘workers according
to their specific desires.

For the second question in the check-list "Does the SPRF
expert system demonstration prototype solve problems in the
natural way the expert scheduler prefers P", the answer was
no. The embedded reasoning process of scheduling did not seem
to work in the natural way preferred by the expert scheduler.
For example, lunches had traditionally been the hardest to
staff ; therefore, lunch work shifts were scheduled first.
However, in a laboratory test, 20 lunch work shifts,
pertaining to the first six job priorities, were not filled
after the scheduling was completed by the SPRF expert system
demonstration prototype (Table 16). On the other hand, only
eight lunch work shifts were not filled after the same

scheduling (Winter Term 1990) was completed by the personnel

104

Table 16. The contrast between manual scheduling of part-
time student workers (MSPS) and scheduling by the SPRF“ expert.
system demonstration prototype on the number of open work
shifts of the completed weekday lunches schedule at Brody
Complex, Michigan State University, Winter Term 1990.

 

 

 

 

 

 

 

 

 

Priority Job liaise Total The timber of The umber of Difference
Possible Open Work Open work of ISPS
Shifts Shifts by Shifts by vs. SPRF
"SP8 SPRF

1 MM" 5 0 S 4
nCONi 5 1 0 1

eFLUT 5 0 0 0

eCOfli S 0 0 0

sTc1 5 0 0 0

dDRl 5 0 O 0

2 nSDT 5 0 0 0
e801 5 0 0 0

sUT1 S 1 l 0 1

sTCZ S 0 0 0

dP&P1 S 0 6 -b

3 nCONZ S 0 0 0
eCONZ 5 0 0 0

dDRZ S 0 0 0

dDR3 S 1 0 1

6 nSDZ S O 0 '0
e802 5 O 0 0

sL-Stock 5 3 3 0

dP&P2 5 2 4 -2

5 nSD3 S 0 0 0
e803 5 0 0 0

sUTZ 5 O 0 0

dbkb 5 0 2 -2

6 nDESS S O -3
dDRS 5 0 0 0

Total 125 8 (6.4%) 20 (16%) -12

 

 

 

 

 

 

 

a SPRF = Scheduling part-time student workers in the university residence hall foodservice
operation.

105
supervisor at Brody Complex. Therefore, to make sure that
critical jobs in lunch scheduling were filled by student
workers, improvements should be implemented in the SPRF expert
system demonstration prototype.

For the third question in the check-list "Are the results
of scheduling organised and presented at the right level of
details that satisfy the schedule's intended use 7", the
answer was yes. The master schedule, indicated in Appendix D,
generated by the SPRF expert system demonstration prototype,
was similar to that generated by the expert scheduler
manually. The only difference was that the student number was
shown in the master schedule generated by the SPRF expert
system demonstration prototype, instead of the student worker
name. However, it would be very easy for the knowledge
engineer to change student worker number into student worker
name. As to the list of all student workers (indicated in '
Appendix E), personal information such as address and phone
number, job preference, work time and day preferences, and the
number of work hour desired need to be included to keep a
complete student worker data file for the use of personnel
management (e.g. rescheduling).

Additionally, the flexibility for'determining the format
and content of outputs would be beneficial, which could be an
improvement in the SPRF expert system demonstration prototype.

For the fourth question in the check-list "Does the SPRF

expert system demonstration prototype help the user in some

106

significant way 7", there was no answer available yet. The
SPRF expert system demonstration prototype seemed to be faster
than manual scheduling when performing the element of term
scheduling function, Schedule, under the same condition
(scheduling 213 student workers at Brody Complex, Winter Term
1990) . Schedule was defined as the process of matching student
workers with available work shifts by considering their past
experience and performance, alternative weekend, job, work
time, and work.day preferences; number of work.hours desired:
in/out time of available work shifts; and class schedule
(Table 5). ,

The total time that the personnel supervisor at Brody
Complex spent in performing the element "Schedule" of term
scheduling during Winter Term 1990 was 44.33 hours. The
personnel supervisor was unable to exactly separate the time
spent in weekday scheduling from. that spent in weekend
scheduling; he estimated that one-eighth of the total time
spent in term scheduling was used to perform weekend
scheduling. In other words, approximate 38.79 hours were spent
in weekday scheduling. Compared to 2.18 hours that the SPRF
expert system demonstration prototype spent for performing the
element. "Schedule" of ‘weekday scheduling ‘under the same
condition (scheduling 213 student workers at Brody Cpmplex,
Winter Term 1990) , the SPRF expert system demonstration

prototype seemed to be faster than manual scheduling.

107

With the same number of shifts and student workers, the
percent of weekly shifts filled by the SPRF expert system
demonstration prototype was 86.86%, while that by the manual
labor scheduling was 92.21% (Table 17). At this point, manual
scheduling was more efficient; however, the efficiency of the
SPRF expert system demonstration prototype could be increased
by further programming improvements.

For the fifth question in the check-list "Is the
interface friendly and clear for different groups of users
(e.g. personnel supervisors, student supervisors) 2", there
was no answer available yet. The SPRF expert system
demonstration prototype did not yet have a friendly and clear
interface. The criteria (7.1-7.9) indicated in Table 8 would

be the goal to achieve in the near future.

, 108
Table 17. The contrast between manual scheduling of part-
time student workers (MSPS) and scheduling by the SPRF' expert
system demonstration prototype on the number of open work
shifts of the completed weekday schedule at Brody Complex,
Michigan State University, Winter Term 1990.

 

 

al Lunch Breakfast Dimer Total Per
it- Day
Total Mariner of 195 100 216 611

Weekday Work Shifts

 

The timber of men Work 22 10 0 32
Shifts lie-mining by MSPS

 

 

Percent of Work Shifts 08.72% 90.00% 100.00: 92.21%
filled by HSPS
The later of Open Work 56 0 0 56

Shifts Remining After
Scheduling by SPRF

Expert System

 

Percent of Work Shifts 72.31% 100.00% 100.00% 06.86%
Filled by the SPRF Expert

System

 

 

 

 

a SPRF: Scheduling part-time student workers in the miversity residence hall foodservice
operation.

Chapter V

CONCLUSIONS

W
W

AI research papers described only scantily the methods
employed in the development of expert systems, which were not
specifically related to foodservice applications. Therefore,
the method used in this study to construct the SPRF expert
system may provide the foodservice industry with a model of
how expert systems could be developed and used.

The process of building an expert system usually involved
interaction between the , expert-system builder, knowledge
engineer, and one or more human experts in some problem area.
However, an intermediary between the knowledge engineer and
human experts (personnel supervisors) was needed by the
present study. The human expert (personnel supervisor) was not
be able to work with the knowledge engineer for the entire
time needed for the expert system development for two reasons.
First, the personnel supervisor was a entry-level position:
people in this position were often promoted to higher
positions after they had been in the position for approximate
6 months. Second, the turnover rate was high in the university
foodservice even at the supervisor level. Therefore, an
intermediary was needed for this study; it may be necessary

for the expert system development in the foodservice industry.

109

110
i s e
W
Discussion of difficulties occurred in the development
of SPRF expert system could be divided into two parts: (1)
inherent limitations of expert systems and (2) limitations of
the expert system building tool "DSPL". Each part is discussed

further below (Sections 5.2.1 and 5.2.2).

WW
Even though expert systems have better ability than

conventional computer programs to mimic thehuman reasoning
process of problem solving (discussed in Section 3.2.2), they
still could not behave totally as humans experts for the
foreseeable future (Vedder, 1987). In this way, a challenge
faced. in. the development. of ‘the SPRF’ expert system is
balancing time efficiency and performance quality to produce.
the most optimal results.

Additionally, the power and utility of the resulting SPRF
expert system depends on the quality of the underlying
representation of human scheduler knowledge. The human
knowledge may not all be mapped into the system; therefore,
eliciting the right amount. of“ knowledge form the human
scheduler and representing that knowledge at the right level
of details, which allows the- developed expert system to
replicate important and valuable parts of the performance of

expert schedulers, becomes another challenge.

111
l! . e" e g‘ '3'1‘1, 5,, ;1_ e ’0 ea

DSPL could not recognize incorrect or inconsistent
knowledge. Therefore it is likely to produce incorrect results
or advice when inconsistencies and/or errors are incorporated
into the knowledge base (e.g. job priority).

DSPL was a programming language for designing expert
systems that performed problem solving on routine design tasks
(see Section 2.5). However, the SPRF problem domain was not
a typical routine design. Some tasks in SPRF problem were of
a classificatory nature. For example, the personnel supervisor
at Brody Complex sorted student workers into lunch, breakfast,
and dinner groups. Some tasks in the SPRF, problem were of
rule-based nature (if. . , then. .) , such as scheduling n student
workers into m jobs (e.g. if the student worker could not fill
in any lunch shifts, then place this student worker on the
underscheduled list).

As to the problem of rule-based nature, the domain
knowledge was usually represented as sets of rules that were
checked against a collection of facts about the current
situation. For example, a rule in the SPRF domain may be that
if the student worker have previous experience on one critical
job, then the student worker is a candidate for this job. Each
time, when the computer reads in one student worker, this rule
would be checked against the previous experience (the fact)
of the student worker. When the IF portion of a rule was

satisfied by the fact (student previous experience), the

112
action specified by the THEN portion was performed. That is,
this student worker would be moved into the candidate list for
the job.

Since DSPL.did not support classification and rule-based
methods for representing knowledge and controlling the process
of problem solving, some tasks. performed by the human
scheduler could not efficiently be modeled using DSPL. In the
present study, the programmer developed a LISP function to
classify student workers into lunch, breakfast, and dinner
group and to do the basic scheduling of n students into m
jobs. The LISP function returned the problem back to DSPL when
the scheduling process (scheduling n students into In jobs) was

done. Thus, DSPL was able to function more effectively.

5,; The Eotgntial Benefits of Using
t F e S s e
The completion of weekday scheduling has suggested that
system development would be possible.Based on the testing of
weekday scheduling in a laboratory environment and data
obtained from collection of time data on manual scheduling of
part-time student workers (discussed in Section 4.2) , the
following potential benefits would be expected in the future
by using a well-developed version of the SPRF expert system

to facilitate scheduling.

. 113

MW

.A human scheduler at different times or schedulers in
different places may reach different conclusions about a
particular scheduling situation (discussed in Section 4.2).
An expert system always provides the same conclusion to the
same problem. This consistency is an obvious advantage in
using the SPRF expert system in matters of fairness among
employees. In addition, the SPRF expert system would be able
to eliminate the occasional inadvertent omissions or errors
that schedulers make (e.g. forgetting to write the shift
planned to give the student worker on his/her individual

schedule sheet).

. ' o m e

The SPRF expert system has a potential to greatly reduce
the time needed to schedule student workers per term and
relieve supervisors from handling problems which, for the
experts, are routine. Consequently, the Department of Housing
and Food Service at MSU would be better able to use its
supervisors' time and abilities on the really tough problems,
such as personnel management, determining the amount of work
force for next term, and adjusting the number of shifts. This
switch in emphasis would also help the human experts by making

their jobs more interesting and rewarding.

114

5. Labo Co 3

The personnel supervisor at Brody Complex:delegated.most
of the scheduling tasks to student supervisors in Fall Term
1989 and Winter Term 1990. The hourly salary for a student
supervisor was approximately $5.00 per hour. As discussed in
Section 4.3.2, the amount of time needed to perform the
element ”schedule" of weekday scheduling manually was 38.79
hours, while the SPRF expert system took only 2.18 hours. In
other words, by using the SPRF expert system demonstration
prototype, 36.61 hours of labor at a cost $5.00/hour (hiring
student supervisor to do scheduling) , $183.05, could be saved.

Overall, savings in labor costs are hard to determine at
this time. A formal cost benefit analysis, including the cost
of this system and of hiring employees to perform scheduling,

should be completed.

5. .4 r rv ' E e ise

The SPRF expert system. can help preserve existing
knowledge. Also, expertise may be extended to all residence
halls at MSU, even foodservice operations at other
institutions employing part-time workers, through use of

portable floppy disks.

Chapter VI

RECOMMENDATIONS FOR FURTHER RESEARCH

The SPRF expert system. demonstration prototype has
completed weekday scheduling (vs. weekend, final, and
volunteer scheduling) in a laboratory environment. This
suggested that expert systems technology could.effectively be
applied to the scheduling part-time student workers problem,
and that the system development would be beneficial for the
university foodservice operation. Therefore, the current SPRF
expert system has passed the stage of demonstration prototype
and has entered the stage of research prototype, and then will
enter the stages of field prototype, production model and

commercial system, as shown in Figure 1.

§ml_B§§§eI§h_EIQEQE¥E§

Further developments in the research prototype stage '

should include the following:

1. To improve the current SPRF expert system in terms
of efficient use of available student workers,
scheduling time, completeness of embedded problem-
solving strategies, and accuracy of the reasoning
process of scheduling (discussed in 4.3).

2. To extend the current SPRF expert system to include
the final scheduling and volunteer scheduling

functions.

115

‘ 116

3. To develop the selective use of the following
elements of scheduling functions:

a. Rescheduling of a single (or few) student
worker.

b. Check schedule

c. Copy schedule

d. File schedule

e. Change and update student database, job
database, and completed schedule database.

4. To establish the user interface. The criteria 7.1-
7.9 in Table 8 (Section 3.2.4.3.1) can act as a
guide for the development of user interface.

5. To test and refine the research prototype by (1)
bringing in additional experts to help validate the
system's accuracy, and (2) using test cases not
encountered by the system during its previous
development (Waterman, 1986).

In addition, one of objectives of all expert systems was
to help less expert professionals (e.g.-trainees) improve the
quality of their judgments and.decisions and, possibly, their
general problem-solving skills (Boritz and Brown, 1986). This
Objective was not included in this study. However, faced with
turn over of labor schedulers (e.g. three personnel
supervisors had taken over the scheduling between Spring Term

1989 and Winter Term 1990 at Shaw Hall, two in Mason-Abbot

117

Hall, and three in Brody Complex), it would be beneficial to
have the SPRF expert system also acting as a trainer.

To be able to play a role in training new schedulers, the
SPRF expert system needs to be equipped with the capabilities
of explaining how the scheduling is done, which would allow
users to explore each facet of their reasoning. Thus, a new
scheduler could be allowed to track, step-by-step, the
analysis leading to the solution or recommended decision.
Also, a new scheduler could use this system to support or
contradict their solution to a problem. In this way, the SPRF
expert system also holds promise to reduce training costs and
ensure consistency in the presentation of problem solving
skills.

As the SPRF expert system performs with adequate
reliability in the laboratory environment, it could pass from

the research prototype to the stage of field prototype.

LAW

In the field prototype stage, the SPRF expert system
research prototype could be revised and refined based on
extensive testing in the user environment (cafeteria
foodservice at a residence hall of MSU) . The further research

at this stage should include:
1. To design and conduct the experiment of collection
of time data on manual labor scheduling by taking

the variables discussed in Section 4.2 into account.

118

2. To design and conduct the experiment of collection
of time data on scheduling by the SPRF expert
system.

3. To evaluate the SPRF expert system on the aspect of
time-saving according to the time data obtained from
1 and 2.

4. To evaluate the SPRF expert system according to the
criteria shown in Table 8. The approach to evaluate
the SPRF expert system has been proposed in Section
3.2.4.3.2.

5. Compare the SPRF expert system scheduling to manual

scheduling in terms of labor cost and investment.

6 ' l C e ' S 5

At the stage of production model, the further work would
be to transfer the whole program from a sun computer system
to an IBM or Apple computer and run the program on-site at
residence halls of MSU. The portability of the completed SPRF
expert system field prototype can be troublesome. DSPL used
to build the SPRF expert system requires LISP or LISP-based
software (and even hardware) environments which are not
readily available for the user. A finished SPRF expert system
field prototype may need to be recorded in a more conventional
programming language (like C) for the sake of portability. To
use a production model on a regular commercial basis, more

effort would be needed, especially in determining the target

119
market, the needs of target market, and identifying existing
or/and possible competitors. In addition, the user interface
may need to be tailored to fit the needs of different groups

of users.

 

GLOSSARY

Artificial intelligence (AI). The sub-field of computer
science concerned with developing intelligent computer
programs. This includes programs that can solve problems,
learn from experience, interpret visual scenes, and, in
general, behave in a way that would be considered
intelligence if observed in a human.

Domain expert. A person who, through years of training and
experience, has become extremely proficient at problem
solving in a particular domain.

Domain knowledge. Knowledge about the problem domain (e.g.
knowledge about scheduling part-time student workers in
a university residence hall foodservice operation).

User. The person who uses the finished expert system: the
person for whom the system was developed.

Expert system. A computer program that uses expert knowledge
to attain high levels of performance in a narrow problem
area. These programs typically represent knowledge by
using heuristics, and examining and explaining their
reasoning processes.

Expert-system-building tool. The programming language and
support package used to build the expert system.

Heuristic. A rule of thumb or simplification that limits the
search for solutions in problem domains that are
difficult and poorly understood.

Inference engine. The part of a knowledge-based system or
expert system that contains the general problem-solving
knowledge. The inference engine processes the domain
knowledge (located in the knowledge base) to reach new
conclusions.

Inference method. The technique used by the inference engine
to access and apply the domain knowledge (e.g. forward
chaining and backward chaining).

120

. 121

Knowledge acquisition. The process of extracting,
structuring, and organizing knowledge from some source,
usually human experts, so it can be used in a program.

Knowledge base. The portion of a knowledge-based system of
an expert system that contains the domain knowledge.

Knowledge engineer.- The person who designs and builds the
expert system. This person is usually a computer
scientist experienced in applied artificial intelligence

methods.

Knowledge engineering. The process of building an expert
system.

Knowledge representation. The process of structuring

knowledge about a problem in a way that makes the problem
easier to solve.

Master schedule. A schedule consisting of the jobs
(including the starting and ending times) that need to
be filled for each meal on each day.

Rule. A formal way of specifying a recommendation,
directive, or strategy, expressed as IF (premise), THEN
(conclusion) or IF (condition), THEN (action).

Search. The process of looking through the set of possible

solutions to a problem in order to find an acceptable
solution.

SPRF. The problem domain of scheduling part-time student

workers in the university residence hall foodservice
operation.

SPRF expert system. The computer program developed by this
study to perform scheduling'of'partetime student workers
in a residence hall foodservice operation.

Symbol. A string of characters that stands for some real-
world concept.

Symbolic reasoning. Problem solving based on the application
of strategies and heuristics to ‘manipulate symbols
standing for problem concepts.

Work shift. A particular job for a particular meal on a
particular day.

APPHDICES

 

 

 

 

 

 

 

 

 

 

 

c cc.c cc.o meHesoonum numbness:

m mo.a mm.a masaonommm

a Hm.m no.m masaonom

N mo.~ oo.m massage» 0» mnucmmuo

on mo.m m~.ma meHasaarum HmaHm

mm om.pa Ho.m~ masaosommm

N om.~ om.~ masaocom smoo

Hm ma.mm sm.ms masomnom

m mmé oo.~ 03828 on 0NHccmuo

as 2.8 8.2. 0:93 Bus mm 2.0 $3
a om.c cc.H mcHasuozom amouasaas

H om . 0 mm . o 0.3602000m

a om.o oo.H masownom maHm

u m 8.0 «To 033:8 38
l v om . N mm. . N 038:8

m Sim 8.». 038:8 8 02.895

Ha om.m «v.5 meaasaorum Hanan

m m m . o S . H 0300.18.31

a mm.a He.a wasnmsom mane

we ma.m ma.~a masomnom

m Hm.a mw.a maaumnom on mNHcmmuo

8 ~99 8.3 0:933 . Ewe ma 80% $2

AcHec cascaosm
umxuoz unmasum Amuse mumxuoz
u0m @0802 .509 umm 600002 mu:0&0Hm m”: can uc0wsum

z 0.59 0mcu0>< 0.58 no 050.5 coHuoch mcHHSBcom no .0952 c509 1.80»
N no a mama

. HI¢ vacg

stasis: 33m .538? £31 u8n<ic8mz am 89.83
200300 063100?“ mHumu0mco no mCHH~60£0m HoocmE :0 Ben 05.... mo coHuo0HHoo 50.3 005300 0000

 

123

.500 .00 00000: 0.50 Mo 05050 0050000 on u0fi0mou @0000 0:0 0050000 005000 NO .5350 u z

 

 

 

 

 

 

 

 

 

 

a co.c oc.c mcHanuonom uuoueaap>
a no.o mo.o mummoamo
o so.m mm.m wasnmcom
a om.o oo.H masoocom op mNHcmmuo
a va.m Hm.a meaaaumaum Hmcam
m Ho.a mo.~ mazaocommm
m ov.H mH.H announce mean
ma o~.oa ma.~a masoonom
N ma.o ~m.o masoonom on mNHcmmgo
«N 8.3 8.: 85:33 58. me .0053 82
Andes nosoamsm
sweeps ucmnsum Amuse mumxugz
2 00a 8802 68. use 8802 $5.00 man can. 0:830
958 0m0u0>< 0.59 no 0:395 coHuoch mcHHsBcom no 00952 509 u00w
N co m moms Acmuscaucooc .Hua xaocwom<

 

 

 

 

 

 

 

 

 

 

 

 

 

m ca mm.mH cm.am meHasuagum enema
mm mm.Hm om..o~ maHasumrum gums. mm. H.mm mmma
a mm.o mm.a whammHmo
m mm.~ mm.m masaonom mean
. mh.o om.a masamsom moose
m oo.a oo.~ masaoaom
m mm.H oo.~ masomnom 00 muHcmmuo
m «H.m m~.¢H maHaauasum umouaseos
m mm.m ma.ma . masamnom
“a o mm.o me.~H meHaaaorum enema
1. m me.o m~.. mumomamo
. mm.o mm.o masoonommm
a mm.~ mm.m masoonom 0.00
e mm.a mm.m masoosom moon
m. mm.m. om.m~ masaonom
m mm.m mm.o. masaonom on muscmmuo
av mm.e~ mc.ma meHaaumaum .numm om. mcaumm mam.
.cHsc nomonsm
09.82 .....mofim .35 23.83
z umm ooammz some umm ooaomz mucmewam mu. new unmasom
0.59 000002.. 0.5m. mo 0.50.5. 0030ch mcHH000c0m .00 000.52 .509. 000.»
N no H 0m00

$500050: 0000m :05...on 730: 30cm 00 000x003 0000000
05010000 50000000 m0 905000200 H0000... so 0000 0.50 no :0H000HH00 Scum 0050000 0000 .mi... 50:009..

125

.0000 000 000000 0.50 00 000080 0050000 00 u0000m00 00000 000. 0050000 050000 00 00500006. n z :

 

 

 

 

 

 

 

 

 

 

0 m0.o m~.o 00:00:00 mmoo
0 m0.o mm.o 00:00:00 000:0 .
0 om.o oo.0 00:00:00
0 m~.o om.o 00:00:0m on 000:0000
: 0:.0 o:.~ m:00::0:00 u00:::09>
m mm.m om.m 00:00:00 mmoo
m mm.0 om.m 00:00:00 000:0
m om.m oo.m _ 00:00:00
0 :c.m 00.00 m:00::0:um 00:00
00 om.: oo.m 00:00:0000
m m~.o om.o 00:00:00 0000
00 mm.m om.:0 .00::0:0m mmoo
m0 m0.m m~.m 00:00:00 000:0
mm mm.mm om.00 00:00:00
: m0.0 m~.~ 00:00:00 on 0:0:0000
m: 00.0: :c.o:0 m:00::0:um .nsym o~0 000:03. omm0
.:00. 00:00:20
00:00: 0:00:00 .mu:v 0:00:03
2 :00 000002 :00: :00 000002 3:050: 000 0:0 0:00:00.
0.5.0. 0m000>¢ 0.5.0. 00 00000.0 00000000 95000000 00 000052 . 8.00.0. .0000.
m 00 0 0mm: . 0000:5088 1?: 0.00.0000...

 

126

 

 

 

 

0 00.: 00.0 00:00:0000
m 00.0 00.0 00:00:00 0000
m 00.0 00.0 00:00:00 0000
m 00.0 00.0, 00:00:00 000:0
mm 00.0 00.00 00:00:00
00 no.0 00.00 00:00:00 00 000:0000
:0 00.00 00.00 m:00:00:00 .5000 com 0000 0000
m 00.0 00.0 00:00:00
0 00.0 00.00 00:00:00 00 000:0000
0 00.0 00.00 0:00:00:00 0000::00>
00 00.0 00.0 00:00:00
:0 00.0 «0.0 0:00:00:00 00:00
0 on.0 00.0 000000000m
0 00.0 00.: 00:00:00 0000
0 00.0 00.0 00:00:00 0000
0 00.0 00.0 00:00:00 000:0
mm 00.0 00.00 00:00:00
0 00.0 00.0 00:00:00 00 000:0000
00 00.00 00.00 m:00:00:00.:00a 000 0:0000 0000
0:000 00000000
000003 0:00:00 000:: 000000:
z 000 000002 000.0. 000 000002 00:2000 000 0:0 0:00:00
0.00.0. 0m0u0>¢ 0.5.0. no 000054. 000000000 m5 0000090 no 000.52 5.009 .000»

 

 

 

 

 

 

 

N no 0. 000m

4000000300 000nm 009000.: 30009000 >000m 00 0.000.003 0000000
0800:0000 000000000 00 000000000000 00000.: 00 0000 000.00 00 000000.300 00.00 00000000 0000

.mu< 000:0000

 

127

.53 uwm @0ch menu mo ago—5m magmumc ou umfimmou 0069.. can. 605330 mmauucm mo ~333an n z «

 

 

 

 

 

 

 

 

 

Em?

*-

 

o oo.c co.c meadauocum nuancuaos
¢H Nh.m om.o¢ masconom
m m¢.o oo.~ wasnmcom op muflcmmuo
ma c~.o~ cm.~q mcwaauocum dmawm
a 3.0 omd 3338
m om.o mm.m masconom mafia
ma vm.~ mm.HH masownom amoo
m ov.m oo.oa mflsumnom xownu
mm ¢m.oa mm.¢v mannmcom
ma «o.m om.m masuwnom ou wwflcmmuo
E. 3.3 36“ Egg 5 omm ~33: 82
o co.o oo.c mcaasuoaum “moucuapa
m mo.o mm.~ masuozom
m mad and 0.38290: ou mnacmmuo
m mm.~ mu.m mcflfisuonom Hanan com Hams mama
:55 @0535
3&0: ucmosum Ga: 39.83
z umm 8802 88. “8 R682 3858 EH can usmnsum
manna mmmumzm 958 no uchE c0305 0530038 mo umnafiz umww

 

N no N «mam

Auoscflucouv .mLa xflncmamc

 

128
Appendix B. Knowledge Representation in DSPL (By Robert

Hauser, Department of Computer Science, Michigan State
University)

The knowledge elicited from the expert led to its
presentation in DSPL. DSPL provides two useful knowledge
representation facilities, agents (e.g. specialist, plan,
task, step, etc.) and a design database. Expert knowledge on
the scheduling process was encoded both in the agent structure
and the design database. The procedural knowledge (scheduling
method) was reflected in the agent structure while the
declarative knowledge (e.g. job priority, the in/out times of
work shifts) was represented in the design database. The other
major function of the design database was to provide the
structures for constructing the schedules. Agents and the
design database are detailed below.

Agents were organized into hierarchies in DSPL to break
down the problem. Major functions had a specialist associated
with them along with their sub-hierarchy of agents. The SPRF
expert system had the term specialist at the top of the
hierarchy and it was responsible for the successful design of
a complete term schedule. Other specialists were responsible
for designing weekday, weekend, lunch, breakfast, and dinner
parts of the term schedule.

The interaction of these agents, their corresponding
functions provided the implementation of the problem solving

method of the expert as determined from the steps of

129
interviewing the expert and analysis of elicited knowledge.
The agents functioned by changing attributes in the design
database. These changes constitute the design activity and
the scheduling process.

The design database contained knowledge used in the
scheduling process of the agents (e.g. job priority, the
in/out times of work shifts) and also the structures for
storing the actual schedules that needed to be filled. Both
kinds of knowledge were represented in attributes of
components. Attributes were very small pieces of knowledge
that could have only one value (e.g. student number).
Components were related groups of attributes (e.g. student
workers).

As the problem solving is expanded beyond term
scheduling, more agents and design database components may be

added.

130

Appendix C. FSA (Foodservice Scheduling Assistant; the SPRF
expert system demonstration prototype) interface and
scheduling strategies (By Robert Hauser, Department of
Computer Science, Michigan State University).

1 FSA Interface

The following sections describe the current working aspects of the FSA software.
A windowing interface based on Xwindows is being added to the software.

1.1 Entering the Master Schedules

The master schedules consist of the jobs, including the starting and ending
times, that need to be filled for each meal on each day. These schedules must
be complete and provided to the FSA software.

Since the schedules may change over time facilities for modifying them must
be provided. The schedules are currently hard coded and not modifiable.

1.2 Entering the Students

The students must all be entered before scheduling. Students may be entered
by selecting the “Modify Schedules/Students” choice from the FSA MENU and
then selecting “Enter New Students”. The user is prompted for the name of the
student file. If the file exists the students are added to the file otherwise the file
is created.

Then the user is prompted for the student last name, followed by the student
number.

Then the student’s schedule must be entered, one day at a time. Each day
is entered as a list of free time periods having a starting time and ending time
for each time period. A sample list for Monday could be “((400 1000) (1400
2000))”. The times are entered as military time without colons. The sample
indicates that the student has free time on Monday from 4am to 10am and from
2pm to 8pm.

The last entry for a student is a list of the student’s previous jobs experience.
The experience is represented as the types of jobs a student has had. An example
would be “(flut tc dr)”.

To finish entering students, the return key is pressed when prompted for
the student’s last name.

Currently entered students are not modifiable.

1.3 Changing Scheduling Parameters

Some parameters to the FSA scheduling software are deeply related to the al-
gorithm itself and modification of these cannot be done. Other parameters
could be modified by knowledgeable users to tune the scheduling. Currently all
parameters are hard coded and not modifiable.

 

131

1.4 Running the Scheduler

To schedule the students that have been entered one just selects the “Run
Scheduler” Option on the F SA MENU. This begins scheduling. When prompted
the user enters the file name of the file containing the students. The rest of the
scheduling is non-interactive.

During the scheduling the time for different parts of the scheduling process
is written to a file “FSA.B.unCase”.

1.5 Generating Output

Upon completion of the scheduling two reports are automatically generated.
The first report is a listing of the master schedule generated with the jobs filled
in with the students that were given that shift. This report is stored in the file
“FSA.WeekSchedule.Report” .

The other report generated is listing of all the students. Each student print-
out includes the shifts in which he or she was scheduled. This report is stored
in the file “FSA.FinishedStudents.Report”.

An optional report, which shows the number of empty and full shifts can
be generated by selecting “Utilities” from the FSA MENU then “Analyze Jobs
Empty/Filled”. The report is stored in the file “FSA.Analysis.Report”.

Any of the files may be printed to the local line printer.

2 FSA Scheduling Strategies

This section, written by Robert Hauser, is an high level description of the
scheduling method used by the I" SA software. The boxes roughly correspond to
the specialist agents in the DSPL problem solver. For each specialist, the plan
that is used is outlined. '

2.1 Term Scheduler

Figure 1 displays the top specialist and its sub-specialists. The FSA Term
Scheduler is responsible for all of the scheduling. The plan it uses calls the
Partition Students specialist, then the week scheduler specialist, and finally the
weekend scheduler specialist.

The partition students specialist loads in the students from the file and places
them into lunch, breakfast and dinner groups by examining their schedules to
determine which meal during the week the student could be scheduled into.
Also experienced students are moved to the front of each list. Other students
are kept in first-come-first-served order.

The weekend scheduler is much the same as the week scheduler( see section
2.2),T here are two major differences. The first is that two separate weekends are
to be scheduled and used alternately during the term. The second is that each

 

132

 

FSA Term
Scheduler

 

 

 

 

V

 

 

 

 

 

 

 

Partition Students I Week Scheduler A Weekend Scheduler

 

 

 

 

 

Figure 1: FSA Term Scheduler

meal of each day is considered separately. That is, Friday’s dinner is scheduled,
then Saturday’s breakfast, then Saturday’s lunch and so on.

2.2 Week Scheduler

Figure 2 shows the sub-specialists of the week specialist. The plan used by
the week specialist is to schedule lunch' then breakfast then dinner. Lunch is
scheduled first because it is the most difficult meal to schedule.

Breakfast and dinner scheduling is done in the same way as lunch.

 

 

 

, Week Scheduler

 

     

 

 

 

 

 

 

 

Dinner

  

Breakfast

 

Figure 2: Week Scheduler

2.3 Lunch

The lunch sub-specialists are shown in figure 3. The plan that the lunch spe-
cialist uses is to call the priority specialist until all of the students in the lunch
group have been scheduled. Then the fixer specialist is called.

Some of the students may not have been scheduled and some of the jobs
may not be full. The fixer specialist notifies the user of any problem students
or jobs.

133

 

 

Lunch

 

 

 

 

Priority

 

Figure 3: Lunch

2.4 Priority

The priority specialist’s plan is to set up a group of jobs to schedule, set up
the number of students needed to try to fill the jobs, and then to call the sub-
specialist prioritygroups, figure 4, to schedule the group of students into the

group of jobs.

 

 

 

Priority

 

 

L...
i—GToup

fl ‘

 

 

Figure 4: Priority

The jobs are listed in priority order in the design database. Therefore, the
most important jobs will be filled first.

2.5 Group

The group specialist, shown in figure 4. is responsible for scheduling a group of
students into a group of jobs. No sub-specialists are called.

The plan is to select a student and a job and schedule the student into the
job until no more students remain. If the student doesn’t fit another job is
selected. If no job is found the start and end times of the shifts are relaxed and
a fit is tried again for the jobs. If the student still does not fit then he/she joins
the next group of students to be scheduled unless this type of failure has already
occurred in which case the student joins the group of students designated for
the next meal.

In the process of selecting a job for the student, the student’s previous ex-
perience is considered. Jobs in which the student has experience are attempted
first.

134

Appendix D. Master schedule (a sample print-out before
completion of user interface).

Friday's Breakfast Schedule

src1 --—----- (1061412 640 1015)
r0151 -------- (1132330 615 1000)
SUTl --------- (1133794 635 1000)
SL-SETUPl ---- (113015 515 930)
SL-SETUPZ ---- (1135736 515 930)
SLSl. --------- (1139871 640 1000)
EFLUT -------- (1104450 640 1015)
scou1 -------- (1132510 640 1015)
scouz -------- (1118725 700 1005)
£331 -------- (1126347 700 1005)
EEHZ --------- (1112655 715 1005)
EBDl --------- (1123875 640 1000)
rBD2 - ------ (1051148 700 1000)
EBDB --------- (1136092 715 1000)
DDR1 ---4 ----- (1134555 710 1030)
DDR2 --------- (1084356 710 1030)
DDR3 - ------- (123 730 1030)
VDDRé -------- (1138492 730 1030)
DPPl --------- (1130239 630 1030)

DPPZ -------- (1114661 715 1030)

 

135

Appendix E. Student worker list (a sample print-out before
completion of user interface).

Student Name Student Number Previous Experience
MAZZUIHI 1135826 (NONE)
Shifts

((BREAKFASTTUESDAY DPPl 630 1030) (LUNCHTUESDAY NCONl 1025 1430)
(LUNCHTHURSDAY NCONl 1025 1430))

Class Schedule

((MONDAY (1455 2200)) (TUESDAY (400 2200)) (WEDNESDAY (1455 2200))
(THURSDAY (400 2200)) (FRIDAY (1455 2200)))

Student Name Student Number Previous Experience
MAYS 1127438 (PP BD DR)

Shifts

NIL

Class Schedule

((MDNDAY (1345 1545)) (TUESDAY (1345 1845)) (WEDNESDAY (1345 1545))
(THURSDAY (1345 1845)) (FRIDAY (1345 1545)))

Student Name Student Number Previous Experience
MAYFIELD 1106216 (NONE)
Shifts

((BREAKFASTMDNDAY 5031 635 1000) (BREAKFASTWEDNESDAY SUTl 635 1000)
(LUNCHFRIDAY DDR7 1145 1530))

Class Schedule
((MONDAY (400 1005) (1455 2200)) (TUESDAY (1455 2200)) (WEDNESDAY (4C2
1005) (1455 2200)) (THURSDAY (1455 2200)) (FRIDAY (400 2200)))

HSTOFRHW

LIST 0? 8318338038

Barnes, R-H- 1980- Wish—end
W 7th ed. New York: John Wiley a

80113 .

Beach, B. L. and Ostenso, G. L. 1969. "Entree Serving Times:
Relationship of Serving Time to System Capacity.”
W 54(4): 290-296.

Block, Arla A., Roach, Faith R., and Konz, Stephan A. 1985.
"Occurrence Sampling in a Residence Hall Foodservice:
Cleaning Times for Selected Vegetables."lg1,i_rna_l_g_f_1hg
MW 85(2): 206-209-

Boehm. Barry W- et a1. 1978. W
mum New York: North-Holland Publishing Company.

Boritz, Efrim J. and Brown, Donald A. 1986. "The Expert-Ease
of Expert Systems." W 119(8): 50-56.

Breuker, Joost and Wielinga, Bob. 1987. "Use of Models in the

Interpretation of Verbal Data.‘I WW1 .
W Ed. Kidd, Alison L. New York:

Plenum Press.

Briggs, Steven and Doney, Lloyd D. 1988. "Can HR Expertise Be
Computerized ?" 23W 67(4): 106-111.

Britton, Scott G. 1987. "Computer-Based Expert System Aids
Underground Mine Planning." _§Qal_Agg& 92(1): 69-70.

Brown, David C. "Routine Design Problem Solving."1987. BBS

W Ed- Gero, J- Reading
(PA): Addison-Wesley.

Brown, David C. and Chandrasekaran, B. 1989. mm

-_,v . t es a- - . t .te-'-s. San
Marcos (CA): Pitman Publisher.

136

137

Buchanan, B. et al. 1983. "Constructing an Expert Systems."
asildins_Exnert_§Y§tenel Ed- Hayes-Roth. F.. Waterman.
Donald A., and Lenat D. Reading (PA): Addison-Wesley.

Burns, R.N. and Carter M.W. 1985. "Work Force Size and
Single Shift Schedules With Variable Demands.’I

Management_§21encel 31(5): 599-607-

Chandrasekaran, B. 1983 . "Towards A Taxonomy of Problem

Solving Types." W (Winter/Spring): 9-17.

Chandrasekaran, B. 1983. "On Evaluation AI Systems for
Medical Diagnosis." W (Summer):- 34-37.

Chandrasekaran, B. 1986. "Generic Tasks in Knowledge-Based
Reasoning: High Level Building Blocks for Expert System

Design." W (Fall): 23-30.

Currie, 12.11. 1977. 379;); Study, 4th ed. London (UK):
Pitman Publishing.

Curtis, John W. 1987. "Robotics and Artificial Intelligence
for the Food Industry." W 41(12): 62-64.

deHaas, Pat. 1983. “Computerizing the Reference Desk
Schedule." W 108(5): 973-975.

Desai, Harsha B. et al. 1987. "Scheduling of Professional
Staff: Microcomputer Integration." W 57(1):
109-116.

DHFS 1988. (The Division of Housing and Food Service, Michigan
State University, East Lansing, MI 48824) "How to
Schedule." Summer.

Fensholt, Carol. 1988. "Labor Scheduling: New Software Enhance

savings and Service." Supermarket_nusines§1. 43(5): 55-
59.

Firdman, H. E. 1986. "Artificial Intelligence: Understanding
the Basic Concepts." W 42(3): 89-95.

Gaschnig, J. 1982. "Application of the PROSPECTOR System to

Geological Exploration Problems." W
10: 301-323.

Gaschnig, J. et a1. 1983. "Evaluation of Expert Systems:
Issues and Case Studies." Bgilging_§xp§;§_§y§§em§;_ Ed.
Hayes-Roth, F., Waterman, Donald. A., and Lenat D.
Reading (PA): Addison-Wesley.

138

Geissman, James R. and Schultz, Roger D. 1988. "Verification
and Validation of Expert Systems." mm 3(2): 26-
33.

Gevarter, William B. 1987. "The Nature and Evaluation of
Commercial Expert System Building Tools. "99mm (May) :
24-40.

Hayes-Roth, F., Waterman, D. A., and Lenat D. 1983. W
W Reading (PA): Addison-Wesley.

Harrison, Patrick R. 1989. "Testing and Evaluation of
Knowledge-Based Systems." t e
Ed. Liebowitz, Jay and De Salvo, Daniel A. Englewood
Cliffs (NJ): Yourdon Press.

Ho, A. and Matthews, M.E. 1978. ”Activity Sampling in Two
Nursing Home Foodservice Systems." W

MW 73(12): 647-653-

Krebs, Valdis E. 1987. "How Artificial Intelligence Assists
HR Managers." W 64(9): 58-62.

Kern, J ., Ila-user, R., and Sticklen Jon. 1989. "Configuration
and Scheduling: A Preliminary Report on Using A Routine

Design Approach." e d' o rs a
W (Grand RapidS. MI) .

Kidd. Alison L- Ed. 1987. W
M New York: Plenum Press.

Kinoglu, Faith. 1986. "Artificial Intelligence: Expert

System Model of the Design Process." Design News. 42(3):
149-156.

Kuipers, Benjamin and Kassirer Jerome P. 1987. "Knowledge
Acquisition by Analysis of Verbatim Protocols."mgw1§_dgg

W Ed. Kidd Alison L- New
York: Plenum Press.

Liebowitz, Jay. 1986. "Useful Approach for Evaluating Expert
Systems." Expert Systems, 3(2): 86-96.

Liebowitz, Jay. 1989. "Problem selection for Expert Systems
Development." St ctu S s e 5 Ed.
Liebowitz, Jay and De Salvo, Daniel A. Englewood Cliffs
(NJ): Yourdon Press.

Lippert, Renate. 1987. "Expert Systems: New Instructional
Tools." A C N v. (San Francisco, CA).

 

. 139

Matthews, M.E., Zardain, M.S., and Mahaffey, M.J. 1986.
”Labor Time Spent in Foodservice Activities in One

Hospital: A 12-Year Profile." W
W 86(5): 636-643-

Mandel “-3- 1978- Mime—Maxine
W 5th ed. Englewood Cliffs (NJ): Prentice-
Hall.

Miller, F.D. et a1. 1985. "The ACE Experiment: Initial
Evaluation of an Expert System for Preventive
Maintenance . " ' ‘

Ed. Richardson, J.J. Park Ridge (NJ): Noyes
Publications.

Naughton, James M. 1989. "A Specific Knowledge Acquisition
Methodology for Expert Systems Development: A Brief Look
at Precision Knowledge Acquisition."

Ed. Liebowitz, Jay and De Salvo, Daniel A.
Englewood Cliffs (NJ): Yourdon Press.

"Computer Solves Labor Scheduling Mess.” 1986. W

W 62(9): 148-152-
Olsen, Michael D. and Meyer, Mary Kay. 1987. "Current

Perspectives on Productivity in Food Service and

Suggestions for the Future." Wise
W 11(2): 87-92-

Otten, Alan L. 1990. "Part-Time Work Is Fine, If It Is What

You Want." Tbe_flall_§treet_leurna11 (April): Bl.
Prerau, David S. 1985. "Selection of an Appropriate Domain
for an Expert System." W (Summer) . 26-
41.
Rauch-Hindin, Wendy B. 1988. e
W Englewood Cliffs (NJ): Prentice-Hall.
pp 1-110.

Rich. Elaine. 1983. AW New York:
McGraw-Hill, Inc.

Roberts, Hilary. 1988. "Expert Systems and the Personnel

Department." Personae; Menegemept, 20(6): 52-56.
Rodammer, Frederick A. and White, K. Preston. 1988. "A Recent
Survey of Production Scheduling." c '

SYSLQRS. Mag, egg beerneeies, 18(6): 841-851.

140

Rothenberg, Jeff et a1. 1987.

Exeluatins.)baauda_§xsten
Wm Santa Monica (CA):
RAND Corporation. ‘

Slagle, James. and Wick, Michael. 1988. "A Method for
Evaluating Candidate Expert System Applications." The

AI_Mcsazine1 9(3): 44-53.

Stinson, Joel P. and Guley, Helen M. 1982. "Use of a Branch
and Bound Algorithm to Schedule Food Production in a
Semi-Conventional Foodservice System." 0

Ameri2an_Dieteti__A§seciatienl 81(11): 562-567-

Taylor, W.A. 1986. "Artificial Intelligence: Potentials and
Limitations." Deeign_3egee 42(3): 75-83.

Ten Dyke, Richard P. 1986. ”Artificial Intell'igence:

Integrating Expert Systems." W 42(3): 131-
133.

The CBORD Group, Inc. 1984. Labor Scheduling System: for IMB
PS/2, PC XT, AT. Ithaca, NY.

Turpin, Bill. 1986. "Artificial Intelligence: Project
Needs." W 42(3):104-113.

Vedder, Richard. 1987. "The Opportunities and Challenges of
Managing Expert Systems." m (January): 14-16.

Waterman, Donald A. 1986. W Reading
(PA): Addison-Wesley. -

Waterman, Donald A. and Peterson, Mark A. 1984. "Evaluating
Civil Claims: an Expert Systems Approach.” Expect
Systems; 1(1): 65-76.

Williams, J. E. and Donaldson, B. 1969. "SCORE: A Management
Evaluation Program for Dietary Departments."J_Q_ur_n_a_l_91
Tne_Anerican_ni_tetic_Asse_iatien1 54(4): 283-289-

Zemel, P. C. and Matthews, M.E. 1982. "Determining Labor
Productivity Time for Roast Entree in Hospital
Foodservice . " QQQIEQLWD—M e

Aesecietien. 81(12): 709-714.

 

1111111111111