L4 . I .. I22... .4 .x . rs: .4 . Ill, ‘I uélfil: - ~ - c}!!! ‘ I .‘vt 1| IL.PJ .‘ Jul ll?! . x 7"“; In?" ‘94” ..h . will: 1‘ - I'M”; ghftfvo I 3%)???)ng to . -II’II...‘ Iii!) » .h ’(‘vt1.'!(|- ll fl] ll), . ll . ‘ Ill «11123151: . \I‘I. £121]. .bfll. {111? -Irf titan. 1%,”) .4“! “50;" mnugll 4 II, V phi? Alta. ll :r‘ v ‘fllvflxuoai 1.1).. 4 -‘Iilz’lluu “E;’;‘all pl '1} . églfiw I l l\ t . nor! ii ’l-nu . f '71! fig .1 l:i:1lll1n ul.h”:.l\1. i . 1-..... t: ’n‘. .199)- I It'll" p1,}; 0 2.!!! Half... it}!!! |r Itizi“! {I}, w l .I l’f‘t'll .III‘I u- “I ‘l f I!" IL». 0 fl D O I ' 5.! . ‘ o | V Ill .‘ I ‘- .I hufllflfl. 5'- h I 1- {will} IIIIIII I!!! l: ”III‘ 0 gull-I‘ll, I . {JIFIhfi'l‘giw [Ill 1.10 .1 [lurk n El}: [filiagilzz '0 . 11‘1"- . t I yuliqti 7).. 4'" i I I! . . 4! iv?!” \(Il ‘c’ul flamingofli. I. 'a {RED ,1! ‘ 1-1:), l..|rll.....v.0laxh.fli«l.bbfli lllvyfilr’h‘h‘i .11 . I I J . pr 0:1...411: ’Jnlwuiyssththllhl’u 1’. ‘ It». )l’u Q | ‘ 1. («lull .r vnlllll . v , L y .11. (VP'JIII ’ “o u. .. » v‘ .‘I ‘ y i. ’i, 4 :11: THE-353 m? 51"?“th uofife Um‘éverafity ,i This is to certify that the dissertation entitled Reflectance Near Infrared Analysis of Dairy Products and Air-classified Bean Flour presented by Khalil I; Ereifej has been accepted towards fulfillment of the requirements for Ph. D. degree in Food Science -2114éév iézaéigzlaé Major professor Date / _ {'9' f 1: / MS U i: an Affirmative Action/Equal Opportunity Institution 0-12771 MSU LIBRARIES —:I_. RETURNING MATERIALS: Place in book drop to remove this checkout from your record. FINES will be charged if book is returned after the date stamped below. REFLECTANCE NEAR INFRARED ANALYSIS OF DAIRY PRODUCTS AND AIR-CLASSIFIED BEAN FLOUR By Khalil I. Ereifej A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Food Science and Human Nutrition I982 pa “Q Cd l 7 CF} ' ABSTRACT REFLECTANCE NEAR INFRARED ANALYSIS OF DAIRY PRODUCTS AND AIR-CLASSIFIED BEAN FLOUR By Khalil I. Ereifej Low-moisture dairy products and air-classified bean powders were analyzed by conventional methods and by near infrared reflectance (NIR). For 41 spray-dried milk samples prepared by premixing whole milk with skim milk at various proportions, the correlation coefficients (r) between the data of the two analytical procedures were 0.7889, 0.9672, 0.9892, 0:9437 and 0.9528 for moisture, protein, fat, ash and lactose contents, respectively. For six random spray-dried milk samples the corresponding r values were 0.8384, 0.9282, 0.9908, 0.9886 and 0.9882, respectively. The r values of 3l samples prepared by mixing whole milk powder with skim milk powder were 0.8208, 0.9844, 0.9956, 0.9846 and 0.9959 for moisture, protein, fat, ash and lactose, respectively. For 6 un- known samples the r values were 0.60l7, 0.9945, 0.9514, 0.8625 and 0.9628 for moisture, protein, fat, ash and lactose contents, respec- tively. Eleven commercially dehydrated milk powders had r values 0.8507, 0.9878, 0.9992, 0.9949 and 0.9888 for moisture, protein, fat, ash and lactose contents, respectively. For 40 commercial cheese powder samples the r values were 0.7494, 0.9352, 0.9803 and 0.7486 for moisture, protein, fat and ash contents, respectively, Thirty air-classified bean flour samples were analyzed for moisture, protein, and ash contents using conventional methods and NIR. The r values were 0.8971, 0.9878 and 0.957l for moisture, protein and ash contents, respectively. Six unknown samples had the corresponding r values: 0.7467, 0.9394 and 0.9699, respectively. F-values obtained from the analysis of variance of conventional and NIR data for each food components were smaller than F-table values in all cases except for the moisture content of the random spray-dried milk samples. The F-values indicated that there was no significant statistical difference between NIR data and that obtained by the con- ventional procedures. The NIR method is rapid (30 seconds), involves little sample pre- paration, and provides a direct read out. The test is not destructive and requires 8-l2 9 sample. Accurate calibration of the NIR instrument with a reliable reference method of analysis is necessary for consistent and meaningful data. T0 My Mother, Noura and My Sister Hilala ii ACKNOWLEDGEMENTS The author would like to express his sincere appreciation and thanks to: Dr. Pericles Markakis, his major professor, for his expert, profess- ional guidance, encouragement, and understanding during the research for this project and in the preparation of the manuscript. Dr. M.A. Uebersax, for providing the bean flours, his advice and help in preparation of this manuscript. Dr. R. Chandan, for his help in preparation of the spray- dried milk samples and reviewing this manuscript. Dr. E. Beneke, for his advice and aid in preparation of the manuscript. Dr. D. Reicosky, for his help in using the GQA-4l, valuable discussion, suggestions and his help in reviewing this manuscript. Sincere appreciation is extended to Dr. J. Gill, Dr. R.C. Nicholas and Talal Hussein for their assistance in the statistical analysis. Special thanks are extended to Dr. Judy Kintner of Commercial Creamary, Spokane, Washington for supplying the cheese samples, to J. Partridge, Nayini, Narsimha and R. Blakeny for their technical help and encouragement, and to Mrs. G. Markakis (from the Michigan Dept. of Agriculture) for providing the commercially dehydrated milk samples. Sincere appreciations and thanks also are extended to Dickey-john Corp., Auburn, IL, for their partial financial support to me as a graduate research assistant. The author wishes to extend his most sincere gratitude and indebt- edness to his mother and sister for their patience and moral support, and all brothers and sisters, especially his brothers, Hilal and Sami for their assistance and encouragement throughout the entire academic study. iv TABLE OF CONTENTS Page DEDICATION ................................................ 11 ACKNOWLEDGEMENTS ......................................... 111 LIST OF TABLES ........................................... vi LIST OF FIGURES .......................................... x INTRODUCTION ............................................. 1 REVIEW OF LITERATURE ------------------------------------- 3 METHODS AND MATERIALS ------------------------------------ 9 A. Sample preparation ----------------------------- 9 Spray-dried milk ------------------------------- 9 Dry-mixed milk --------------------------------- 10 Cheese powder ---------------------------------- 1o Air-classified bean flour ---------------------- 10 Commercial dry milk ---------------------------- l0 8. Sample analysis -------------------------------- ll Nitrogen determination by the micro-Kjeldahl method --------------------------------------- 11 Fat Determination .............................. 12 Instrumentation -------------------------------- 12 C. Instrument Calibration --------------------------- 13 0. Statistical analysis --------------------------- 16 RESULTS AND DISCUSSION ----------------------------------- l7 SUMMARY .................................................. 59 REFERENCES ............................................... 71 Table Table Table Table Table Table Table Table Table Table Table Table Table 10. ll. 12. l3. LIST OF TABLES Analysis of spray dried milk by conventional and NIR procedures. ------------------------------------- Analysis of dry-mixed milk powders by conventional and NIR procedures. -------------------------------- Analysis of cheese powders by conventional and NIR procedures. ........................................ Analysis of air-classified bean flour by convention- al and NIR procedures. ............................. The best four reflectance pulse points selected by the regression analysis for dehydrated milk, cheese and bean flour. ------------------------------------ Multilinear regression coefficients (K values) used for NIR analysis of milk powder, cheese and bean flour. --------------------------------------------- Mean and range of the content in several constituents of foods analyzed by conventional methods and subse- quently subjected to NIR analysis. ----------------- Analysis of variance of moisture content of spray dried milk analyzed by conventional and NIR methods. Analysis of variance of fat content of spray dried milk. .............................................. Analysis of variance of fat content of spray dried milk. .............................................. Analysis of variance of ash content of spray dried milk. .............................................. Analysis of variance of lactose content of spray dried milk. ---------------------------------------- Analysis of variance of moisture content of dry- mixed milk analyzed by conventional and NIR methods. vi Page 18 I9 20 2l 22 23 25 46 46 46 47 47 47 LIST OF TABLES (continued) Table 14. Table Table Table Table Table Table Table Table Table Table Table Table Table Table 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. Analysis of variance of protein content of dry- mixed milk analyzed by conventional and NIR methods. Analysis of variance of fat content of dry—mixed milk analyzed by conventional and NIR methods. ----- Analysis of variance of ash content of dry-mixed milk analyzed by conventional and NIR methods. ----- Analysis of variance of lactose content of dry- mixed milk analyzed by conventional and NIR methods. Analysis of variance of moisture content of cheese powder analyzed by conventional and NIR methods. --- Analysis of variance of protein content of cheese powder analyzed by conventional and NIR methods. —-- Analysis of variance of fat content of cheese powder analyzed by conventional and NIR methods. --- Analysis of variance of ash content of cheese powder analyzed by conventional and NIR methods. ---------- Analysis of variance of moisture content of air- classified bean flour analyzed by conventional and NIR methods. ................................... Analysis of variance of protein content of air- classified bean flour analyzed by conventional and NIR methods. ----------------------------------- Analysis of variance of ash content of air- classified bean flour analyzed by conventional and NIR methods. ----------------------------------- Analysis of spray dried milk, dry-mixed milk and air-classified bean flour unknown samples by con- ventional and NIR methods. ------------------------- Number of samples, correlation coefficient (r), slopes and Y-intercepts of the linear regression of unknown samples analyzed by conventional and NIR methods. --------------------------------------- Analysis of variance of moisture content of spray- dried milk unknown samples analyzed by conventional and NIR methods. ................................... Analysis of variance of protein content of spray dried milk unknown samples analyzed by conven- tional and NIR methods. ---------------------------- vii Page 48 48 48 49 49 49 50 50 50 51 51 44 45 52 52 LIST OF TABLES (continued) Page Table 29. Table Table Table Table Table Table Table Table Table Table Table Table 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41. Analysis of variance of fat content of spray dried milk unknown samples analyzed by conven— tional and NIR methods. ------------------------------ 52 Analysis of variance of ash content of spray dried milk unknown samples analyzed by conventional and NIR methods. ----------------------------------------- 53 Analysis of variance of lactose content of spray dried milk unknown samples calculated by difference and predicted by NIR method. ------------------------- 53 Analysis of variance of moisture content of dry-mixed milk unknown samples analyzed by conventional and NIR methods. --------------------------------------------- 53 Analysis of variance of protein content of dry-mixed milk unknown samples analyzed by conventional and NIR methods. ---------------------------------------- 54 Analysis of variance of fat content of dried-mixed milk unknown samples analyzed by conventional and NIR methods. ----------------------------------------- 54 Analysis of variance of ash content of dry-mixed milk unknown samples analyzed by conventional and NIR methods. ----------------------------------------- 54 Analysis of variance of lactose content of dry-mixed milk unknown samples calculated by difference and predicted by NIR methods. ---------------------------- 55 Analysis of variance of moisture content of air- classified bean flour unknown samples analyzed by conventional and NIR methods. ------------------------ 55 Analysis of variance of protein content of air- classified bean flour unknown samples analyzed by conventional and NIR methods. ........................ 55 Analysis of variance of ash content of air-classified bean flour unknown samples analyzed by conventional and NIR methods. ------------------------------------- 56 Analysis of commerCial navy bean air-classified flour by conventional and NIR methods before and after grinding for 2 minutes. ------------------------------ 58 Analysis of commercially dried milk samples by con— ventional and NIR methods. ------------------- . ....... 60 viii LIST OF TABLES (continued) Table 42. Table 43. Table 44. Table 45. Table 46. Table 47. Table 48. Table 49. The best four reflectance pulse points selected by regression analysis for commercially dried milk. ---- Multilinear regression coefficients (K values) used for NIR analysis of commercially dried milk. -------- Number of samples (n), correlation coefficients (r), slopes, y-intercepts, and the equation, of the linear regression between the conventional and NIR values of commercially dried milk. ------------------ Analysis of variance of moisture content of commer- cially dried milk analyzed by conventional and NIR methods. -------------------------------------------- Analysis of variance of protein content of commer- cially dried milk analyzed by conventional and NIR methods. -------------------------------------------- Analysis of variance of fat content of commercially dried milk analyzed by conventional and NIR methods. Analysis of variance of ash content of commer- cially dried milk analyzed by conventional and NIR methods. ---------------------------------------- Analysis of variance of lactose content determined by difference and NIR method. ....................... ix Page 61 62 63 64 65 66 67 68 Figure Figure Figure Figure Figure Figure Figure Figure Figure LIST OF FIGURES Page . Relationship between % moisture determined by the oven method and by the NIR method in spray dried milk. Bars show the 95% confidence limits of the linear regression. ---------------------------------- 26 . Relationship between % protein determined by micro- Kieldahl and by the NIR methods in spray dried milk. Bars show the 95% confidence limits of the linear regression. ----------------------------------------- 30 . Relationship between % fat determined by the Mojonnier and by the NIR methods in spray dried milk. Bars show the 95% confidence limits of the linear regression. -- 34 . Relationship between % ash determined by dry ashing and by the NIR methods in spray dried milk. Bars show the 95% confidence limits of the linear regression. ----------------------------------------- 37 . Relationship between % lactose calculated by differ- ence and predicted by NIR method in spray dried milk. Bars show the 95% conficence of the linear regression. 42 . Relationship between % moisture determined by the oven method and by the NIR method in dry-mixed milk. Bars show the 95% confidence limits of the linear regression. ----------------------------------------- 27 . Relationship between % protein determined by micro- Kjeldahl method and by the NIR method in dry-mixed milk. Bars show the 95% confidence limits of the linear regression. ---------------------------------- 3l . Relationship between % fat determined by Mojonnier method and by the NIR method in dry-mixed milk. Bars show the 95% confidence limits of the linear regression. ----------------------------------------- 35 . Relationship between % ash determined by dry ashing method and by the NIR method in dry-mixed milk. Bars show the 95% confidence limits of the linear regression. ----------------------------------------- 38 LIST OF FIGURES (continued) Figure 10. Figure Figure Figure Figure Figure Figure Figure Figure 11. 12. 13. 14. 15. 16. 17. 18. Relationship between % lactose calculated by difference and predicted by the NIR method in dry- mixed milk. Bars show the 95% confidence limits of the linear regression. ------------------------- Relationship between % moisture determined by the oven method and by the NIR method in cheese powder. Bars show the 95% confidence limits of the linear regression. --------------------------------------- Relationship between % proteins determined by micro Kjeldahl method and by the NIR method in cheese powder. Bars show the 95% confidence limits of the linear regression. -------------------------------- Relationship between % fat determined by Mojonnier method and by the NIR method in cheese powder. Bars show the 95% confidence limits of the linear re- gression. ----------------------------------------- Relationship between % ash obtained by dry ashing method and by the NIR in cheese powder. Bars show the 95% confidence limits of the linear regression. Relationship between % moisture determined by the oven method and by the NIR in air-classified bean powder. Bars show the 95% confidence limits of the linear regression. -------------------------------- Relationship between % proteins determined by micro Kjeldahl and by the NIR in air-classified bean powder. Bars show the 95% confidence limits of the linear regression. -------------------------------- Relationship between % ash determined by the dry ashing and by the NIR methods in air-classified bean powder. Bars show the 95% confidence limits of the linear regression. ------------------------- Near infrared reflectance instrument used in this research. ......................................... xi Page 43 39 INTRODUCTION Near infrared reflectance spectroscopy has been developed by Norris and Hart (1965), to measure the moisture content of grains and oilseeds. The technique was later expanded to measure other food constituents, such as protein, oil, starch, sugar and fiber. ,The measuring system consists of interference filters to isolate selected wavebands of near infrared energy, a photosensor, a signal conditioning amplifier, and a computer for collecting and analyzing the data. In 1971, NIR was introduced to the grain industry as a means of rapid analysis for moisture, oil and protein (Rosenthal, 1971). At least three instrument manufacturers, Neotec, Technicon and Dickey-john, made available instruments of increasing sofistication for the analysis of grains and oilseeds. In 1978, NIR was introduced to the baking industries in the United States, as a high speed analytical technique of flour and a quality con- trol tool. In the last fifteen years, many workers used the NIR technique to estimate the concentrations of several constituents of agricultural pro- ducts (Norris et_al,, l976, Stermer gt_al, 1977, Watson gt_al,, 1976, Williams and Starkey, 1980). The work reported here describes an effort to use NIR for the deter- mination of moisture, protein, fat, ash and lactose contents in low- 2 moisture dairy products and moisture, protein and ash contents in air- classified bean flours. REVIEW OF LITERATURE Norris and Hart (1965), investigated the water absorption bands at 0.76, 0.97, 1.18, 1.45 and 1.94 n for the spectrophotometric measure- ment of water in seeds and grains. These efforts led to the develop- ment of a near infrared reflectance (NIR) technique, which originally aimed at determining the moisture content of agricultural products, but later was expanded to the measurement of protein, oil, starch and other constituents in these products. Hymowitz et_al_(l974), estimated the protein and oil content in corn, soybean, and oat seeds by NIR using the grain analyzer manufac- tured by Dickey-john Corp. These authors also studied the effect of sample grinding time on the grain analyzer readings for protein and oil content. Their most important findings were: a) the high correlation between the NIR and Kjeldahl data for protein content, and; b) the lack of statistical significance between grinding time and protein or oil content estimates. Pomeranz and Moore (1975), compared six methods of protein deter- mination: Kjeldahl, biuret, dye binding, alkaline distilation, NIR by GAG-2 (Dickey-john) and NIR by GQA (Neotec). They reported a high correlation by all the six methods for protein determination in wheat. Williams (1975) applied NIR to the analysis of several cereal grains and oil seeds including hard red spring wheat, hard and soft wheat flour, barley, oats, rapeseeds and soybeans. Protein content 3 4 was determined by Kjeldahl and by two different NIR instruments. He concluded that, the introduction of NIR instrumentation represents a rapid routine analysis of cereal grains and oil seeds for oil, protein and moisture, and that sample preparation and accurate calibration of the NIR instrument are very important. Watson gt_al, (1976) evaluated the GQA-l and grain analyzer com- puter (GAC) for protein determination in hard red winter wheat. They reported correlation coefficients of 0.979 with GQA and 0.982 with GAC for protein estimation between Kjeldahl and NIR. GQA was easier to calibrate and operate than the GAC. Watson gt_al, (1977) developed a regression equation for protein determination by Kjeldahl and NIR using five classes of wheat. It was found that the slope of the regression equation depended on the wheat class and that the effect of wheat class on the regression equation was not related to the particle size distribution of the ground sample. Stermer gt 31, (1977) attempted to estimate the moisture of whole grain corn and sorghum using the neotec GOA-41 instrument. The corre- lation coefficients between the NIR values and those by the oven method were 0.997 for whole corn and 0.974 for whole grain sorghum. Rubenthaler and Bruinsma (1978) estimated the amino acid lysine in wheat samples using NIR. They used the Technicon Infraalyzer inter- faced with the Hewlett-Packard 9815A programmable calculator, and found a high correlation between the lysine values predicted by NIR with those determined by amino acid analysis. Rosenthal (1978) reported the absorption spectra of various food components with the emphasis on the best absorption wave lengths. In addition to agricultural products he noted a wide scope for the use of 5 NIR in industrial and clinical applications. Several commercial models of NIR instruments are manufactured by Dickey-john, Neotec and Technicon Corporations. The Neotec Feed Quality Analyzer model 51 provides optical data at a greater number of wave lengths and directly prints percentage of pro- tein, oil, moisture and fiber. Dairy products have become an important part of American diet. In these products the amount of the yield is related to the total solid content of fluid milk. Hence the need for rapid and accurate measurement of fluid milk constituents is essential. Hooton (1978) emphasized the versatility of NIR in several busi- ness areas as receiving, manufacturing, shipping, laboratory and inven- tory. 0n the other hand, he pointed out several industrial areas as grain handling, milling, mixed feeds, dry corn milling, and residual oil in corn germ cake, where NIR rapid analyses are very important. Norris (1978) showed that NIR measurements are not limited by the instrument noise alone. Sample preparation is the greatest source of variability and grinding is important. Williams and Thompson (1978), investigated the effect of granularity of the hard red spring wheat using NIR analysis for protein and moisture content and found several factors effecting granularity, such as genetic constitution, chromosome number, wheat type, variety, soil, grinder and grinding conditions as well as sample size. The researchers recommended the following steps to achieve accurate results: 1) Use of high speed hammer mill for sample preparation. 2) Optimum mean particle size for NIR analysis of hard red spring wheat is between 180-220 u. 3) At least fifty samples to be used for calibration. 4) A wide range of the content of food constituent is necessary for calibration of the instrument. 6 Williams gt_al, (1978) applied the NIR technique for protein and moisture testing in pulse breeding programs to improve both yield and quality of pulses in the dry and tropic areas. Pulses were obtained from different research institutes all over the world and were analyzed for protein content by Kjeldahl and were subsequently analyzed by NIR using the Neotec Quality Grain Analyzer model 31. They found correlation co- efficients from 0.89 to 0.96 between Kjeldahl and NIR. Miller gt_al, (1978) investigating the protein content in hard red winter wheat samples found excellent reproducibility in the values ob~ tained by NIR. Williams (1979) investigating the possibility of screening wheat for protein content and hardness, used two sets of wheat samples which varied in Kjeldahl protein content and hardness. Hardness was assessed by the particle size index (PSI) test. All wheats were ground with a Burr mill and an Impeller-type mill, and passed through a 1.0 mm screen. Neotec GOA-31 was calibrated against protein content and PSI with the Burr mill and for protein with the Impeller-milled samples. He reported that protein content was predictable to within 0.31% in the Impeller- ground wheat samples and within to 0.70% in the Burr milled samples, and the PSI was predicted to within less than two units. He further analyzed five classes of hard wheat by one calibration and another five soft wheat classes by different calibration and showed that analysis of hard wheat was more accurate than that of the soft wheat. Birth (1979) reviewed the measurement of food quality by radio- metric methods. He concluded that the evaluation of any new method relays on statistical analysis and found that spectral reflectance or transmittance data generated by computation of derivative spectra can 7 be analyzed by multiple linear regression to develop the best equation for food quality prediction. Giangiacomo gt_al, (1981) employed the NIR technique to measure the concentration of fructose, glucose and sucrose in model systems. Subsequently, measurements were made to estimate the same sugars in dried apple tissues. The correlation coefficients of the actual concentrations versus the predicted values were 0.995, 0.994 and 0.986 for fructose, glucose and sucrose, respectively in the model systems comprising 20 samples. But when they tried to use the prediction equations to esti- mate the sugars in the dried apple samples, the corresponding correla- tion coefficients were 0.70, 0.55 and 0.90 for fructose, glucose and sucrose, respectively.' Fernandez (1981) using the GOA-41 estimated the nitrogen content of several dried tissues of bean plants (seed, pod wall, leaf blade, petiole, steam and root) with a correlation coefficient ranging from 0.873 to 0.973, between Kjeldahl and NIR. Shenk gt_al, (1981) using the spectra-computer Neotec 6100 evalu- ated the acid fiber, neutral fiber, lignin, cellulose, Ca, P and K in forage. In addition, 90 samples of Canadian wheat were also evaluated for their protein content. They concluded that this instrument provides an acceptable means of evaluating forages and grains. The accuracy of NIR depends on the successful completion of the following: 1. Selection of a representative set of samples from the popu- lation. 2. Accurate laboratory analysis of the quality parameters of interest. 3. Accurate NIR data. 4. 5. 8 Appropriate transformations of the NIR data for each quality parameter to be predicted. Appropriate wavelengths for the whole population. Chief sources of error in near infrared reflectance testing in- clude: 1. 2. NONCD-hw Selection of calibration samples. Accuracy of standard chemical analysis used in calibration or monitoring. Particle size and particle size distribution. Homogeneity of ground sample. Moisture status of samples. Sample storage. Uneven or inconsistent loading of cell. METHODS AND MATERIALS Spray-dried milks, milk powders prepared by dry-mixing, cheese powders, commercially dried milk and air-classified bean flours were used in this research. All samples were analyzed in duplicates. The data are reported on dry weight basis for air-classified bean powder and on "as is" basis for dairy products. A. Sample Preparation. _§pray-dried milk Forty one 3-liter samples were prepared by mixing 0+3000, 75 + 2925, 150 + 2850 ...... and 3000 ml of whole pasteurized milk (3.06% fat) + 0 ml of skim milk (0.11% fat). The mixtures were stirred, spray-dried and samples of the powders were collected in glass jars, packed in two- layer polyethylene bags, sealed and stored at room temperature until the time of analysis. A spray drier manufactured by Swanson Evaporator Company, Harvey, IL, was used. The inlet air temperature was 350°F, and that of outlet air temperature was 150°F. The atomizer air pressure was 35 PSI and similar air pressure was exerted on the feed tank. The purpose of mixing two kinds of milk was to establish a wide range of fat concentrations which is necessary for calibrating the grain quality analyzer (GOA—41). The forty one spray-dried samples were used for calibrating the instrument. Some other samples were prepared in a similar way and later were used as unknowns. lO Dry-mixed milk samples Thirty one samples were prepared by mixing 0, l, 2, 3....and 30 g of commercially spray—dried whole milk (Valley Lea Dairies, Inc., South Bend, Indiana) with 30, 29, 28, 27......and 0 g non-fat dry milk (Golden Guernsey Dairy, Sparta, WI). These samples were stored in glass jars until the proximate analysis were performed. The purpose of mixing the two kinds of dry milk powder was to obtain samples with a wide range of concentrations for calibrating the GOA—41 instrument. Some unknown samples were prepared in a similar way. Cheese_powder Forty samples of cheese powder were provided by Commercial Creamery Co., Spokane, WA. These cheese powders were representative of the powders sold in the industry for snack seasoning. All samples were used to calibrate the GOA-41 instrument. Air-classified bean flour Thirty air-classified bean flour samples were provided by Dr. M. Uebersax (Food Science Department) which were used to calibrate the instrument. Several other samples were used as unknowns. Also three commercial samples were analyzed as unknowns. The samples were kept in polyethylene bags until they were analyzed. Commercial dry milk Eleven commercially dried skim and whole milk were provided by the Michigan State Department of Agriculture, Laboratory Division. Two milk samples were bought from the market in E. Lansing area and were used as unknowns. All samples except the bean flour were ground for two minutes in a Mitey-mill, a high speed rotating blade type. The powders were passed 11 through a 100 mesh sieve to enforce a uniformed particle size, which was later packed and used for calibrating the GOA-41 instrument and proximate analysis. B. Samplg_ana1ysis Moisture and ash contents of all samples were determined by the oven method and dry ashing respectively according to AOAC (1975). Lac- tose content of the milk powders were calculated by difference. Nitrogen determination by the micro-Kjeldahl method Approximately 50 mg of each sample were digested for one hour in duplicate according to AOAC (1975). Sulfuric acid of 1.84 specific gravity was used for digestion. Potassium sulfate and mercuric oxide were added as catalysts. After cooling the flasks, the sides were rinsed with deionized water and digestion continued for another hour. The digests were transferred into the distillation apparatus by using approximately 10 ml deionized water. The diagested mixture was neutralized with 15 ml of 50% NaOH solution, containing 5% sodium thio- sulfate. The liberated ammonia was steam-distilled into 5 ml of 5% boric acid solution, containing 4 drops of methyl red-methylene blue indicator (2 parts of 0.2% methyl red in alcohol with one part of 0.2% methylene blue in alcohol). The distillation was continued until the volume in the receiving flask reached 25 ml. The ammonium borate complex was titrated with 0.02 N HCl which had been accurately standard- ized against tris-hydroxy amino methane (THAM) as a primary standard. Nitrogen was calculated from the following formula: % . (ml “Cl-ml blank) (normality of HCl) (14.007) X 100 mg of sample N 12 Protein content was calculated as follows: % Protein = %N x 6.25 for bean flour. % Protein - %N x 6.38 for dairy products. Fat determination Approximately 2 g of the dairy product samples were transferred carefully to a Mojonnier flask and analyzed according to Mojonnier (1925). The sample was mixed with 8.5 ml deionized water. Two m1 of ammonium hydroxide was added to neutralize the acidity, followed by adding 10 ml of ethyl alcohol. The fat was extracted with 25 ml anhydrous diethyl- ether, flasks were stoppered with a rubber stopper and shaken vigor- ously for 2 minutes. About 25 ml of petroleum ether was added to the mixture, followed by shaking for 2 minutes, and the flasks were set aside to assure complete separation of the two layers. The upper layer was carefully poured into previously pre-weighed can, second extraction was performed by adding 5 ml ethyl alcohol, 25 ml ethyl ether and 25 ml petroleum ether to the flask containing the sample mixture. The upper layer was combined with the previous extract. The fat extractant was evaporated by placing the can on a steam bath. Further, the cans were dried in the vaccum oven at 100°C for 20 minutes, followed by cooling for 5 minutes in a desiccator. The cans were re-weighed and the fat content was calculated. Instrumentation The Neotec Grain Quality Analyzer model 41 (GQA—4l) interfaced with a teletype computer that could be connected with the main computer at Michigan State University was used in this research. The GOA-41 is built around a Carry model 14 prism—grating infrared monochromator. The sample is packed into the sample cell and a quartz 13 glass window covers the smooth surface of the sample. A spring-loaded pressure plate is used to hold the sample smoothly against the glass window. The sample is illuminated through the window and the diffused re- flected light is collected with four lead sulfide cells placed at equal distances around the sample. A solid teflon plate is used as a reflec- tance reference. The signal from the detectors is fed to the computer after it has been amplified with a logarithmic response amplifier which is digitized. The sample is scanned in the useful range of the infrared, 1.2 - 2.5 u at a scanning speed of 10 nm per second and the reflectance values are recorded as dR/R, where dR is the differential coefficient of a line tangential to the absorption curve peak, R is the absolute reflectance. A change in a food constituent concentrations results in a change in the corresponding dR. This mathematical model is referred to as dR/R. C. Instrument Calibration The GOA-41 (Figure 18) was calibrated for moisture, protein, fat, ash and lactose (by difference) content of the dairy products under study. Similarly, the instrument was calibrated for moisture, protein, and ash content of the air-classified bean flour. The calibration was performed as follows: 1. The GOA-41 was connected to the computer to collect and analyze the reflectance measurements. 2. Three samples of one product were selected: among those to be used for calibration of the instrument one had the highest content of the food constituent of interest (moisture for instance), the second sample had the lowest content, the 14 - Infrared lamp Lens Wide band light beam. Tilting filters Photo detector Sample cup i -—— Teflon reference Figure 18. Near infrared reflectance instrument used in this research. 15 third was randomly selected to be in between the two extremes. Samples were placed in the sample cell, introduced to the radiation beam one after another, the reflectance values were amplified, digitized and recorded. Four reflectance values were obtained at each pulse point, where the maximum response of the system takes place at a certain wave length. The 300 reflectance values for each sample were fed to the computer along with the corresponding analytical values for each food constituent. The computer selected four wave lengths (pulse points) which best correlated with the analytical values for each food component. All of the calibration samples were now read by the GOA-41 computer at the four selected pulse points for each food constituent. The computer performed a polynomial regression analysis and provided the regression coefficients, K values, for the following prediction equation: - dR dR dR dR % Food Constituent - K0+K1(R;Y+K2(RE(+K3(R§°+K4(RZI Where: K0 is the intercept of the multilinear regression line. K1, K2, K3 and K4 are the regression coefficients at four different wavelengths dR is the differential co- efficient.R is the absolute reflectance value at a particular wavelength. 16 5. Adjustment of K0 The intercept K0 was adjusted by inserting the pulse points and K values of the particular food component into the GOA-41 computer. All samples were measured again and a reading was recorded as % food constituent. The difference between the mean analytical values and the mean regression values was calculated and was substrated or added to the K0 value. The new calculated K0 was considered the most suit- able value to analyze the product for that food constituent. The new calculated Ko along with the rest of K values and pulse points were inserted into the GOA-41 computer for routine analysis. 0. Statistical analysis The data obtained by conventional analytical procedures and those estimated by NIR were subjected to correlation analysis and analysis of variance according to Cochran (1957). The 95% confidence limit intervals were calculated according to Snedecor (1955). RESULTS AND DISCUSSION The tables 1, 2, 3 and 4, show the moisture, protein, fat and ash contents of the spray-dried milk, dry-mixed milk, cheese powder, and air- classified bean flour respectively by the conventional and NIR procedures. All food constituents were reported on "as is“ except for the air— classified bean flour which was calculated on a dry basis. The lactose content in dairy products was calculated by difference. The analytical values were used to calibrate the GOA-41 and later used to estimate the NIR values, which are shown in the Tables 1, 2, 3, and 4. The best four pulse points which were selected by the regression analysis are shown in Table 5. The P0 is constant and is 80 according to Neotec GOA-41 manual. The P1, P2, P3 and P4 are shown for each food component in all the pro- ducts. The pulse points were selected from the relationship between the reflectance of a particular food constituent and the change in the wave length at the NIR region. To estimate the percentage food constituent content in a sample, the pulse points (AS) and the K values (Table 6) of that particular food component must be fed back to the computer of the GOA—41. The sample is introduced in the sample cup to the NIR light, the instrument will measure the reflectance and the computer will solve the prediction equation for % food constituent, and the result is displayed or recorded. The means and ranges of the food constituent contents analyzed by both 17 18 ..... 22.2 22.22 22..2 22.. 22... 22.2 2..22 .2.22 22.2 .. 22... o... 2...2 .2..2 .2.2 .2.2. 22.2 22.22 22.22 22.. o. 22... 22.2 2..22 2...2 22.. .2... 22.2 22.22 .2..2 .2.. 22 .2... 22.2 22..2 22..2 2..2 22.2. 22.2 22.22 .2..2 .2.. 22 .2... 22.2 .2.22 2...2 22.. .2... 22.. 22.22 22.22 22.. 22 22.2. .2.2 2..22 o...2 .2.. 2..2. .2.2 22.22 22..2 2... .2 22.2. 2... .2..2 2..22 22.. , 22.2. 22.2 22..2 22.22 22.. 22 22.2. .2.2 .2.22 22.22 22.2 .2.2. 22.2 22.22 22.22 22.2 .2 2o... 22.. 22..2 22..2 22.2 22... 22.2 22..2 .2..2 .2.2 22 .2... 2a.. 2..22 2..22 2... 22.2. 22.2 22.22 .2.22 22.. 22 22... 22.2 22.22 2..22 o..2 22.2. 22.2 .2..2 .2.22 22.2 .2 22.2. 22.2 22.22 22.22 22.. 22... 22.2 22..2 22.22 22.. 22 22... 22.2 22.22 22.22 22.. 22... .2.2 ....2 22.22 2... 22 22.2. 2... 22.2. 22.22 .2.. 22.2. 22.2 2...2 .2.22 22.. 22 22.2. 22.2 22.2. 22.22 .2.2 ...2. 22.. 22.22 22.22 2..2 22 22.2. ...2 22.22 2..22 22.. 22... 2... 22.2. 22.22 .2.2 22 22... 22.2 2...2 22.22 22.. 22.2. 2... 2...2 22.22 22.2 22 22.2. 2... 2...2 22.22 ...2 22.2. 22.2 ...22 22.22 22.. .2 .2.2. 2... 22.2. 2..22 .... 22.2. .2.2 22.2. 2..22 22.2 22 22.2. 2... 2..2. 22.22 2..2 .2 2. 22.2 22.2. 22.22 22.. 22 22.2. 2... 2.... .2.22 22.2 22.2. 22.2 2..2. 22.22 .2.2 .2 2..2. 22.2 22... 22.22 o..2 22.2. 22.. .2.2. .2.22 22.2 02 22.2. 22.. 22.2. 22.22 22.2 22.2. .2.2 .2.2. 22.22 22.. 2. 22.2. 22.2 22... 22.22 22.2 .2.2. .2.2 22... 22.22 .2.2 2. ...22 o... 22... 2..22 22.2 22.2. 22.2 .2.2. .2.22 22.2 2. 22.2. 22.2 22.2. 22.22 22.. 2.... .2.2 22... 22.22 2..2 2. 22.2. 2... 22.2. 22.22 2..2 22.2. 22.2 2.... 22.22 2..2 2. .2.2. 22.. ...2. 2..22 22.2 2..22 .2.2 22.2. 22.22 2..2 .. 22.2. .2.2 22... 2..22 .o.. 2..22 22.2 22.2. 22.22 2..2 2. 22.22 22.. .2.2. 22.22 22.2 ....2 22.2 22.2 22.22 22.2 2. 22.. 22... 2..22 22.. 2... 22.2. .2.22 22.. .2 22.2. 22.2 .2... 22.22 .2.. .2.2. 22.2 2.... 22.22 2... o. .2.2. 22.2 22.2. 22.22 22.2 22.2. 22.2 .2.2. 22.22 22.2 2 22.2. 2... 2o... 2..22 22.2 ...2. 2... .2.2. .2.22 22.2 2 22.2. 22.2 2..2. 22.22 22.2 2..2. 22.2 22... 22.22 22.2 2 . 22.2. 22.. 22.2. 22.22 22.2 2.... .2.2 22.2. 22.22 22.. 2 22.22 22.2 22.2 2..22 22.2 22.2. .2.2 22.2 .2.22 .2.. 2 22.2. 22.2 c... 22.22 22.2 22.2. 22.2 22.2. 22.22 .2.2 . 22.22 22.2 22.2 .2.22 a... oo..2 .2.2 .2.2 2o..2 2... 2 ....2 22.2 22.2 ...22 22.2 ...22 22.2 .2.2 22.22 22.2 2 .2... 22.2 22.. 2..22 22.2 .2... 22.2 .o.. .2.22 .2.2 . a easy 2 2095:: 22.22.. 2 .2. 2 2.2 2 2.2.2.2 2 22: 2 . u .2ouwwn 22. .2. 2 2.2 2 2.22.22 2 c2: 2 2.2.22 “—2 -¢°wd80>¢°u .2326322. «:2 .2... .oco.2co>¢ou#a|2:¢ 3.2.2412... so 2.22.223 ._ o2: ..... ....2 ...2. ...2 ..... ..... 22... ...2 .2 ..... .... ....2 ....2 .... ..... .... ....2 2.... ...2 .2 ..... .2.. ....2 2.... ...2 2.... .... .2... ....2 ...2 .2 ...2. .2.. 2.... ....2 ...2 ...2. 2... 22... ....2 .... .2 ..... 2... ...22 2.... 2... ..... .... ...22 .. .2 2... .2 22... .... ....2 ....2 .... 22... 2... ...2. 2.... ...2 .2 ..... .... 22... ....2 .2.2 .2... 2... ....2 ..... ...2 .2 ..... .... 2.... ....2 .2.. ..... 22.. ....2 ....2 .... .2 ..... .... 22.2. 22... ...2 .2... .... 22... ....2 ...2 2. ..... .... ..... ..... ...2 .2... .... ..... 2.... ...2 .2 ..... .... 2.... .2.2. ...2 ..... .... ...2. ....2 .2.2 .. .2... ...2 ..... 22.2. ...2 22... .2.. ...2. ...2. ...2 .2 .2... ...2 ..... .2.2. .... ..... .2.. ..... 2.... .... .. ..... 2... 2.... ....2 ...2 ...2. .... ..... ....2 ...2 .. ...2. ...2 ..... ....2 .... ..... .... 2.... ..... .2.2 2. n: ...2. .... 22.2. ....2 ...2 ...2. .... ..... ....2 .... .. 1. 2.... .... ...2. ....2 2... ..... ...2 2. 2. .2... ...2 .. ..... ...2 2.... 22... 2... .2... .... ..... ....2 .2.2 .. ..... .... .... ....2 .... ..... 2... ..... ....2 .... .. ..... .... ..... ...2 .2... .... ....2 ...2 2. ..... 2..2 .... .2... 2... ..... ...2 2... ..... .... .. ..... ...2 .2.. ....2 ...2 ..... .... .... .2... ...2 .. ..... ...2 .2.. .2... ...2 2.... .... ...2 .2... ...2 . ..... .... .... .. .2 ...2 ..... .2.. .... ....2 2... . ..... ...2 22.. ..... ...2 22... ...2 .2.. 22... .... 2 ..... ...2 .... ....2 ...2 ....2 .... .... 22.2. 2... . ..... ...2 .... ..... ...2 .2... .... .... .2.2. ...2 . ...2. .... 2... ...2. .2.2. .2.. 2... ....2 . ...2. o... ...2 2.... ...2 ...2. 2... .... ...2. .... . .2... ...2 .... ...2. 2... .2... .... .... ..... .... . ..... .... .... .2.2. ...2 ..... .... 2... ....2 .... . . L0 B: m...... . ... 2 ... 2 ....... 2 .2. . ....uuu”wwn n.. .2. 2 ... . ....... . .2. 2 2.“..M «.2 ..co..:o>coo ..o.:uouo.. «.2 cc. ...o...o>cou AMI2souzo. ...n v.2.-.a.. 2. 2_.N_.=< .~ .22 20 ...2 ....2 ...2. .2.. 2... ..... ...2. .... .. 2... .2... ..... ...2 .... ....2 ...2. .... .. ...2 ..... .2... .... .... ..... .2... .2.. .2 .... 2.... 2. 2. .... 2... ....2 2.... 2... 2. .... .2... ..... .... ...2 ....2 ....2 .2.. .. .... ....2 ....2 2... 2... ...2. .2... 22.. .. 2..2 ..... 2.... .... .... ..... ..... 2... .2 .... ....2 ...2. 2... .... 2.... ..... .2.. .2 2... .2.2. .2... ...2 ...2 ..... ..... .... .2 2... ..... ..... 2... ...2 2.”.. ..... 2... .. 2... ..... ...2. .2.. .2.. ....2 .. 2. ...2 .2 .... .2... ....2 .... 22.. ....2 2.... .2.. .2 .... 22.2. ..... .. . 2. . ...2. .2 .. .... 22 .2.. 2.... 2.... ...2 .... ..... ...22 ..H. .2 ...2 ...2. .2... .... .... ..... 2.... .2.. .2 ...2 .2.2. ....2 .2.. .. . ...o. ....2 .... .2 .... ..... ..... ... .... .2. .. ..... .. . 2. .2.. 2.... ...2. 2... .. . .2... 2. 2. .... 22 .2.. ...2. .2... .2.. .2.. 2.... ..... .... .2 .2.. ..... 2.... .... ...2 ....2 ..... .... .2 o... 2.... .2 .2 ...2 .. . 2.... ...22 .... .. .... ..... ..... .... 2. . .. .. ..... ...2 .. .2.. .2... .2... .o.. ...2 ...2. ...2. ...2 2. .... 22... ..... ... .... .2... 2.... .... .. 2.... .2... .2.. .2 .. ....2 ...2 .. .... ...2. 2.... .... ...2 ..... 22 2. ..h. .. 2... .2... ...2. .2.. 2... ...2. ..... 2... .. .... ....2 ...2. 2... .... 22... ...2. 2... .. 2... ....2 .2... .... .... ...22 2.... .2.. .. .... ..... .. .. .... ...2 ....2 .. .. .... .. .... ..... ...2. .... 2... .. .2 .. .. .... . .... ..... ....2 ...2 2... ..... ....2 .... . .2.2 ..... 2..2. .... .... ..... .2... .... . .... ....2 ...2. .... .... 2.... 2.... ...2 . .... 2.... .2... .2.. .... .. .2 .. .2 .2 . . ....2 ...2. .2.. .2.2. ...2. 22.. . .... .2... ....2 .... 2... 2.... 2.... ...2 2 ...2 ....2 2.... ...2 ...2 ....2 ..... .2.. 2 .... ....2 ...2. .... .2.. 22... 2.... ...2 . 2 2 .2. 2 2.. 2 ....... 2 . z 2 ..< 2 ... 2 ....... 2 . z 2 ...s.. 2.2 ...o...o>cou 0.23.. ..ogavouosm 2.2 we. ..c¢.ucu>:ou Nd 2Louzoq «...2. 2° 2.2N—222 .n a...» 21 Table 4. Analysis of air-classified bean flour by conven- tional and NIR procedures. Bean type Flour Class Conventional NIR ' F 1. z z z ‘2‘" Black Beans H20 Protein Ash H10 Protein Ash Granular '__TU.65 25.75 4.167710.10 23.29 74127 Fine I 9.44 35.27 5.20 8.74 35.91 5.30 Coarse I 9.63 23.17 3.84 8.77 23.47 4.15 Fine II 8.85 37.09 5.62 8.27 37.56 5.71 Coarse II 9.85 20.52 3.53 9.32 18.71 3.45 Fine High Protein 9.36 36.55 5.36 8.65 35.45 5.39 Pinto Beans Granular 8.35 26.80 4.28 8.79 27.62 4.73 Fine I 8.33 39.96 6.14 7.81 39.85 5.91 Coarse I 8.64 20.94 3.34 8.28 18.85 3.64 Fine II 8.35 34.10 5.12 8.27 34.75 5.51 Coarse II 8.47 18.64 2.91 8.13 15.57 3.16 Fine (1 + II) 8.40 38.28 6.04 8.11 38.21 5.83 Navy Beans Granular 8.88 23.11 3.84 8.77 25.62 4.55 Fine I 7.87 39.14 5.72 8.05 40.78 5.95 Coarse I 8.80 20.29 3.22 8.07 19.59 3.62 Fine II 7.76 39.78 5.57 7.33 41.31 6.03 Coarse II 8.21 17.16 2.76 7.66 16.02 3.25 Fine (I + II) 7.66 40.41 5.63‘ 8.04 40.24 5.98 Roasted Navy Beans Granular 7.13 27.12 3.70 7.86 28.93 4.56 Fine I 7.49 39.87 5.46 7.40 43.18 5.91 Coarse I 7.46 20.47 3.23 7.08 19.38 3.35 Fine I 7.02 40.40 5.30 7.19 42.18 5.97 Coarse II 6.92 17.33 2.83 7.06 15.91 2.99 Fine (I + II) 7.04 42.36 5.40 7.69 41.39 5.94 Tempered Roasted Navy Beans Granular 7.92 28.11 3.91 7.78 28.98 4.49 Fine I 7.25 44.95 5.34 7.54 43.89 6.00 Coarse I 7.78 21.21 3.67 7.11 20.37 3.40 Fine II 7.16 42.69 5.15 7.60 41.64 5.96 Coarse II 7.33 17.81 3.42 7.54 17.81 3.14 Fine (I + II) 7.11 47.03 5.19 7.19 43.27 6.01 22 Table 5. The best fbur reflectance pulse points selected by regression analysis for dehydrated milk, cheese and bean flour constituents. Product Pulse points Food Component Pot P1 , P2 P3 P4 Wet-mixed milk Moisture 80 265 611 “903 963 Protein 80 594 635 927 960 Fat 80 601 635 914 967 Ash 80 605 629 909 965 Lactose 80 275 914 938 965 Dry-mixed milk Moisture 80 306 628 912 968 Protein 80 274 306 612 906 Fat 80 566 609 906 961 Ash 80 245 267 926 977 Lactose 80 311 611 906 959 Cheese-powder Moisture 80 233 271 303 596 Protein 80 233 307 596 962 Fat 80 243 258 927 979 Ash 80 245 267 926 977 Air-classified bean flour Moisture 80 270 567 599 912 Protein 80 273 312 599 899 Ash 80 604 626 934 962 *Po - Number of revolutions of the filters per minute. 23 Table 6. Multilinear regression coefficients (K values) used for NIR analysis of milk powder. cheese and bean flour. Product Food Component ‘ K0 .K1 K2 K3 K4 Net-mixed milk Moisture 4.3 300.7 -ll6l.5 402.5 2430. Protein 28.3 890.6 -944.2 -159.4 -1324. Fat 15.8 -2512 1524.9 242 -3264.3 Ash 6.1 117.4 180.2 -125.1 194 Lactose 49.4 4627.9 2304.3 -1015.4 1615. Dry~mixed milk Moisture 3.9 352.5 -1072.3 -334.6 -70.1 Protein 4.7 394.7 365 266.5 -2407.8 Fat 30.9 2279.5 5410.8 4550 532.3 Ash 3.0 18 339.9 -294 .1.7 Lactose 43.4 -529.3 1130.1 -661.3 -1182.3 Cheese powder Moisture 2.1 107.9 147.2 -539.1 -28.5 Protein 32.8 354.3 2106.2 246.1 -1461.6 Fat 2.7 128.9 -1296.9 86.4 -2134.7 Ash 6.7 71.2 147.1 -40.3 92.3 Air-classified bean flour Moisture 7.9 -150.1 -148.8 -157.4 -1430.4 Protein 28.4 1506.1 -1824.1 5629.4 1693. Ash 7.1 -279.3 319.3 -56.3 2. 24 conventional and NIR methods are shown in Table 7. A wide range is necessary for accurate calibration of GOA-41 instrument. The relation— ships between moisture content obtained by the oven method and those predicted by the NIR of the spray-dried milk, dry-mixed milk. cheese powder, and air-classified bean flour are shown in Figures 1, 6. 11 and 15. The correlation coefficients (r) for the moisture content of spray- dried milk, dry-mixed milk, cheese powder. and air-classified bean flour were 0.7889. 0.8208. 0.7494 and 0.8971. respectively. Hymowitz 93 El: (1974), working with corn, soybeans and oats reported r values for moisture as 0.839, 0.944 and 0.621. respectively. The narrow range of moisture content of the products, might have contributed to the relatively low moisture correlations compared to those obtained for protein, fat and ash. The regression lines between Kjeldahl and NIR data for the protein content of the same four products are shown in Figures 2, 7. 12 and 16. The correlation coefficients were 0.9672. 0.9844, 0.9352 and 0.9571 for spray-dried milk, dry-mixed milk, cheese powder and bean flour, respectively. The fat content determined by Mojonnier and NIR methods were also highly correlated. The corresponding r values were 0.9892. 0.9954 and 0.9803 for spray-dried, dry-mixed milks. and cheese powder, respective- ly. The regression lines are shown in Figures 3. 8 and 13. The Figures 4. 9. 14 and 17 show the relationships between % ash determined by dry ashing and NIR. The correlation coefficients obtained were: 0.9437 for spray-dried milk, 0.9846 for dry—mixed milk, 0.7486 for cheese powder and 0.9571 for bean flour. The low correlation co- efficient 0.7486 for the ash in cheese powder may be due to the uneven 25 Table 7. Mean and range of the content in several constit- uents of foods analyzed by conventional methods and subsequently subjected to NIR analysis. Food Constituents Mean Range Spray dried milk Moisture 2.64 0.64 4.99 Protein 27.43 23.91 33.54 Fat 16.68 1.01 26.49 Ash 6.51 5.72 7.36 Lactose 46.55 41.01 55.19 Dry-mixed milk Moisture 2.75 2.22 3.11 Protein 28.89 24.83 33.51 Fat 13.69 1.03 26.49 Ash 6.94 5.98 7.94 Lactose . 47.76 40.04 54.55 Cheese powder Moisture 4.03 1.45 6.01 Protein 19.03 13.69 24.88 Fat 38.30 22.36 73.86 Ash 8.13 2.85 9.79 Air-classified bean flour Moisture 8.17 6.92 10.65 Protein 30.88 17.16 47.03 Ash 4.51 2.76 6.03 26 % Moisture (NIR) 7 r Y = 0.5928 + 0.7963 x 5 J- r = 0.7889 n = 41 ////// 5 F 4 ' i-° 'F . 3 n O . O ‘ . ‘b 2 ’ L . «t 0 l I L I I o T" 2 3 4 s % Moisture (oven method) Figure 1. Relationship between % moisture determined by the oven method and by the NIR method in spray dried milk. Bars show the 95% confidence limits of the linear regression. 27 3.4 Y = 0.5761 + 1.1859 X 3 2 i r = 0.8208 v , I; n = 30 , '2 V30' ‘. O 0 O 2 to 3 4.4 U! $32.8 . N 2.6 b 2.4 0 0 2.2 2.6 3.0 3.4 % Moisture (Oven Method) Figure 6. Relationship between % moisture determined by the oven method and by the NIR method in dry- mixed milk. Bars show the 95% confidence limits of the linear regression. % Moisture (NIR) Figure 11. 28 -< I - 1.5062 + 0.6253 X 0.7494 1 ll 40 0 2 4 % Moisture (oven method) Relationship between % moisture determined by the oven method and by the NIR method in cheese powder. Bars show the 95% confidence limits of the linear regression. 11 % Moisture (NIR) Figure 15. 29 -1.4783 + 1.2051 X 0.8971 so L 7.5 9.5 11.5 % Moisture (oven method) Relationship between % moisture determined by the oven method and by the NIR in air-classified bean powder. Bars show the 95% confidence limits of the linear regression. % Protein (NIR) 3O 35 Y = 1.6199 + 0.9381 X r = 0.9672 0 n = 41 30 25 N 000 Figure 2. 35 25 30 % Protein (micro-Kjeldahl) Relationship between % protein determined by micro-Kjeldahl and by the NIR methods in spray dried milk. Bars show the 95% confidence limits of the linear regression. % Protein (NIR) 35 ..< II 1 II 33 31 31 29 27 25 23 0 0 24 Figure 7. 0.9506 + 0.9773 X 0.9844 31 26 28 30 32 34 36 % Protein (micro-Kjeldahl) Relationship between % proteins determined by micro-Kjeldahl method and by the NIR method in dry-mixed milk. Bars show the 95% confidence limits of the linear regression. 32 31 ( Y = 0.3214 + 1.0093 X r = 0.9352 n = 40 (NIR) 21 - c '8 4.) o s. m a! ll _ ‘P 0 Al I l 0 11 21 31 O A Protein (micro Kjeldahl) Figure 12. Relationship between % proteins determined by micro Kjeldahl method and by the NIR method in cheese powder. Bars show the 95% confidence limits of the linear regression. 33 46 Y = 2.0132 + 0.9422 X . C r = 0.9878 . o n = 30 o 38 (NIR) % Protein. 22 14 0 16 32 48 % Protein (micro Kjeldahl) Figure 16. Relationship between % proteins determined by micro Kjeldahl and by the NIR in air-classified bean powder. Bars show the 95% confidence limits of the linear regression. % Fat (NIR) 30 20 10 34 Y = 0.9969 + 0.9336 X .1 o O r = 0.9892 n = 41 I; "1. O o '1. ‘2’.- i O. Q - l O O O O O 0 10 20 30 % Fat (Mojonnier) Figure 3. Relationship between % fat determined by the Mojonnier and by the NIR methods in spray dried milk. Bars show the 95% confidence limits of the linear regression. 30 25 20 % Fat (NIR) Figure 8. 35 = -0.0831 + 0.9915 x ' = 0.9954 ° C = 30 C C . C C C '0 O . O C C O C 5 10 15 20 25 30 % Fat (Mojonnier) Relationship between % fat determined by Mojonnier method and by the NIR method in dry-mixed milk. Bars show the 95% confidence limits of the linear regression. % Fat (NIR) 80 72 36 1.3981 + 0.9479 X 0.9803 40 Figure 13. 28 40 52 64 76 % Fat (Mojonnier) Relationship between Mojonnier method and cheese powder. Bars limits of the linear % fat determined by by the NIR method in show the 95% confidence regression. %.Ash (NIR) 1 -< II II 3 II Figure 4. 37 1.2065 + 0.8090 X 0.9437 41‘ ‘ / 6 7 % Ash (Dry Ashing) Relationship between % ash determined by dry ashing and by the NIR methods in spray dried milk. Bars show the 95% confidence limits of the linear regression. 38 7.7 Y = 4.0361 + 0.4455 X r = 0.9846 0 n = 30 ° \1 % Ash (NIR) 6.7 0 6 7 8 % Ash (Dry Ashing) Figure 9. Relationship between % ash determined by dry ashing method and by the NIR method in dry- mixed milk. Bars show the 95% confidence limits of the linear regression. 10 % Ash (NIR) m Figure 14. 39 1.9754 + 0.7808X 0.7486 , 39 4 6 3 10 % Ash (Dry ashing) Relationship between % ash obtained by dry ashing method and by the NIR in cheese powder. Bars show the 95% confidence limits of the linear regression. % Ash (NIR) Figure 17. 40 1.9754 + 0.7808 X 0.9571 -. 30 3 4 5 6 % Ash (Dry Ashing) Relationship between % ash determined by the dry ashing and by the NIR methods in air- classified bean powder. Bars show the 95% confidence limits of the linear regression. 41 distribution of values, the vast majority of which were clustered in the 8% to 10% range. The relationships between % lactose obtained by difference and those estimated by NIR instrument are shown in Figures 5 and 10. The r values were 0.9528 and 0.9959 for spray-dried and dry-mixed milk, respectively. The unknown samples of spray-dried milk, dry-mixed milk and bean flour were prepared in the same way as the calibrating samples and were sub- jected to NIR analysis using the pulse points and K values shown in Tables 5 and 6. The samples were later analyzed by the conventional methods. Table 25 shows the analysis by conventional and NIR methods. The correlation coefficients are shown in Table 26. the moisture had lower r value in comparison to protein, fat. ash and lactose. The r values were 0.8384. 0.6017, and 0.7467 for spray-dried milk, dry- mixed milk and bean flour, respectively. The r values of the other food constituents were higher and ranged from 0.8625 to 0.9908. The analysis of variance for the calibrating samples are shown in Tables 8 to 24. The F-value was calculated for each food component (moisture, protein, fat, ash and lactose). It was found that all F- values were smaller than F-tables at 10% probability level. It shows that there was no significant statistical difference at 10% probability level between the data obtained by conventional methods and those pre- dicted by NIR, and also demonstrates the capability of NIR technology in predicting moisture. protein. fat, ash and lactose contents of the dairy products and bean flour. The analysis of variance of the unknown dairy products and bean flour samples are shown in Tables 27 to 39. The F-value of all food constituents except moisture of spray-dried milk provide another 56 52 % Lactose (MIR) .5 G) 44 40 Figure 5. 42 0.3333 + 0.9979 X t . 0.9528 ///// 41 44 48 52 56 % Lactose (By difference) Relationship between % lactose calculated by difference and predicted by NIR method in spray dried milk. Bars show the 95% confidence of the linear regression. % Lactose (NIR) 55 50 45 40 43 O Y = 0.1404 + 0.9984 X 0 r = 0.9959 0 n = 31 Figure 10. 55 45 50 % Lactose (By difference) Relationsihp between % lactose calculated by difference and predicted by the NIR in dry- mixed milk. Bars show the 95% confidence limits of the linear regression. 44 ao.v - mm.~n - m~.o - ~c.~n - o a... - o_.o~ so.» ao.v - ~n.m~ o_.o m mo.n - aa.—~ oc.o .e.n - ao.- oo.m e na.e - ~c.~n Na.» om.e - ~m._n cm.m n a~.m - op.~n .o.o o~.o - om.~c an.» N _~.c - oo.v~ ~—.a _~.¢ - m~.- ~_.m _ Lao—e coma ua.~.nma_u-gv< - av.“ ~a... «n.on me.~ . c~.~ co.o mm.on o~.n o Nonae mv.~ o~.o— mo.on ~_._ oa.oc no.5 ac..— oo.on -.n m o~._m mm.“ mm.“ -.—n co.~ no.cm cm.~ mm.“ nn._n sh.n e as ~m mm.“ so.m -.~n eo.~ um.—m om.~ vn.m an._n _~.v n ~o.nm no.“ cm.n o~.nn .o.n c..~m -.~ nu.n .o.~n oc.o N ce.nm no.“ oc.~ an.nn c_.n co.~m m~.~ .m.~ ~—.~m mc.c _ x—.E sax—E-»so pn.~q mn.o - m_.om n~.~ oo.~v m~.o - om.- oo.e m amnov ~m.o - me.- no.~ ma.~o an.u - ~o.- mo.e m on me oo.m sv.o_ om.- oo.~ ov.¢v nm.m o~.m— m—.a~ om.e e o_.~m ~m.~ oo.o— ~m.nn oo.~ oo.nm mo.“ o~.o_ o—.nn om.¢ m .m.om Na.“ on... .o.nn na.n ov.cm No.“ nc.~_ .m.~m wo.e m om.om mm.“ a..— mo.nn mn.n mn.~m no.“ —_.— mc.¢n mo.o _ 1| Aoucososw—v r: omcuuaa a gm< a u.u.u :.ouoga a ON: a amouuad a gm< a «an a :.uuotn,u o~z a a..: uo—Lu a-Lam gossaz ¢~z ..co.ucu>:ou u—anm .nuoguos ¢_z can .oco.ucu>¢ou an no—aeuu exocxca Lao—u econ uu—u.mua_u-c_a cc. x—vn cox.a-msu .x_.e no.gu sauna uo m_u»—~c< .mN o_nap 45 Table 26. Number of samples, correlation coefficient (r), slopes and Y-intercepts of the linear regression of unknown samples analyzed by conventional and NIR methods. Food constituents n r slope y-Intercept Spray dried milk Moisture 6 0.8384 3.1079 -8.8589 Protein 6 0.9282 0.9023 3.3147 Fat 4 0.9908 0.8996 0.2969 Ash 6 0.9886 0.5767 2.8222 Lactose (by difference) 6 0.9882 0.8269 12.6024 Dry-mixed milk Moisture 6 0.6017 1.2248 -2.4784 Protein 6 0.9945 1.9893 -30.4649 Fat 6 0.9514 1.0393 -0.0521 Ash 6 0.8625 0.2538 5.6759 Lactose (by difference) 5 0.9628 0.8975 6.2889 Air-classified bean flour Moisture 5 0.7467 0.7071 2.5689 Protein 6 0.9394 0.7789 5.4563 Ash 6 0.9699 0.7346 1.3100 46 Table 8. Analysis of variance of moisture content of spray dried milk analyzed by conventional and NIR methods. Source of Degree Sum of Mean F- F-tabulated variance of squares square statistic, a=0.10 freedom Treatment 1 0.1018 0.1018 0.0754 2.75 Experimental 80 0.0169 1.3506 error Total 81 0.0179 1.4524 Table 9. Analysis of variance of protein content of spray dried milk. Source of Degree Sum of Mean F- F-tabulated variance of squares square statistic a=0.10= freedom Treatment 1 0.1598 0.1598 0.0368 2.75 Experimental 80 0.0543 4.3415 error Total 81 0.0556 4.5013 Table‘HL Analysis of variance of fat content of spray dried milk. Source of Degree Sum of Mean F- F-tabulated variance or squares square statistic ¢1= 0.10 freedom Treatment 1 0.3367 0.3367 0.0080 2.75 Experimental 80 0.5249 41.9921 error Total 81 0.5226 42.3288 47 Table 11. Analysis of variance of ash content in spray dried milk. Source of Degree Sum of Mean F- F-tabulaum variance of squares square statistic. o=0.10 freedom Treatment 1 0.1980 0.1980 0.8536 2.75 Experimental 80 0.0029 0.2320 error Total 81 0.0053 0.4300 Table 12. Analysis of variance of lactose content of spray dried milk. Source of Degree Sum of Mean . F- F-tabulated variance of squares square statistic a=0.10 freedom Treatment 1 1.1139 1.1139 0.0879 2.75 Experimental 78 0.1624 12.6674 error Total 79 0.1745 13.7813 Table 13. Analysis of variance of moisture content of dry-mixed milk analyzed by conventional and NIR methods. Source of Degree Sum of Mean F. F-tabulated variance of squares square statistic «=0.10 freedom Treatment 1 0.0025 0.0025 0.0382 2.79 Experimental 60 0.0011 0.0675 error Total 61 0.0012 0.0700 _— 48 Table 14. Analysis of variance of protein content of dry-mixed milk analyzed by conventional and NIR methods. Source of Degree Sum of Mean F- F-tabulated variation of squares square statistic 4‘0-10 freedom Treatment 1 ‘ 1.4130 1.4130 0.1730 g 2.79 Experimental 60 0.1361 8.1635 error ' Total 61 0.1569 9.5765 Table 15. Analysis of variance of fat content of dry-mixed milk analyzed by conventional and NIR methods. Source of Degree Sum of Mean F— F-tabulated variation of squares square statistic “=0-10 freedom . Treatment 1 1.4342 1.4342 0.0242 2.79 Experimental 60 0.9872 59.2342 error Total 61 0.9946 60.6684 Table 16. Analysis of variance of ash content of dry-mixed milk analyzed by conventional and NIR methods. Source of Degree Sum of Mean F- F-tabulated variation of squares square statistic a=o,in freedom Treatment 1 0.5748 0.5748 2.1180 2.79 Experimental 60 0.0045 0.2714 error Total 61 0.0139 0.8462 49 Table 17. Analysis of variance of lactose content of dry-mixed milk analyzed by conventional and NIR methods. Source of Degree Sum of Mean F- F-tabulated variance of squares square statistic 030-10 freedom Treatment 1 0.7436 0.7436 0.0441 2.79 Experimental 60 0.2807 16.8429 error Total 61 0.2883 17.5865 Table 18. Analysis of variance of moisture content of cheese powder analyzed by conventional and NIR methods. Source of Degree Sum of Mean F- F-tabgla6ed variance of squares square statistic ‘ 3’ ~ freedom Treatment 1 0.1110 0.1110 0.2005 2.75 Experimental 78 0.0071 0.5536 error Total 79 0.0084 0.6646 Table 19. Analysis of variance of protein content of cheese powder analyzed by conventional and NIR methods. Source of Degree Sum of Mean F- T-tabulated variance of squares square statistic a=0.10 freedom Treatment 1 4.8609 4.8609 0.5683 2.75 Experimental 78 0.1097 8.5529 error Total 79 0.1698 13.4138 50 Table 20. Analysis of variance of fat content of cheese powder analyzed by conventional and NIR methods. Source of Degree Sum of Mean F- F-tabulated variance of squares square statistic a=O.lO freedom Treatment 1 7.9632 7.9632 0.0800 2.75 Experimental 78 1.2751 99.4540 error Total 79 1.3597 107.4172 Table 21. Analysis of variance of ash content of cheese powder analyzed by contentional and NIR methods. Source of Degree Sum of Mean‘ F- F-tabulated variance of squares square statistic. a=0.10 freedom Treatment 1 0.2063 0.2063 0.1170 2.75 Experimental 74 0.0238 1.7629 error Total 75 0.0263 1.9692 Table 22. Analysis of variance of moisture content of air-classified bean flour analyzed by conven- tional and NIR methods. Source of Degree Sum of Mean F- fF-tabulated variance of squares square statistic a=0.10 freedom Treatment 1 0.4018 0.4018 0.5639 2.84 'Experimental 58 0.0123 0.7125 error Total 59 0.0189 1.1143 51 Table 23. Analysis of variance of protein content of air- classified bean flour analyzed by conventional and NIR methods. Source of Degree Sum of Mean F- F-tabulated variance of squares square statistic a=0.10 freedom Treatment 1 0.7216 0.7216 0.0072 2.84 Experimental 58 1.7262 100.1174 error Total 59 1.7091 100.8390 Table 24. Analysis of variance of ash content of air- classified bean flour analyzed by contentional and NIR methods. Source of Degree Sum of Mean F- T.tabulated 'variance of squares square statistic a=0-10 freedom Treatment 1 1.4322 1.4322 1.1724 2.84 Experimental 58 0.0211 1.2215 error Total 59 0.0451 2.6537 52 Table 27. Analysis of variance of moisture content of spray-dried milk unknown samples analyzed by contentional and NIR methods. Source of Degree Sum of Mean F- F-tabulated variance of squares square statistic 0=0.05 freedom ' . Treatment 1 3.3920 3.3920 7.9045 4.96* Experimental 10 0.0429 0.4291 error - Total 11 0.3474 3.8211 *Significant at 51 probability level. Table 28. Analysis of variance of protein content of spray dried milk unknown samples analyzed by conven- tional and NIRS methods. Source of Degree Sum of Mean F- F-tabulated variance of squares square statistic 0=0.10 freedom Treatment 1 0.2945 0.2945 0.0325 3.29 Experimental 10 0.9044 9.0437 error Total 11 0.8489 9.3382 Table 29. Analysis of variance of fat content of spray dried milk unknown samples analyzed by conven- tional and NIR methods. Source of Degree Sum of Mean F- F-tabulated variance of squares square statistic a=0.10 freedom Treatment 1 3.8364 3.8364 0.0553 3.29 Experimental 10 11.5566 69.3398. error Total 11 10.4537 73.1762 53 Table 30. Analysis of variance of ash content of spray dried milk unknown samples analyzed by conven- tional and NIR methods. Source of Degree Sum of Mean F- F-tabulated variance of squares square statistic a=0.10 freedom Treatment 1 0.1140 0.1140 0.2671 3.29 Experimental 10 0.0427 0.4269 . f error : Total 11 0.0492 0.5409 . Table 31. Analysis of variance of lactose content of spray dried milk unknown samples calculated by dif- ference and predicted by NIR method. Source of Degree Sum of Mean F- F-tabulated variance of squares square statistic a=0.10 freedom Treatment 1 53.1723 53.1723 1.9472 1 3.29 Experimental 10 2.7304 27.3043 error Total 11 7.3161 80.4766 Table 32. Analysis of variance of moisture content of dry- mixed milk unknown samples analyzed by conven- tional and NIR methods. Source of Degree Sum of Mean F- .F-tabulated variance of squares square statistic- a=0.10 freedom Treatment 1 . 2.7840 2.7840 1.4676 3.29 Experimental 10 0.1897 1.8969 error Total 11 0.4256 4.6809 54 Table 33. Analysis of variance of protein content of dry- mixed milk unknown samples analyzed by conven- tional and NIR methods. Source of Degree Sum of Mean F- F-tabulated variance of squares square statistic a=0.10 freedom Treatment 1 1.0384 1.0384 1.1425 3.29 Experimental 10 0.0909 0.9088 error Total 11 0.1770 1;9472 Table 34. Analysis of variance of fat content of dry-mixed milk unknown samples analyzed by conventional and NIR methods. Source of Degree Sum of Mean F- F-tabulated variance of squares square statistic a=0.10 freedom Treatment 1 0.1365 0.1365 0.0106 3.29 Experimental 10 1.2870 12.8699 error Total 11 1.1824 13.0064 Table 35. Analysis of variance of ash content of dry- mixed milk unknown samples analyzed by conventional and NIR methods. Source of Degree Sum of Mean F- F-tabulated variance of squares square statistic a=0-10 freedom Treatment 1 0.0507 0.0507 1.1192 3.29 Experimental 10 0.0045 0.0452 error Total 11 0.0087 0.0959 55 Table 36. Analysis of variance of lactose content of dry- mixed milk unknown samples calculated by difference and predicted by NIR method. Source of Degree Sum of Mean F- F-tabulatEE variance of squares square statistic o=0.10 freedom Treatment '1 2.7667 2.7667 1.0988 3.46 Experimental 8 0.3147 2.5178 error Total 9 0.5872 5.2845 Table 37. Analysis of variance of moisture content of air- classified bean flour unknown samples analyzed by conventional and NIR methods. Source of Degree Sum of Mean F- F-tabulated variance of squares square statistic 0‘0-10 freedom Treatment 1 0.1440 0.1440 0.8326 3.46 Experimental 8 0.0216 0.1729 error Total 9 0.0352 0.3169 Table 38. Analysis of variance of protein content of air- classified bean flour unknown samples analyzed by conventional and NIR methods. Source of Degree Sum of Mean F- F-tabulated variation of squares square statistic a=0.10 ' freedom Treatment 1 5.6994 5.6994 0.1483 3.29 Experimental 10 3.8421 38.4210 error Total 11 4.0110 44.1204 56 Table 39.’ Analysis of variance of ash content of air- classified bean flour unknown samples analyzed by conventional and NIR methods. ‘4 Source of Degree Sum of Mean F- F-tabulated variance of squares square statistic. ‘p=0.10 freedom Treatment 1 0.0374 0.0374 0.0525 3.29 Experimental 10 0.0712 0.7120 error Total 11 0.0681 0.7494 57 evidence that both NIR and conventional procedures are identical and there is no significant statistical difference at the 10% probability level. The F-value for moisture in the spray-dried milk (Table 27) was significant at the 5% probability level, and might be due to the changes in moisture content from the time of calibration of GOA-41. To test the capability of GOA-41 prediction, some commercial bean flours were tested for their moisture, protein and ash contents. The samples were subjected to NIR analysis using the corresponding pulse points and K values (Tables 5 and 6), and later subjected to conventional analysis. The discrepancies were small. Table 40 shows the analysis of commercial bean flour by con- ventional and NIR methods. The commercial bean flour samples were ground for 2 minutes using the Dickey-john grain grinder to see any differences due to grinding. The samples were subjected to NIR analysis and the data are shown in Table 40. The extra 2 minutes of grinding did not improve the predictability. Eleven commercially dried milk powder samples were obtained from the Michigan State Department of Agriculture in E. Lansing. Their moisture. protein. fat. ash and lactose content was estimated by NIR using the pulse points and K values of the spray-dried milk samples, and were later analyzed by the conventional methods. It was found that the discrepancies between the NIR and conventional values were very large. The pulse points and K values of the dry-mixed milk (Tables 5 and 6) were used, but the differences between the estimated and analytical- values were large. This may be due to the different physical character- istics of the commodities resulting in different reflectance spectra. 58 me.ot cp.m mn.~ mm.¢1 mm.~ mm.~ :m<& m_.o+ No.m_ mp.mp m¢.O1 oc.op m_.m_ c_mpomex m¢.F+ em.o mm.w op._+ PP.N -.m o :x HH :oPgooLm camp x>mz Rm.o- No.o mm.m em.o1 mm.m mm.m :m<& om.o+ _m.oe FF.F¢ Pm.o+ om.oe PP._¢ :wmuomax Pm.o1 mo.w -.~ mm.o- mo.m um.“ o :& HH + H cowuumgm coma x>wz op.o- mm.~ ¢~.~ NN.o- mm.m ¢N.N cm>mz mocmgmwwro mHz pocowucm>coo mucmgmewro «Hz Pmcovucm>cou pcmauwumcou coo; acvucwgm gmum< mcrccrgm mgommm cowuuotm gzopm :mmm .mmuzcve N Low mcvccvem gmuwo can mgoemn muogpms mfiz can Pocovucm>cou xa czar» no?$?mmm_u-gwm camp >>mc Powugmesoo we mwmzpcc< .08 mpnmh 59 Milk, in industry is never dried with its original moisture content. but evaporated to concentrations of 45—52% total solids prior to spray drying, and hot-air temperature up to 750°F (400°C) may be used for drying with secondary cool air introduced lower in the drying chamber. The temperature of 480“F (250°C) appears to be the maximum without the cool air system. Skim milk powder is dusty, and to overcome this problem. the powder is agglomerated thereby improving its wetability and dispersibility characteristics. When the eleven commercial milk powders were used for calibration of the GQA-4l. pulse points and K— values were obtained which were different from those of the spray- dried and dry-mixed samples prepared in the laboratory. This indicates that the spectral characteristics of the commercial and laboratory samples were different and explains the failure of predicting the composition of the former using the calibration data of the latter. However, when the calibration data of the eleven commercial samples (Tables 42 and 43) were used to predict the composition of two unknown commercial samples (Table 41), the agreement between analytical and pre— dicted values for protein, fat, ash and lactose were very good for the 13 commercial samples of Table 41. The correlation coefficients between the analytical and NIR data were 0.8507, 0.9878. 0.992. 0.9949 and 0.9888 for moisture, protein. fat. ash and lactose contents, respectively (Table 44). The analysis of variance for each food constituent of the commer- cially dried milk are shown in Tables 45 to 49. F—values for moisture, protein, fat, ash and lactose were not statistically significant at 10% probability level. 60 .mmpnsam czocxcze x__5 22u «N.mm mm.N «N.o m_.mm oo.N Nm.mm oN.N e_.p m¢.mm o_.e uneco: mupmnosm «m— mo.mm Nw.N No.p NN.pm om.N cN.mm .m.N m~.p om.pm cu.m cowuocgau aa< 4N. om.mm mo.c pm.mN co.mN pm.m mN.wm pm.m mm.cN NN.mN po.m = e _P pm.xm om.m mm.mN mm.mN em.m eo.mm «0.9 pm.oN <_.mN No.m e = o_ mm.Nm po.o pm.mN ma.eN Nm.e co.mm P—.w NN.oN m¢.mN mo.m e e a Nm.mm mm.m —P.NN mo.eN um.m Nm.Nm No.9 mo.oN mm.mN mN.m = = m mN.Nm mm.m om.oN .m.mN om.m oo.Nm Nm.m mN.oN mm.mN Nm.m = = N Nn.mm Nm.m we.NN we.mN mm.n mo.mm mm.m mN.oN m~.mN me.m muuauogm m_aucgw> o oN.om oo.N Nm.o m—.—m N¢.N Nm.cm mm.N e_.p mN.—m oo.N cowuacgou m ma..m em.~ mm.” NN.mm mo.m m~.mm mm.~ p_._ Nm.mm ow.m aa< c mm.om mo.~ mm.o mm._m .o.¢ em.mm _o.N eo._ Nm._m mm.m m.gmmogx m oN.mm om.N No.o we.Nm mm.m oN.vm No.N No.o mm.Nm NN.¢ muwcmpnap =ou .mcozuoe a”: van _a:o.»:o>=ou xn mo_asmm x—ws um_gu >_~o_ugmseou to m_mapu:< ._e «pack 61 Table 42. The best four reflectance pulse points selected by regression analysis for commercially dried milk. Food Constituent P0 P1 P2 P3 . P4. 1 Moisture 80 288 611 903 962 1 Protein 80 264 286 601 961 3 Fat 80 614 915 954 978 1 Ash 80 248 309 620 973 % Lactose 80 597 624 906 964 62 Table 43. Multilinear regression coefficients (K values) used for NIR analysis of commercially dried milk. Food Constituent K0 K1 K2 K3 K4 3 Moisture 4.7 673.0 1762.7 429.2 -171.5 1 Protein 26.5 1311.3 2113.4 -2l68.l -245.8 E 1 Fat 0.70 -343.4 266.4 365.9 -536.1 i 3 Ash. - 7.3 13.4 -163.2 -463.2 105.5 x Lactose 75.6 -2648.7 -2124.8 -741.8 1224.2 63 xmpco.P + Nmum.o-n> Nmu¢.o- mFFo.P wmmm.o F. mmouqu xom8o.P + mme.O1n> mmmN.o- cm8c._ m8mm.o FF cm< xmmoc.P + mnap.o-n> onP.o- wwoo.F Nmmm.o FF we; xm8No.P + mmom.o-n> mmom.o- m8No.p m~mm.o PP cwmpoga xFNoo.p + mpoo.o n> mpoo.o FNoo._ Nomm.o PP weaumwoz copuoaou uqmugmugr-z maopm L c pcmaupumcou noon .xpws caret appavogmssou yo mmzpo> mHz use —o=ovpcm>:ou one :mmzumn cowmmmgmwt goocvp any 4o cowumacm mgu .ucn muamugmucvua .maopm .ALV mucwvuvmymou coruopmggou .Acv mmpanm mo Lassaz .88 mpnm» 64 Table 45. Analysis of variance of moisture content of commer- cially dried milk analyzed by conventional and NIR methods. Source of Degree of Sum of Mean of F- F-Tabulated variance freedom squares squares statistic a=0.10 Treatment 1 0.2002 0.2002 0.4021 2.97 Experimental error 20 9.9563 0.4978 Total 21 10.1565 0.6980 65 Table 46. Analysis of variance of protein content of commercially dried milk analyzed by conventional and NIR methods. Source of Degree of Sum of Mean of F- F=tabulated variance freedom squares squares statistic o=0.10 Treatment 1 0.3876 0.3876 0.0272 2.97 Experimental error 20 284.8026 14.2401 Total 21 285.1903 14.6277 66 Table 47. Analysis of variance of fat content of commercially dried milk analyzed by conventional and NIR methods. Source of Degree of Sum of Mean of F- F-tabulated variance freedom squares squares statistic o=0.10 Treatment 1 0.0236 0.0236 0.0001 2.97 Experimental error 20 3582.227 179.1113 Total 21 3582.2506 179.1349 67 Table 48. Analysis of variance of ash content of commercially dried milk analyzed by conventional and NIR methods. Source of Degree of Sum of Mean of F- F-tabulated variance freedom squares squares statistic o=0.10 Treatment 1 0.0001 0.0001 0.00012 2.97 ExpErimental error 20 16.4244 0.8212 Total 21 16.4245 68 Table 49. Analysis of variance of lactose content determined by difference and NIR method. Source of Degree of Sum of Mean of F- F-tabulated variance freedom squares squares statistic o=0.10 Treatment 1 1.0341 1.0341 0.0131 2.97 Experimental error 20 1582.8499 79.1424 Total 21 1583.8790 80.1765 SUMMARY Forty one spray-dried milk samples prepared in the laboratory were analyzed for moisture. protein, fat. ash and lactose contents by conven- tional methods and by NIR. The Neotec GOA-41 instrument was used for the NIR method. The optimum pulse points and the regression coefficients (K-values) of each constituent were obtained and used to calibrate the instrument. The correlation coefficients (r values) between analytical and NIR values of the samples were 0.7889, 0.9672, 0.9892, 0.9437 and 0.9528 for moisture. protein, fat, ash and lactose contents, respectively. Some unknown samples prepared in a similar way were subjected to analysis by the two procedures. The r values for moisture, protein, fat. ash and lactose determined by conventional methods and estimated by NIR were 0.8384. 0.9282, 0.9908, 0.9886 and 0.9882, respectively. Thirty one milk powders prepared by mixing whole milk powder and skim milk powder were analyzed by both procedures. With pulse points and K values characteristic for these powders the following r values were obtained: 0.8208. 0.9044, 0.9954, 0.9846 and 0.9959 for moisture. protein. fat. ash and lactose, respectively. Some unknown samples pre- pared in the laboratory by dry-mixing were analyzed by both techniques and the r values were 0.6017. 0.9945. 0.9514, 0.8625 and 0.9628 for moisture. protein, fat. ash and lactose, respectively. The pulse points and K values of the dry-mixed powders were different from those of the 69 70 spray-dried samples, Forty cheese powder samples obtained from the industry were analyzed by both procedures. The r values were 0.7494, 0.9352. 0.9803 and 0.7486 for moisture. protein. fat and ash, respectively. Thirty samples of air-classified bean flour were analyzed by both methods. The correlation coefficient values were 0.8971. 0.9878 and 0.9571 for moisture. protein and ash content. respectively. The r values for six unknown bean flours were 0.7467. 0.9394, and 0.9699 for moisture. protein and ash. respectively. The predictability of the NIR method for three commercial bean flours was very good. Eleven commercially dried milk samples were analyzed by both tech- niques. The reflectance characteristics were different from those ob- tained from the milk powders prepared in the laboratory. apparently due to different drying treatments. However. when the instrument was calibra- ted with eleven commercial milk powders. the r values were 0.8507, 0.9878, 0.9878, 0.992 and 0.9949 for moisture. protein. fat, ash and lactose, respectively. Two additional commercial dry milk samples were analyzed by NIR and conventional procedures and showed satisfactory agreement between analytical and predicted values. Analysis of variance was performed for all food constituents in the dairy and bean samples. The calculated F-values were smaller than F-table in all products. (except for moisture in unknown spray-dried milk). indi- cating no significant difference between the conventional and NIR methods. This work indicates that the NIR technology holds considerable prom- ise for the analysis of dairy and bean powders by a quick. non-destructive, and non-polluting method. REFERENCES Association of Official Analytical Chemists, 1975. Methods of Analysis. AOAC, Washington, DC. Birth, G.S. 1979. Radiometric measurement of food quality - a review. J. Food Sci. 44, 949. Cochran, N.G. and Cox, G.M. 1957. Experimental Designs. 2nd Ed., John Wiley & Sons, Inc., New York, London, Sydney. Fernandez, J.A.Q. 1981. The effect of accumulation and remobilization of C-assimilate and nitrogen on abscission, seed development, and yield of common bean (phaseolus vulgaris L.) with different archi- tectural forms. Ph.D. Dissertation, Department of Crop and Soil Sciences, Michigan State University. Giangiacomo, R., J.B. Magee, G.S. Birth, and G.G. Dully. 1981. Predict- ing concentrations of individual sugars in dry mixtures by near in- frared reflectance spectroscopy. J. Food Sci. 4g, 531. Hooton, D.E. 1978. The versatility of near infrared reflectance devices. Cereal Foods World 24, 176. Hymowitz, T., J.W. Dudley, F.I. Collins and C.M. Brown. 1974. Estima- tion of protein and oil concentrations in corn, soybean, and oat seed by NIR. Crop. Sci. 44, 713. Miller, 8.5., Y. Pomeranz, W.0. Thompson, T.W. Nolan, J.W. Hughes, 6. Davis, N.G. Jackson and D.W. Fulk. 1978. Interlaboratory and intra— laboratory reproducibility of protein determinatoin in HRW wheat by Kjeldahl and near infrared procedures. Cereal Foods World 24, 198 . 71 72 Mojonnier Bros. Co. 1925. Mojonnier Milk Tester Instruction Manual. Chicago, IL. Norris, K.H. and Hart, J.R. 1965. Direct spectrophotometric determin- ation of moisture content of grains and seeds. lg; Principles and Methods of Measuring Moisture in Liquids and Solids. V01. 4, p. 19. Reinhold Publishing Corp., New York. Norris, K.H. 1978. Near infrared reflectance spectroscopy - The Present and future. 453 Cereals '78: Better Nutrition for the World's Millions. Sixth International Cereal and Bread Congress, published by AACC, May 1978. Pomeranz, Y. and R.B. Moore. 1975. Reliability of several methods for protein determination in wheat. Baker's Digest 42, 44. Rosenthal, R.D. 1978. An introduction to near infrared quantitative analysis. Presented at the 1977 annual meeting of the American Association of Cereal Chemists. Rubenthaler, G.L. and B.L. Bruinsma. 1978. Lysine estimation in cereals by near infrared reflectance. Crop Sci. 12, 1039. Shenk, 0.5., I. Landa, M.R. Hoover and M.0. Westerhaus. 1981. Descrip- tion and evaluation of near infrared reflectance spectra-computer for forage and grain analysis. Crop Sci. 21, 355. Snedecor, G.W. and N.G. Cochran. 1955. Experimental Methods Applied to Experiments in Agriculture and Biology. 5th ed., The Iowa State College Press, Ames, Iowa. Stermer, R.A., Y. Pomeranz and R.J. McGinty. 1977. Infrared reflectance spectroscopy for estimation of moisture of whole grain. Cereal Chem. 54, 345-351. 73 Watson, C.A., D. Carville, E. Dikeman, G. Diagger and G.D. Booth. 1976. Evaulation of two infrared instruments for determining protein con- tent of hard red winter wheat. Cereal Chem. §§, 214. Watson, C.A., W.C. Shuey, 0.0. Banasik, and J.W. Dick. 1977. Effect of wheat class on near infrared reflectance. Cereal Chem. §4, 1264-1269. Williams, P.C. 1975. Application of near infrared reflectance spectro- scopy to analysis of cereal grains and oilseeds. Cereal Chem. §2, 561. Williams, P.C., S.G. Stevenson, P.M. Starkey, and G.C. Hawtin. 1978. The application of near infrared reflectance spectroscopy to protein- testing in pulse breeding programs. J. Sci. Agric. 22, 285. Williams, P. and B.H. Thompson. 1978. Influence of whole meal granular- ity on analysis of HRS wheat for protein and moisture by near in- frared reflectance spectroscopy. Cereal Chem. §§, 1014. Williams, P.C. 1979. Screening wheat for protein and hardness by near infrared reflectance spectroscopy. Cereal Chem. §§, 169. Williams, P. and P.M. Starkey. 1980. Influence of feed ingredients upon the prediction of protein in animal feed-mixes by near infrared spectroscopy. 0. Sci. Food Agric. 21, illlllll 3056 199 93 v” U"0 "2 '1 N“ Aimed mli HI