Nghiên cứu sữa part 5

be degraded or inactivated by chemical, physical, or biological means Aflatoxin M1 was decreased by 45% when 0.4% potassium bisulfite was used at 25°C for h.327 The bisulfites may cause the oxidation to a bisulfite free radical that reacts with the dihydrofuran double bond of aflatoxin to give sulfonic acid products Combinations of hydrogen peroxide, riboflavin, heat, and lactoperoxidase were used to see if aflatoxin M1 could be inactivated in milk.296 The best procedure resulted in 98% inactivation of aflatoxin M1 after use of 1% H2O2 plus 0.5 mA/ riboflavin followed by heating at 63°C for 20 When milk was treated with 0.1% H2O2 plus U of lactoperoxidase and held at 4°C for 72 h, 85% of aflatoxin M1 was inactivated These authors postulated that either singlet oxygen or hypochlorous acid were involved in the destruction of the aflatoxin Some physical methods have been experimented with to determine if they are viable options for detoxifying milk Bentonite was added to milk in 0.1 to 0.4 g/20 ml for h at 25°C It absorbed 65 to 79% of aflatoxin M1;328 however, the removal of bentonite from milk could cause some problems Yousef and Marth329'330 reported that 0.5 ppb of aflatoxin M1 could be degraded by 100% in milk after a 60-min exposure to UV at a wavelength of 365 nm at room temperature The temperature increased by 15°C during the 60-min treatment When 1% hydrogen peroxide was added to the milk and it was irradiated for 10 min, total destruction of the 0.5 ppb aflatoxin M1 was noted.330 Degradation of aflatoxin M1 by UV energy followed first-order reaction kinetics and was not affected by enzymes present in the milk In addition to physical and chemical methods, mycotoxins can be degraded by other microorganisms Flavobacteriwn aurantiacum in a concentration of X 1010 cells/ml completely degraded 9.9 fjug of aflatoxin M1AnI during h at 3O0C.328 The mechanism by which this bacterium degrades aflatoxin is not known Some microorganisms, such as L lactis subsp lactis, can convert aflatoxin B into aflatoxicol and other metabolites that are either nontoxic or less toxic than B 326 - 328 Degradation in other foods by other microorganisms is reported by Doyle et al.328 Feed can also be detoxified before it is fed to dairy cows General reviews on methods to detoxify feeds have been published.328-331 One example was reported by Price et al.332 who ammoniated cottonseed meal to reduce the amount of aflatoxin B fed to cows When ammoniated feed was consumed, aflatoxin M1 was below the limits of detectability; however, when untreated feed was consumed, the level of aflatoxin M1 increased to about /xg/L in days When the contaminated feed was removed from the diet and treated feed consumed, the level of aflatoxin M1 became nondetectable again The Food and Drug Administration authorizes ammoniation of feeds in Arizona, California, Georgia, and North Carolina, but it has not declared this treatment as being safe for all states to use.331 If measures to prevent the growth of mold and aflatoxin formation in feed commodities fail, then detoxification with ammonia can reduce aflatoxin by 97 to 98% This ammoniation detoxification process is already used in different countries Mold growth and subsequent mycotoxin production can be prevented by use of antifungal agents, such as sorbates, propionates, and benzoates Ray and Bullerman333 have reviewed the agents that prevent both mold growth and mycotoxin production 5.7.4 Regulation of Mycotoxins in Foods The presence of mycotoxins, especially aflatoxins, in foods and feeds can cause potential harm to humans and animals; therefore, many countries have developed regulations to control the amount of mycotoxins that can be in foods, or feeds Under the United States Federal Food, Drug, and Cosmetic Act, aflatoxins are considered poisonous or deleterious substances.334'335 This falls under Section 402(a)(l) of the act The Food and Drug Administration (FDA) established a guideline in 1965 that included an action level of 30 ppb aflatoxin in foods and feeds.334'335 This action level was lowered to 20 ppb by 1969 In 1977 and 1978, aflatoxin M1 was detected in market milk in the southeastern United States and in Arizona; hence, an action level of 0.5 ppb aflatoxin M1 was then set for fluid milk.335 Over 50 countries now have legislation for the presence of aflatoxins in foods and feeds.290 Tolerances range from to 20 ppb depending on the country and may be for either aflatoxin B1 or the total amount of aflatoxins B1, B , G1, and Gl2 Several countries also have set tolerances for aflatoxin M1 in milk and dairy products ranging from to 0.5 ppb.290 Van Egmond336 summarized data from 66 countries on the planned, proposed or existing legislation for aflatoxins B1, B , G1, G2, and M1 in foods, feeds, and milk and dairy products Other mycotoxins, namely chetomin, deoxynivalenol, ochratoxin A, phomopsin, T-2 toxin, stachyobotriotoxin, and zearalenone are regulated in some countries.290-336 The acceptable tolerance levels depend on the country and the food or feed Several surveys have been done to determine whether toxigenic molds or mycotoxins are present in milk and dairy products The results of some of these surveys will be summarized Bullerman337 examined both domestic and imported cheese for mycotoxin-producing molds Penicillium species were isolated from 86.4 and 79.8% of the domestic and imported cheeses, respectively Aspergillus species were isolated only from 2.3 and 5.4% of the domestic and imported cheeses, respectively CIadosporium, Fusarium, and other genera made up the rest of the molds isolated from these cheeses Toxigenic species—P cyclopium, P viridicatwn, A flavus, and A ochraceus—were found in only 4.4% of domestic and 4% of imported cheese When 118 imported cheeses from 13 countries were analyzed, had aflatoxin M1 in levels of 0.1 to ppb.338 Kivanc339 found that 65% of molds isolated from Van hereby and pickled white Turkish cheeses were Penicillium species, and fewer than 4% were Aspergillus species The rest of the molds were species of Mucor, Geotrichum, Candidum, and Trichoderma No aflatoxin was detected in any of the cheeses Blanco et al.340 analyzed commercial UHT-treated milk over year in Spain and found that 30% of the samples contained 0.02 to 0.1 ppb aflatoxin M1 Most contaminated samples were detected in summer and autumn Wood341 examined 182 samples of milk and dairy products in the United States and found no measurable aflatoxin in them From these studies, it appears that the presence of aflatoxins in milk and dairy products is very low and most samples meet the tolerance or action levels established for them The presence of molds and mycotoxins in dairy products and animal feeds will continue to be a concern until the health effects in humans and animals are better understood The control of mold growth in foods and feeds will be important to prevent mycotoxin production New and improved analytical methods will help to monitor the level of mycotoxins in foods and feeds 5.8 Microbiology of Starter Cultures Starter cultures are those microorganisms (bacteria, yeasts, and molds or their combinations) that initiate and carry out the desired fermentation essential in manufacturing cheese and fermented dairy products such as yogurt, sour cream, kefir, koumiss, etc In cheesemaking, starters are selected strains of microorganisms that are intentionally added to milk or cream or a mixture of both, during the manufacturing process and that by growing in milk and curd cause specific changes in the appearance, body, flavor, and texture desired in the final end product Progress in dairy starter culture technology and advances in the scientific knowledge regarding the nature, metabolic activity, and behavior of starter cultures in milk, whey, and other media have provided new and improved starter cultures for the dairy industry Research dealing with plasmid-mediated functions of starter cultures and mechanism of genetic exchange has led to utilization of recombinant DNA and other technologies for improvement of dairy starter cultures, particularly regarding development of bacteriophage-resistant strains In this section, general information about starter bacteria is given Several excellent reviews64-75-342"345 have been published and may be consulted for further details regarding starter bacteria 5.8.1 Terminology The fermentation of lactose to lactic acid and other products is the main reaction in the manufacture of most cheese and fermented dairy products Consequently, dairy starter cultures are also referred to as lactic cultures or lactic starters In the dairy industry, single or multiple strains of cultures of one or more microorganism are used as starter cultures The taxonomy and scientific nomenclature of the lactic acid bacteria have been recently modified, for example, lactic streptococci, S cremoris, S lactis, and S diacetylactis are now classified in the genus Lactococcus and referred to as Lactococcus lactis subsp cremoris, L delbrueckii subsp lactis, and L lactis subsp lactis biovar diacetylactis, respectively However, for the sake of convenience, the older names will be retained here The nomenclature and some distinguishing characteristics of dairy starter cultures are listed in Table 5.19 There are two main types of lactic starters: the mesophilic (optimum growth temperature of about 300C) and the thermophilic (optimum growth temperature of about 45°C) Mesophilic cultures usually contain S cremoris and S lactis as acid producers and S diacetylactis and Leuconostocs as aroma and CO producers Thermophilic starters include strains of thermophilus, and, depending on the product, Lactobacillus bulgaricus, L helveticus, or L lactis Often, a mixture of thermophilic and mesophilic strains is used as a starter culture for manufacturing Italian pasta- Table 5.19 NOMENCLATURE AND SOME DISTINGUISHING CHARACTERISTICS OF DAIRY STARTER CULTURES3 Growth Organism Current Nomenclature Morphology 100C + 450C Type Lactic Isomer Percent Lactic Acid Produced in Milk Fermentations Citrate Metabolism Glucose Streptococcus lactis Lactosoccus lactis subsp lactis GM + cocci Streptococcus cremosis L lactis subsp cremoris GM -f cocci Streptococcus diacetylactis L lactis subsp lactis var diaceytilactis GM + cocci Leuconostoc cremoris L mesenteroides subsp cremoris GM + cocci Streptococcus thermophilus S salivarius subsp thermophilus GM + cocci Thermophilic Lactobacillus bulgaricus L delbrueckii subsp bulgaricus GM + rods Thermophilic D(~) 1.8 + Lactobacillus helveticus L helve tic us GM + rods Thermophilic DL 2.0 + a After Tamine,64 Cogan and Accolas.75 -f = positive reaction by > 90% strains — = negative reaction by > 90% strains (d) = delayed reaction Mesophilic 0.8 Mesophilic 0.8 + Mesophilic 0.8 + + Mesophilic 0.2 + Galac- Lactose tose Mal- Sutose crose (d) + + + + D(-) (d) 0.6 + + (d) NH3 from Arginine Table 5.20 LACTIC STARTER CULTURES, ASSOCIATED MICROORGANISMS, AND THEIR APPLICATIONS IN THE DAIRY INDUSTRY Lactic Acid Bacteria Associated Microorganisms Products Mesophilic Streptococcus lactis, Streptococcus cremosis, S lactis var diacetylactis, Leuconostoc cremosis S lactis var diacetylactis, Penicillium camemberti, P roqueforti, P caseicolum, Brevibacterium linens Cheddar, Colby Cottage cheese, Cream cheese, Neufachatel, Camembert, Brie, Roquefort, Blue, Gorgonzola, Limburger Thermophilic Streptococcus thermophilus, Lactobacillus bulgaricus, L lactis, L casei, L helveticus, L plantarum, Enterococcus faecium Candida kefyr, Torulopsis, spp., L brevis, Bifidobacterium bifidum, Propionibacterium fureudenreichii, P shermanii Parmesan, Romano, Grana Kefir, Koumiss yogurt, Yakult, Therapeutic cultured milks, Swiss, Emmenthal, Gruyere Mixed starters S lactis, S thermophilus, E faecium, L helveticus, L bulgaricus Modified Cheddar, Italian, Mozzarella, Pasta Filata, Pizza cheese filata type cheese Some thermophilic starters, such as those used in Beaufort and Grana cheese, contain only lactobacilli,75 whereas some fermented milks made with thermophilic starters also contain Lactobacillus acidophilus, L bulgaricus, and bifidobacteria for their healthful and therapeutic properties.346 Table 5.20 lists the common starter cultures and their applications in cheese and fermented dairy products The lactic starter cultures are also subdivided into two groups: defined cultures and mixed cultures Defined cultures constitute starters in which the number of strains is known The concept of defined starter culture, mainly pure cultures of Streptococcus cremoris, was developed in New Zealand to minimize the problem of open textures in cheese thought to be caused by CO produced by flavor-producing strains in mixed cultures The application of defined cultures did control the open texture problem, however, and they were prone to slow acid production due to their susceptibility to bacteriophage.75'347 The use of pairs of phage-unrelated strains and culture rotation to prevent buildup of phage in the cheese factory were practiced to minimize the potential for phage problems.75-347 Eventually, the use of multiple strain starter and factory-derived phage-resistant strains was made to control the phage problem.345-347-348 Lactic starter cultures are also categorized based on flavor or gas production characteristics64'75 for example, B or L cultures (for Betacoccus or Leuconostoc) contain flavor and aroma producing organisms, for example, Leuconostoc spp D cultures contain Streptococcus diacetylactis', BD or DL cultures contain mixtures of both Leoconostoc and S diacetylactis strains and O cultures not contain any flavor/aroma producers but contain S lactis and S cremoris strain This nomenclature is commonly used in the Netherlands.349 Often, the lactic starters routinely used in dairy plants without rotation are called P (practice) cultures as opposed to L (laboratory) cultures which have been subcultured in the laboratory The P cultures are not usually affected by their own phages, and unlike L cultures, they can recover following the attack of so-called ' 'disturbing" phage 5.8.2 Function of Starter Cultures 5.8.2.1 Production of Lactic Acid The primary function of lactic starter culture is the production of lactic acid from lactose The lactic acid is essential for curdling of milk and characteristic curd taste of cultured dairy products The manufacturing procedures for cheese and other fermented dairy products are designed to promote growth and acid production by lactic organisms The production of lactic acid is also essential for development of desirable flavor, body, and texture of cheese and cultured dairy products The rate of lactic acid production during the cheesemaking is affected by the temperature, calcium and phosphorus content of milk, the type and amount of starter culture used, etc Lactic acid production also results in a decrease of lactose in cheese and whey The presence of excessive lactose in the cheese is undesirable because it can be metabolized by nonstarter bacteria during ripening and lead to flavor and body defects in cheese The mechanisms of lactose metabolism differ considerably in different lactic acid bacteria,350 Streptococcus lactis employs the phosphoenol pyruvate phosphotransferase system (PES/PTS) to transport lactose which is hydrolyzed to glucose and galactose and metabolized by the glycolysis and tagatose pathways, respectively Leuconostoc spp and thermophilic lactobacilli, on the other hand, transport lactose by a permease system It is hydrolyzed to glucose and galactose by /3-galactosidase and further metabolized Lactose metabolism by different starter cultures is reviewed elsewhere.52'54'75343'351"353 5.8.2.2 Flavor and Aroma and Alcohol Production In addition to production of lactic acid, starter cultures also produce volatile compounds, for example, diacetyl, acetaldehyde, and ketones responsible for the characteristic flavor and aroma of cultured dairy products Flavor-producing starter cultures metabolize citric acid to produce CO2 which is necessary for " e y e " formation in some cheeses Some starter cultures, mainly yeast, produce alcohol, which is essential for the manufacturing of kefir and koumiss 5.8.2.3 Proteolytic and Lipolytic Activities The starter cultures produce proteases and lipases which are important during the ripening of some cheese Protein degradation by proteinases is necessary for active growth of starter cultures as most lactic acid bacteria require amino acids or peptides for their growth Proteinase negative (Prot") strains of lactic starters depends on PrOt+ strains in a multiple strain culture for growth in milk 5.8.2.4 Inhibition of Undesirable Organisms The production of lactic acid lowers the pH of the milk and inhibits many spoilage organisms as well as pathogens A number of metabolites produced by lactic cultures can limit the growth of undesirable organisms, for example, Ibrahim354 reported that lowering the pH with lactic acid in a simulated Cheddar cheese making resulted in the inhibition of S aureus Rapid growth and acid development by lactic acid bacteria suppress growth of many spoilage and pathogenic bacteria Besides lactic acid, production of H2O2 and acetic acid by some starter cultures, particularly those containing Leuconostoc or S diacerylactis, can also inhibit pathogenic bacteria.354 The amount of H2O2 produced by lactic acid bacteria may not be adequate in itself to control undesirable organisms in milk However, it can allow the enzyme lactoperoxidase (LPS) to react with thiocyanate (SNC") and produce hypothiocyanate (OSCN"), which can inhibit various pathogens including S aureus, E coli and Campylobacter jejuni?55 Certain strains of S lactis produce nisin, which is inhibitory to various organisms including species and strains of the genera Bacillus, Clostridium, Listeha, etc However, the application of the nisin-producing strains as cheese starters is limited because of their slow acid production and susceptibility to bacteriophages There is considerable interest in developing nisin-producing cultures that may be suitable for use in the dairy industry Several lactic acid bacteria, particularly streptococci, are capable of producing bacteriocins that inhibit Gram-positive pathogens such as Clostridium or Listeria However, the application of these strains as cheese starters may be limited because they inhibit other closely related strains in a cheese starter 5.8.3 Growth and Propagation Lactic starter cultures are generally available from commercial manufacturers in spray-dried, freeze-dried (lyophilized), or frozen form Spray-dried and lyophilized cultures need to be inoculated into milk or other suitable medium and propagated to the bulk volumes required for inoculating a cheese vat as follows: Stock culture spray dried, freeze dried, frozen Intermediate culture Mother culture Bulk culture Intermediate culture Cheese vat Many larger dairy plants develop their own cultures However, preparing and maintaining bulk cultures requires specialized facilities and equipment Much research and development in the starter culture technology has been aimed at designing specialized growth media for starters, protecting the starter cultures from sublethal stress and injury during freezing, and minimizing the theat of bacteriophage during starter culture preparations The specialized systems for starter culture propagation include the Lewis system, the Jones system, the Alfa-Laval system, etc.64 The Lewis system356 utilizes reusable polyethene bottles fitted with Astell rubber seals and two-way needles The growth medium (10 to 12% reconstituted, antibiotic-free skim milk) is sterilized in the mother culture bottle The stock culture is incubated through a two-way needle by squeezing the stock culture bottle The bulk starter tank used in the Lewis system is pressurized to allow heating of the growth medium in the sealed vessel The top of the tank is flooded with 100 ppm sodium hypochlorite solution to prevent any contamination during the inoculation of bulk starter The Jones system uses a specially designed bulk starter tank.64 Unlike the Lewis system, this tank is not pressurized The bulk starter tank is inoculated by providing the intermediate starter through a special narrow opening and a ring of flame or steam is used to prevent any contamination during the inoculation of bulk starter Recently, a combination of the Lewis/Jones system has been developed in the United Kingdom that improves on the Lewis technique of aseptic culture transfer and economizes by using cheaper, nonsealed tanks as in the Jones system The details of the combined Lewis/Jones system have been described ty Tamime.64 The Alfa-Laval system uses filtered-sterilized air uner pressure, for transferring the culture The mother and intermediate cultures are prepared in a special unit called a "viscubator" and transferred to the bulk starter tank using compressed air.64-357 5.8.3.1 pH Control Systems There are two main reasons for using pH control systems in propagating bulk starter cultures: (1) to minimize daily fluctuations in acid development and thereby prevent "over-ripening" of the starter, and (2) to prevent the cellular injury that may occur to some starters when the pH of the medium drops below 5.0 In the pH control systems, the acid produced by the starter culture is neutralized to maintain the pH at around 6.0 The external pH control system, developed by Richardson et al.,358'359 uses wheybased medium fortified with phosphates and yeast extract The pH is maintained at around 6.0, by intermittent injection of anhydrous or aqueous ammonia, or sodium hydroxide This system has been used successfully in the United States for production of most American-style cheeses The internal pH control system, developed by Sandine et al., 360 " 363 uses a wheybased medium containing encapsulated citrate-phosphate buffers that maintain the pH at around 5.2 Unlike in the external pH control system, no addition of ammonia or NaOH is necessary The internal pH control system is available as the phase (Rhdne-Poulenc—Marschall Products Division) and In-Sure (Chr Hansen's Laboratory, Inc.) and is used in the United States and Europe for a variety of cheeses and fermented products such as buttermilk.64 5.8.3.2 Phage Inhibitory and Phage-Resistant Medium (PIM/PRM) The PIM/PRM were developed following observations of Reiter64 that bacteriophage of lactic streptococci were inhibited in a milk medium lacking in calcium Hargrove 364 reported on the use of phosphates to sequester free calcium ions in milk or bulk-starter medium for inhibition of bacteriophage The effectiveness of phosphates in the formation of PIM/PRM for phage control was confirmed by Christensen 365 " 467 The PIM/PRM consisting mainly of milk solids, sugar, buffering agents such as phosphates and citrates and yeast extract have been widely used in the United States, Canada, and Europe for about 20 years 345 However, the effectiveness of the PIM/PRM in inhibiting bacteriophage and stimulating growth of the starter culture media is somewhat limited, 64 Despite the absence of calcium, some phages can infect the the starter culture at its optimum growth temperature Also, phosphates in the PIM/PRM can cause metabolic injury to some starter cultures The preparation of active bulk starter culture free of phage contamination is essential for cheese manufacturing However, poor practices promoting phage contamination still exist in many commercial operations 345 ' 368 Factors important in bulk starter preparation and ways of minimizing bacteriophage problems in cheese factories have been reviewed by Huggins 345 and by Richardson 368 5.8.4 Inhibition of Starter Cultures The inhibition or reduction in activity of lactic starter culture results in consequences ranging from ' 'dead vat'' or slow vat to production of poor quality cultured products Also, sluggish starter culture produces acid at a slow rate and fails to control spoilage and potentially pathogenic bacteria The primary cause of inhibition of starter cultures is the bacteriophage Control of the bacteriophage problem depends on understanding of critical factors affecting phage infection and growth in lactic starter cultures, 369 factors dealing with bulk starter culture production, factory design, sanitation, and whey processing 345 Lactic starter cultures are very sensitive to antibiotic residues in milk, 171 ' 370 " 372 for example, 0.01 IU/ml of penicillin may inhibit a mesophilic lactic starter and a yogurt culture.64 The sensitivity of starter culture to a specific antibiotic residue depends on the species or strains of the starter culture, the antibiotic preparations, and the test for determining antibiotic concentrations The problem of antibiotic residue is primarily associated with their use in mastitis therapy in the dairy cow and failure to withhold the milk from cows treated with antibiotics This problem is currently receiving much attention in the United States dairy industry Residues of detergents and sanitizers used in the dairy industry for cleaning and sanitation may also inhibit starter culture growth and activity The effects of commonly used cleaning compounds such as chloride, quaternary ammonium compounds, and alkaline detergents on the activity of various dairy starter cultures have been studied in detail 373 - 374 Proper cleaning and sanitation, particularly adequate rinsing, is important in minimizing the inhibition of starter culture growth and activity by residues of cleaners and sanitizers Occasionally, inhibition of the growth of starter culture may be caused by naturally occurring antibacterial compounds present in milk For example, lactin and the lactoperoxidase system (LPS) have been reported to cause inhibition of certain lactic cultures.357'375-376 5.8.5 Genetic Engineering for Improving Starter Cultures Recent advances in the genetics of lactic acid bacteria, particularly progress in our understanding of the basic processes relating to transport, metabolism, and genetic regulation of sugar utilization, bacteriocin production, and phage resistance have created many opportunities for applying genetic engineering techniques for improving dairy starter cultures.12 In the past, fast-acid-producing and bacteriophage-insensitive strains were obtained through natural selection and mutation processes However, many of these strains were unstable due to spontaneous loss of properties, apparently due to the loss of plasmid(s) The understanding of the functional properties of plasmids and of the mechanisms of genetic exchange and gene expression in lactic streptococci will allow the cloning of desirable traits into dairy starter cultures It is now well established that mesophilic lactic starter cultures harbor plasmids of diverse sizes and that some of these plasmids code for several major functions of lactic streptococci (Table 5.21) The knowledge of plasmid-mediated functions and plasmid transfer systems may be used to develop specific starter cultures that may: Table 5.21 PLASMID-UNKED METABOUC FUNCTION OF MESOPHIUC STREPTOCOCCP Function Reference Sugar utilization LeBlank et al 377 Gasson and Davies McKay 55 Gonzalez and Kunka 373 Proteinase activity McKay 5 Kok et al Kempler and McKay Citrate utilization Bacteriocin production Schenvitz et al 381 Scherwitz and McKay Nisin production McKay and Baldwin Bacteriophage resistance McKay and Baldwin 383 Sanders and Klaenhammer 384 Chopin et al 385 Sanders 80 a Adapted from McKay.55'56 138 Polzin, K M., J S Horng, and L L McKay 1991 Construction of a lactococcal integration vector using a plasmid encoding temperature-sensitive maintenance / Dairy Sci 74: (Suppl 1):121 139 Raya, R R., G L De Antoni, D C Walker, and T R Klaenhammer 1991 Construction of a phage 0adh-mediated site-specific insertional vector, and chromosomal integration in Lactobacillus gasseri ADH / Dairy Sci 784: (Suppl 1):122 140 McKay, L L., K A Baldwin, and E A Zottola 1972 Loss of lactose metabolism in lactic streptococci Appi Microbiol 23:1090-1096 141 McKay, L L 1983 Functional properties of plasmids in lactic streptococci Antonie van Leeuwenhoek 49:259-274 142 McKay, L L., K A Baldwin 1978 Stabilization of lactose metabolism in Streptococcus lactis Ql Appi Environ Microbiol 36:360-367 143 Leenhouts, K J., J Gietema, J Kok, and G Venema 1991 Chromosomal stabilization of the proteinase genes in Lactococcus lactis Appi Environ Microbiol 57:2568-2575 144 McKay, L L., K A Baldwin 1990 Applications for biotechnology: present and future improvements in lactic acid bacteria FEMS Microbiol Rev 87:3-14 145 Froseth, B R., and L L McKay 1991 Development and application of pFMOll as a possible food-grade cloning vector J Dairy Sci 74:1445-1453 146 Von Wright, A., S Wessels, S Tynkkynen, and M Saarela 1990 Isolation of a replication region of a large lactococcal plasmid and use in cloning of a nisin resistance determinant Appi Environ Microbiol 56:2029-2035 147 Wessels, S., and G Strandevej 1990 Nisin resistance is a genuinely selectable marker for foodgrade cloning in the lactococci FEMS Microbiol Rev 87:P36 148 Takiguchi, R., K Aoyama, and H Hashiba 1990 Development of food grade host-vector system in Lactobacillus helveticus subsp jugurti FEMS Microbiol REv 87:P10 149 Ross, P., F O'Gara, and S Condon 1990 Thymidylate synthase gene from Lactococcos lactis as a genetic marker: an alternative to antibiotic resistance genes Appi Environ Microbiol 56:21642169 150 Leenhouts, K J., J Kok, and G Venema 1991 Lactococcal plasmid pWVOl as an integration vector for lactococci Appi Environ Microbiol 57:2562-2567 151 Harlander, S K 1989 Food biotechnology: yesterday, today, and tomorrow Food Technol 43:196-206 152 Food and Drug Administration 1990 Direct food substance affirmed as generally recognized as safe; chymosin enzyme preparation derived from Escherichia coli K-12 Fed Reg 55:1093210935 153 Flamm, E L 1991 How FDA approved chymosin: a case history BiolTechnol 9:349-351 154 Simons, G., G Rutten, M Homes, and W M de Vos 1988 Expression and secretion vectors for the production of bovine chymosin in lactic streptococci J Dairy Sci 71 (Suppl 1):83 155 Van de Guchte, M., J Kodde, J M B M Van der Vossen, J Kok, and G Venema 1990 Heterologous gene expression in Lactococcus lactis ssp lactis: synthesis, secretion, and processing of the Bacillus subtilus neutral protease Appi Environ Microbiol 56:2606-2611 156 Somkuti, G A., D K Y Solaiman, T L Johnson, and D H Steinberg 1991 Transfer and expression of a Streptomyces cholesterol oxidase gene in Streptococcus thermophilus Biotech Appi Biotechnol 13:238-245 157 Executive Office of the President, Office of Science and Technology 1986 Coordinated framework for regulation of biotechnology Fed Reg 51:23302-23309 158 Kosikowski, F V 1982 Cheese and Fermented Milk Foods, 2nd edit F V Kosikowski and Associates, New York 159 Food and Drug Administration 1986 Statement of policy for regulating biotechnology products Fed Reg 51:23309-23313 160 United States Department of Agriculture 1991 Proposed USDA guidelines for research involving the planned introduction into the environment of organisms with deliberately modified hereditary traits Fed Reg 56:4134-4151 161 International Food Biotechnology Council 1990 Biotechnologies and food: assuring the safety of food produced by genetic modification Regulat Toxicol Pharmocol 12:S1-S196 162 Harlander, S K 1991 Social, moral, and ethical issues in food biotechnology Food Technol 51:152-161 163 Corey, B 1990 Bovine growth hormone: harmless for humans FDA Consumer 24:17-18 164 MacKenzie, D 1989 Can biotechnology pick up the pinta? New Scientist 124:32-33 165 Roush, W 1991 Who decides about biotech? The clash over bovine growth hormone Technol Rev 94:28-36 CHAPTER Computer Applications: Expert Systems Robert L Olsen 3.1 Introduction, 106 3.1.1 Artificial Intelligence and Expert Systems, 106 3.1.2 Relationship to Traditional Programming, 108 3.2 Knowledge-Based Architecture, 109 3.2.1 Knowledge Representation, 109 3.2.2 Searching and Inference Strategies, 113 3.2.3 Uncertainty, 116 3.3 Building Expert Systems, 117 3.3.1 Feasibility, 117 3.3.2 Knowledge Acquisition, 118 3.3.3 Tool Selection, 120 3.4 Expert Systems and Process Control, 121 3.4.1 Preexpert System Developments, 121 3.4.2 Expert System Applications, 123 3.4.3 Knowledge Representation in Process Control, 126 3.4.4 Commercial Examples, 127 3.5 Business and Manufacturing Operations, 128 3.5.1 Physical Goods Management, 128 3.5.2 Time Management: Planning and Scheduling, 130 3.5.3 Computer Integrated Manufacturing, 132 3.6 Quality Management Applications, 138 3.6.1 Quality Control Programs, 138 3.6.2 Laboratory Systems, 140 3.6.3 Quality Defect Analysis, 142 3.7 Strategic Operations, 143 3.7.1 Simulation, 143 3.7.2 Research and Development, 146 3.7.3 Training, 149 3.8 Future Trends, 150 3.9 References, 151 3.1 Introduction This chapter explains what expert systems are, how they work, and their strengths and limitations Expert systems have four major applications The first is for intelligent process monitoring and control This involves managing a real-time system by interpreting incoming data and taking suitable actions A real-time system refers to a computer system that can respond to incoming data fast enough to continue the process at the desired speed The second application is diagnosis or troubleshooting The main task of a diagnostic system is to locate the cause of an observed defect This type of expert system application has been most widely used The third application is instruction or learning In the area of computer-assisted instruction, expert systems can be used as intelligent tutors The fourth application is design and configuration This involves constructing a solution from a set of components given a set of constraints Complete computer systems are configured or assembled according to a customer's requirements using an expert system Sections 3.4 through 3.8 describe applications of these areas in the food industry However, the sections are organized according to subject areas and not according to the above categories Dairy processing will be emphasized where possible Because expert systems are often added to existing traditional computer systems, descriptions of those traditional areas will be included 3.1.1 Artificial Intelligence and Expert Systems Computers were first used as calculating devices As computer languages developed, computers were used to process symbols representing numbers The symbols could also be text or even concepts Early computers worked well with numbers and physical quantities In performing numerical data operations, the computer had specific instructions on dealing with separate components of data However, not all human thinking and decision-making depended on numbers The computer was aware of positions of data but not of the internal meaning of words The word " i s " in the phrase ' 'temperature is cold" has no significance in its relationship to the other words in the phrase It became apparent that existing tools of numerical reasoning were inappropriate for modeling intelligence In order for artificial intelligence to develop, new languages, or at least new structures, were required that could manipulate symbols and understand meanings of words, as well as numbers Symbolic processing was then developed to model thought and reasoning The term "artificial intelligence" (AI) was proposed by John McCarthy in naming the Summer Research Project at Dartmouth College in 1956 "Artificial intelligence is the study of mental facilities through the use of computational models."1 Areas of AI research include robotics, natural language comprehension, machine vision, and expert systems Expert systems are computer-based processes that can assist a user in drawing conclusions from a set of facts, and in prescribing how the conclusions should be used.2 Expert systems are able to solve problems of scientific and commercial importance using human expertise that has been built into the program One of the earliest expert systems was developed in the 1960s by research workers at Stanford University They were asked by the National Aeronautics and Space Administration to construct a computer program to help analyze the soil chemistry on Mars using a mass spectrometer In order to analyze the spectra, the expertise of a soil chemist had to be encoded into the program Although the system turned out to be too large for the spacecraft, it led to the development of the successful DENDRAL expert system Significant features of DENDRAL distinguishing it from previous expert system attempts were its highly specific rules and narrow area of expertise Another widely known expert system developed at Stanford is MYCIN MYCIN can diagnose bacterial infections and prescribe treatments for them Current applications of expert systems span a wide array of subjects They include topics as diverse as optical fiber cable design, container loading, preventive maintenance of diesel engines, control of a cement kiln, machine operation planning, airline scheduling, forest fire prevention, analysis of computer systems performances, DNA restriction mapping, and legal advisement Major fields of business, science, education, medicine, and agriculture all have had many workers developing and evaluating expert systems The discovery that earlier general purpose reasoning programs were limited led to the development and success of these narrowly focused expert systems Once closely constrained, practical problems were solved with success, it became apparent that basic knowledge representation and reasoning techniques could be useful in solving commercial problems Some critics have suggested that the term ''expert systems" is limiting because the same general methods can be used to create intelligent behavior without the human expert These same critics argue that generally available knowledge or information, rather than expert knowledge, can be accessed and manipulated, and made more useful through the inference process To accommodate this attitude, the term "knowledge-based systems" has been proposed and is often observed in the literature For the sake of consistency the term "expert system" will be used in this text The development of expert systems has enriched traditional programming Expert system procedures are embedded in programs such as symbolic compilers, multiwindow high-resolution displays, and object-oriented software Expert systems use symbolic data rather than numeric data Symbols can take the form of characters or digits Numeric and symbolic processing programs both use symbols, but conventional computer programs use numeric or computational operations Expert system programs differ from conventional programs in that they understand relationships between symbols Databases and word processors appear to be symbolic on the surface, but manipulation of all data is numeric Although expert system programs use symbols that are numbers according to ASCII code, expert system programs involve symbolic concepts and interconnected symbols at a higher level Conventional programs are more rigidly designed With an expert system language or tool, the information-containing portions of the program are kept apart from the inference mechanism This allows portions of the program to be changed without affecting the whole program In addition, relationships can be derived that might exist but that are not apparent from examining data by usual means Programs often reveal relationships unrecognized by human experts Adaptive expert systems can exchange operations in response to past experiences In the case of adaptive expert systems, the logic techniques are similar to a mechanical clock in which the hands are continually adjusted until they finally arrive at the right time Adaptive or self-learning systems derive rules from experience They start with blank rules and guess answers for given input Rules are modified depending on whether they are wrong or right 3.1.2 Relationship to Traditional Programming The strength of an expert system lies more in its knowledge than in new programming techniques With expert systems the data structure is referred to as the knowledge base The difference between a conventional data structure and the knowledge base lies mainly in the complexity and organization of the information being stored Databases store only definite facts Knowledge bases contain definite facts as well as rules and probablistic information Knowledge bases can deal with uncertain information whereas databases cannot Knowledge bases can handle more complex relationships between groups of data In contrast to conventional programs, expert systems are able to use information based on experience or intuition Expert systems can be updated more easily than conventional programs They are also able to provide explanations on their line of reasoning In nonbranching segments of traditional programs, the flow of execution always moves forward The lines of program are executed one after another Although an expert system program also moves from one line to the next, the rules not follow a predefined sequence It would appear that the program can be executed in a forward or a reverse direction Even if the lines of code are not actually read backwards, the concept of independent rules is useful and gives the system much flexibility Because the restrictions of execution flow are reduced or eliminated with expert systems, the difference between input and output data decreases Traditional programs usually have input and output variables such as f(x) = 2x = y where x is defined as the input and y is defined as the output Methods of organizing expert systems data structures include bits, bytes, alphanumeric symbols, words, sentences, and frames Programming designs are comprised of data structures and methods to control computations Whereas in traditional programming these are closely associated, with expert system methods, the data structure is more independent of the controlling element Although the differences between traditional programming and expert systems are emphasized here, expert systems procedures are used in most complex software systems Many regard them as more of a programming tool rather than a separate area of computer science Expert system principles are often embedded in commercial software with no mention of the terms expert system or AI Terms such as 'intelligent" and "smart" are sometimes used instead Any case in which the program reasons about knowledge, applies it, and communicates it to others can be referred to as an expert system The technology is becoming more commonplace More and more it is becoming simply a part of standard programming procedure 3.2 Knowledge-Based Architecture Knowledge is information that allows the expert systems to make decisions An expert system is as powerful as its effective use of knowledge Knowledge is represented in a knowledge structure These structures can be compared to data structures; only instead of data, they store and work with knowledge Knowledge representations include the various ways in which knowledge can be organized and displayed Inference refers to the process of combining facts and rules A visual form of the inference process is to construct a tree of possibilities The branches of the tree show the different paths a line of reasoning may take As much of the knowledge in expert systems is subjective and qualitative, the degree of confidence that can be placed in the knowledge needs to be expressed Dealing with knowledge that is less than totally true or false introduces uncertainty The ability to deal with uncertainty is a major advantage expert systems possess 3.2.1 Knowledge Representation Concepts used in knowledge representation are called objects Once a set of objects is selected, the objects need to be defined and their relations described To accomplish this it is useful to consider basic components of knowledge These include naming, describing, organizing, relating, and constraining.3 Names are used for identification purposes and for clarity Names need to be unique to avoid confusion To make a name more unique, more detail can be added, or a special code or property can be assigned Addresses or telephone numbers are commonly used for this purpose The function of naming is usually performed by nouns Objects can be described by their various properties Because a large number of properties could be found for most objects, they are usually limited according to the particular application of interest The function of describing is usually performed by adjectives or values The main purpose in organizing is to assist in locating information Filing systems have developed over many years to accomplish this purpose A common problem occurs when the same information can fit within different files If a researcher finds information on the effects of milk heat treatment on cheese yield, does it go in a file called cheese yield, or into a file called heat treatment? Expert systems provide methods for organizing knowledge into categories Categories contain objects with similar properties Relationships link objects together Moisture content of cheese is related to fat content of milk The way in which objects are related is the basis for logical rule construction If/Then rules express relationships between objects For example, if ice cream is coarse, then storage temperature is high Relationships may also be established with frames using slots or inheritance A frame is similar to a database record and slots are similar to fields Constraints place limits on values They are useful in limiting the values objects can be assigned For example, percent moisture cannot be greater than a predefined value Constraints help exclude invalid data from the system, such as the same cheese made on two different days Knowledge can be represented in the form of facts, semantic networks, rules, frames, and objects Facts are segments of information A fact combined with an If/Then rule can draw a conclusion Facts may be entered into a knowledge base prior to program execution They can be provided directly through the user interface Facts may also be obtained from external devices such as a temperature probe in real-time and from external databases containing information such as dairy herd records or fill weights A semantic network is a symbolic data structure using nodes and arcs in a graphical notation With this type of notation, each node represents an object or attribute while the arcs represent relationships between the nodes The nodes are drawn as boxes, ovals, or circles The links are drawn as arrows connecting the nodes Figure 3.1 represents a node and arc graph Nodes in one semantic network can connect with nodes in other networks This allows the principle of inheritance to take place A problem with semantic networks is that they can become difficult to manage as their size increases Rules are the most common form of knowledge representation They seem to correspond to the way in which experts use and discuss knowledge Rules are somewhat similar to IF/THEN statements used in conventional programming languages IF (set of conditions) = premise THEN (set of conditions) = conclusion Agitator Boiler Steam Pasteurizer Sanitary MtIk Tank Company Milk Silo Temperature Gauge Figure 3.1 A semantic network with node and arc notation Processing Vat Blue Moon Dairy The conclusion is reached only if the premise is true With rule-based programs a rule interpreter uses a pattern matching procedure to see if the IF conditions in the rules match up with information in the knowledge base Rules are modular and independent in nature They contain many possible paths to other procedures To outline all of the possible pathways showing the relationships between various rules would be difficult using conventional programming Conventional programs are so structured as to be difficult to change Several problems exist with rules Rules may lack variation and are unstructured They can be difficult to use successfully in representing cause-and-effect knowledge because too many rules and too much effort are required to get all effects of the causal model Large numbers of rules are difficult to manage Although rules are considered to be independent, it is true mainly in the sense that they are unsequenced However, when they are written, the expert is often thinking of them in terms of other rules This may create unnoticed relationships A new rule can violate a previously established relationship, resulting in a nonsense inference Frames are templates or patterns for clusters of related knowledge Related items of knowledge are grouped together Frames package both data and procedures into knowledge structures A typical frame is made of a name, parents of the frame, slots and their values, and attached predicates A generic representation and an applied example of frames are shown in Figures 3.2 and 3.3 Frames contain slots that can be filled with values Fields in a database can contain data, attributes, and descriptions Slots in frames contain these and also additional information such as rules, hypotheses about a situation, questions to ask users, graphical information, explanatory information, and debugging information An inheritance method helps organize the frames into parent-child relationships Frames are related through hierarchies A frame for 2% milk can inherit knowledge from a parent frame for raw whole milk Lower frames inherit the knowledge from Frame: Parent: Value: Slot: If needed: Predicate or If added: Predicate Figure 3.2 A generic frame representation Frame: Pasteurized low-fat milk Parent: Raw whole milk Slot: Fat Value: 2% Slot: Protein Slot: Standard Plate Count Value: 3.4% Value: 100 cfu/ml if needed: Predicate Quantity in silo - see inventory If added: Predicate Fat > 2% - hold & restandardize Figure 3.3 An applied frame representation higher frames, but they not actually contain the information Frame slots describe characteristics of objects of the system being designed such as placement, connections, distances, various pertinent data, and operations that can be performed on the object Computational power is added to frames by allowing information to be attached to each frame This information includes instructions about how to use the frame and what to if expectations are not achieved As shown in the example, if the antibiotic slot has a value that is positive, the milk is rejected If the quantity to process slot requires a value, the order database will have to be accessed Generally speaking, constructing a frame-based system is more difficult than constructing a rule-based system Scripts are similar to frames but form more of a scenario or outline of particular events Because a group of activities is associated with a script, results of a specific activity can be predicted An example of a script might be all the events associated with presenting a taste panel An abbreviated script is shown in Figure 3.4 Object-oriented methods are closely related to semantic nets, frames, and scripts Sets of objects are self-sufficient modules that contain the information needed to handle a given data structure The modules are situated in networks or hierarchies enabling rapid inheritances of information from one class to another Objects are denoted by unique names Usually, additional identification is required, such as date, lot number, or variety, to locate a specific individual In describing objects, some or all of these additional properties are required Some properties are shared by more than one object Fat content is determined for many dairy products A property may be of more interest to one individual or department than another Cheese appearance is a critical factor for the quality-control supervisor, but of less importance to the distribution manager After naming and describing objects, it is necessary to organize them into categories A common method of organization is to describe a specific object as it relates to a more general class In the knowledge Codes: Script: Taste Panel Track: Hedonic Props: J•Judge T - Technician Preparation Counter Sink Refrigerator Booths Samples Scorecards Water Entry Condition Results J: Invited J: Has Time J: Wants Compensation J: Is Compensated T: Has Data T: Has Fewer Samples Event Event Event Event Enter Panel Read Instructions Taste Sample Mark Card Figure 3.4 A possible taste panel script domain of a dairy processing plant, different sets of objects can be used The objects involved in cheese quality control would be different than the objects dealing with waste disposal, energy utilization, inventory control, or patron incentive programs As it becomes apparent that some object sets overlap with others it is necessary to define the objects and their interactions An interesting feature of objects is that their relevant properties differ depending on the existing conditions The pertinent properties of milkfat in premium ice cream are different from those of unwanted milkfat residues in milk lines Also, basic objects can be modified according to current interests A generic object for frozen dessert can be modified for reduced fat, alternative bulking agents, or other novel ingredients An object-oriented taste panel is shown in Figure 3.5 Because of their similarities, the trend has been for object-oriented and framebased systems to combine their strengths into one structure 3.2.2 Searching and Inference Strategies Inference is the process by which new facts are derived from old facts The inference engine contains the inference and control strategies of the system It combines facts and rules to arrive at conclusions The inference engine tries to establish whether a goal statement is true or false It needs to decide the order in which rules will be processed at each stage of the reasoning process When a consultation with an expert system is begun, the inference engine searches the knowledge base to see if it can Mike Technician (send message) Need Sample (send message) Read Instructions (send message) Bring Sample (send message) Finished Judge Judge Ellen (send message) Compensation Technician Technician Joan Cashier Ron Cashier Figure 3.5 An object-oriented taste panel session reach a conclusion and make a recommendation The pathway the inference engine will follow is not known in advance It depends on the response given back to the questions the computer generates A series of rules can join together to form a line of reasoning Graphically, this can have the appearance of a network structure or a tree The line of reasoning leads to a goal or a fact Viewing problems in a network array or decision tree is useful because of the complexity of many problems It helps to illustrate the problem clearly It soon becomes apparent that there is more than one pathway to the goal All possible lines of reasoning are called a network Networks are observed in many situations An example would be a dairy delivery route (Fig 3.6) A delivery network would connect various stores and warehouses around a city and neighboring communities The connections can vary just as the connection for a network or a rule tree seem to vary In designing the inference strategy, two main approaches could be taken: forward chaining and backward chaining A forward chaining strategy starts at the plant and continues until the last destination is reached If the line of reasoning begins at the last delivery point and works backwards toward the starting point, it is called backward chaining Forward chaining or goal-oriented reasoning is carried forward from available facts It is expected that the deduction of new facts will eventually lead to East High School Fred's Market Highland Elementary Ella's Delicatessen Sandy's Daycare Savemore Foods University Hospital Figure 3.6 Components of a dairy delivery route the goal The inference engine cycles through rules until one is found whose premise matches a true fact Forward rules try to prove goals in their premise IF cheesemaker needs flake salt (goal) THEN receiving buys flake salt (if successful adds this to fact base) If no direction or a very general goal is provided in forward chaining, poor efficiency is observed An advantage of forward chaining is that goals can be generated as conclusions are found Forward chaining is also advantageous where the system has to interpret a set of incoming facts Often the facts are supplied interactively by the user Backward rules try to prove goals in their conclusion receiving buys flake salt (begins with goal) IF cheesemaker needs flake salt (tries to prove—becomes another goal) Backward chaining takes a goal as a hypothesis and tries to prove subgoals by working backward from the goal Each subgoal becomes a hypothesis during the reasoning process The THEN or ELSE part of the IF/THEN statement of the rule is checked first to see if it matches the desired goal The premise of the rule is then examined to see if its truth can be deduced from the knowledge base Backward chaining helps to solve diagnostic problems in which the conclusion is known and the causes sought Suppose the objective is to efficiently find a route from University Hospital to Sandy's Daycare (Fig 3.6) One could either work forward from University Hospital or work backward from Sandy's Playhouse Using forward chaining there is only one path from University Hospital to Sandy's Daycare With backward chaining, there are several paths originating from Sandy's Daycare Because these other paths not lead to University Hospital it makes no sense to pursue them unnecessarily Consider a second route going from Fred's Market to Highland Elementary Departing from Fred's Market in a forward direction can result in a useless sidetrack down to University Hospital Working backward from Highland Elementary will result in fewer distractions From Sandy's Daycare, if Savemore Foods and Ella's Delicatessen are examined, they will be found incorrect and routes beyond them will not be pursued The decision of which pathway and strategy to follow depends on the connections, destination, origin, and all the constraining factors In the delivery example, a number of route patterns or networks can be constructed Constraining factors include speed limits, one-way streets, preferred delivery times, size of delivery, changing customer base, other stops, traffic conditions, size of truck, size of load, and distance The inference engine decides which pathway is best by deciding which is shortest, most efficient, and most accurate 3.2.3 Uncertainty An expert system that mimics human intelligence in real-world situations must be able to deal with uncertainty Information that is incomplete leads to uncertainty Although knowledge can be improved and made complete, much knowledge is inherently imprecise Areas such as the medical, biological, and agricultural fields suffer from having too little data and too much imperfect knowledge Rigorous probability principles no longer apply Because a valid statistical record of data is required to utilize probability theory, alternatives need to be considered These include certainty theory and fuzzy logic With certainty theory inexactness is represented as a confidence factor between and 100 The use of these values to indicate partial truth is known as fuzzy logic Often the value has to be estimated Consideration of the context of the relationship is important Old as it relates to Cheddar cheese is much different compared with old relative to Cottage cheese Also, the relationship between sharp flavor and age is arbitrary Although months or perhaps months is defined as the age for a sharp cheese, samples will vary considerably Because judges' opinions may differ and may change, it is necessary to allow for uncertain decisions Uncertainty values are determined for specific events and then combined to arrive at an uncertainty value for compound or complex events Once a statement is assigned a confidence level there are more precise ways of combining and weighing statements that have differing confidence levels One method of combining uncertainty is to use Bayes's Theorem Although not valid as a statistical probability approach to this application, the theorem is useful to simply combine inexact values The likelihood of one hypothesis over another can be based on the strength of the given evidences using Bayes's Theorem The formula is given as follows: LR(H:E) = P(E:H)/P(E:H') where LR is the likelihood ratio, H is a particular hypothesis, E is the evidence or event, and H' is the false hypothesis Stated in words, the probability or likelihood of the hypothesis given an event is equal to the probability of the event given the hypothesis is true, divided by the probability of the event given the hypothesis is false For example, the probability of developing high-moisture cheese (hypotheses), knowing the starter culture is slow developing acid (event), can be calculated if the probabilities of the event, given the true and false hypotheses, are known The false hypothesis is that the cheese is not high in moisture The theory behind the statistics used in decision-making with expert systems has been widely discussed Because no rigorous theory has been developed in AI for decision-making, expert system decision procedures are often merely a combination of logical inference and probability theory Techniques used to deal with uncertainty in expert systems have not yet employed the most satisfactory statistical methods.4 However, AI is increasingly turning to statisticians in seeking solutions to problems in uncertainty 3.3 Building Expert Systems 3.3.1 Feasibility Although expert systems are useful for many applications, there are some problems that are better solved using a conventional program In addition, there are very complex or abstract applications in which computer utilization is unpractical In order to determine if an application is suitable for an expert system to analyze, the following criteria can be considered: (1) the problem solution requires cognitive reasoning, which took a human expert at least 10 years to acquire: (2) the problem domain is self contained and the boundaries are well defined; (3) the problem usually involves the application of more variables than an average human can retain in active memory at one time; (4) an expert exists and is available; (5) the expert's knowledge still has value and may soon be lost to the company; (6) there is a measurable incentive to the company, which may relate to accuracy, timeliness, consistency, or training of new employees After the potential application of an expert system of a problem is determined, it is useful to consider the advantages and disadvantages of carrying out the project This procedure is desirable for persuading management to support the project and to protect against unrealistic expectations Several advantages of expert systems include the following: Limited or difficult to access expertise can be captured The problem-solving knowledge of a group can be combined into a computerized system The process of constructing the knowledge base helps organize and formalize expert knowledge and decision-making processes The consistency of the human improves by the expert becoming properly organized The expert becomes more repeatable and consistent [...]... level as in nonfiltered milk Hickey et al 55 0 reported that strains of Lactococcus lactis subsp lactis and L lactis subsp cremoris produced more lactic acid in UF retentates of 5: 1 and 2 .5: 1 compared to whole milk Although more acid was produced, the corresponding pH values did not decrease accordingly Other researchers reported similar results.229 -54 9 '55 155 2 The increased buffering capacity in the... riboflavrn, niacin, pantothenate, and biotin by 85, 71, 87, 82, and 84%, respectively .55 3 Free amino acids also decreased by 50 to 98% in five-fold retentates Mistry et al 55 4 found that neither mineral nor vitamin B addition to 2- to 2.4-fold retentates produced significant increases in lactic production by L lactis subsp cremoris or L lactis subsp lactis Qvist et al .55 5 made Havarti cheese from UF five-fold... skim milk powder .56 8 Drew and Manners569 showed that processing at 50 to 55 °C reduced the bacterial population in RO concentrates; however, psychrotrophs grew at about the same rate in RO and raw milk at 5 C Cromie et al .56 6 reported that preheating milk to 50 0C before RO of 2:1 reduced the psychrotrophic, proteolytic, lipolytic, and coliform bacteria, and yeasts and molds by 16 to 50 % In RO concentrates... membranes .56 1 "56 5 Bisulfite was not an effective sanitizer of unclean membranes because it needed a pH of 3 .5 which resulted in corrosion and pitting of stainless steel fittings and rubber gaskets .56 2 Even if the membranes were clean, none of the sanitizers (50 ppm available chlorine, 0.2% hydrogen peroxide, acid anionic surfactant at pH 2 .5) were completely effective because of circulation problems .56 3... processing of milk and found that the DX2\°c w a s 0.6 and 58 s, respectively Z values were 10 and 9.6 K, respectively D values of Bacillus stearothermophilus spores in sterilized milk were 22.4 , 3 .5, and 0.37 min at 1 15. 5°C, 121.10C, and 126.6°C, respectively .57 9 The spores for these experiments were produced at 55 °C in trypticase soy broth (pH 7.1) with 25 ppm of calcium, 31 ppm of iron, 30 ppm of manganese,... buffering capacity .54 4 Growth of enteropathogenic E coli could be prevented in Camembert cheese if milk was preacidified to pH 5. 9 and an active starter was used5 45 or if partial fermentation followed by diafiltration to reduce the buffering capacity was used .54 6 Salmonella typhimurium var Hillfarm grew in retentate concentrated twofold at 7 and 100C, but S aureus grew only at 100C .54 0 When grown in... 15 C the counts increased to 104 to 106 yeasts and molds/g .51 9 In several surveys of yogurts, yeasts belonging to the genera Candida, Kloeckera, Kluyveryomyces, Pichia, Rhodotorula, Saccharomyces, and Torulopsis have been isolated.486 '52 0 "52 2 KcKay523 isolated Yarrowia lipolytica from yogurt In all of these studies few of the isolated yeasts were able to ferment lactose Only Kluyveryomyces species520 "52 2... increased solids content compared to skim milk kept the pH higher 55 755 8 Addition of 1% yeast extract to the permeate stimulate growth of the Lactococcus spp Cheddar cheese whey permeate was used successfully to propagate strains of L lactis subsp lactis and L lactis subsp cremoris over several transfers for Colby cheese manufacture .55 7 The pH and bacterial count from Colby cheese made with a two-fold... problems Burton574 reviewed 35 years of research and development in UHT processing of milk and dairy products The bacteriology of UHT processing, especially resistance of spores to high temperatures, has been reviewed by Brown and Ayres5 75 and Burton .57 6 Two major concerns of UHT processing of milk and dairy products are the heat resistance of bacterial spores and bacterial or native enzymes, particularly... costs Reinbold and Takemoto 559 showed that Bacillus megaterium, Rhodopseudomonas sphaerroides, and Kluyveromyces fragilis could reduce the BOD of permeate from 15, 500 mg/L to 158 0 mg/L Further research is needed on the reduction of BOD in permeate by bacteria and yeasts Another concern of using UF technology is the ability to properly clean and sanitize the membranes after use .56 0 Several reports have