LAIT-XPERT VACHES: An Expert System for Dairy ... - Semantic Scholar

LAIT-XPERT VACHES: An Expert System for Dairy Herd Management D. PELLERIN

Departement de zootechnie R. LEVALLOIS

Departement d’economie rurale G. ST-LAURENT

Departement de zootechnie J.-P. PERRIER

Departement d’economie rurale Universite Lava1 Quebec, PQ, Canada G1K 7P4

ABSTRACT

An expert system called LAITXPERT VACHES, developed to evaluate technical management of dairy enterprises, was tested using case data. The expertise of the system was provided from information obtained from interviews of three dairy management or nutrition experts. LAIT-XPERT VACHES contains over 950 rules and runs on IBM-compatible personal computers. It calculates objectives in milk production, fat and protein production, feeding cost, reproduction, and other areas. In addition, it detects problems and high performance according to these objectives; researches the causes of problems in herd management, feeding, genetics, health, housing, and other areas; and lists conclusions by sector. Using a monthly report of 10 farms registered in the DHI program of Quebec, LAIT-XPERT VACHES issued 92.3% of the conclusions also issued by experts. However, the experts revealed only 53.3% of conclusions reached by the expert system. With Agn-Lait reports of three farms, all conclusions of LAITXPERT VACHES were validated by the experts. These results demonstrated that use of an expert system makes it possible

Received June 21, 1993. Accepted March 14. 1994. 1994

J Dairy Sci 77:2308-2317

to obtain analyses of dairy performance data equivalent to those of human experts. Wey words: expert system, dairy herd management) Abbreviation key: PATLQ = Programme d’analyse des troupeaux laitiers du Q u k c (Qutbec DHI program). INTRODUCTION

During the last decade, an increasing amount of data has been collected on dairy farms, resulting in more complex reports, which are difficult to evaluate by farmers or even by dairy advisors. The use of an expert system, a new technology derived from artificial intelligence, could be very useful as a decision support system. In agriculture, expert system research was initiated in the early 1980s, and an important number of applications have emerged or are being studied (1, 2, 3, 5 , 6, 8, 11, 12, 13, 15, 16, 17, 18, 20, 21, 23, 24). Expert systems have not been applied yet on dairy farms, but some prototypes exist or are under development for diagnosis of mastitis (l), for evaluation of reproductive management (7), for detection of stray voltage problems (9), for detection of ventilation problems (22), for detection of milking equipment problems (4), for economic and financial evaluation of farms (14), and for management support (10). Schmisseur and Gamroth (19) applied expert system technology to provide manage2308

EXPERT SYSTEM FOR DAIRY HERD MANAGEMENT

ment advice to dairy operators. In this paper, we report another pioneering effort to apply expert system technology to technical management of dairy enterprise. MATERIALS AND METHODS

An expert system for technical management of dairy enterprise, LAIT-XPERT VACHES, was developed using the MIMI shell program (Institut National de la Recherche Agronomique, Grignon, France). This expert system runs on standard configured IBM-compatible personal computers with 640 K of base memory. Knowledge base and expertise were provided by three dairy management experts from the private and public sectors and by our own research for some technical criteria. A series of individual interviews with these experts allowed us to identify various concepts (criteria and problems) that they used to evaluate dairy herd performance. Conceptualization of the knowledge gathered during those meetings resulted in the development of a prototype. A second series of interviews, during which all three experts were present, allowed us to define precise thresholds for all criteria used and the ways to represent the relationships among these criteria. After each interview, new criteria were added to provide an expert system prototype ready to be validated or evaluated. Information needed for such evaluation was obtained from monthly reports of Programme d'analyse des troupeaux laitiers du Qutbec (PATLQ), the DHI program of Quebec, or from reports produced by Agri-Lait or Conseil-Lait software (Agri-Gestion Laval, Universitt Laval, QuCbec, PQ, Canada). As shown in Tables 1 to 6, LAIT-XPERT VACHES analyzes two kinds of criteria: 1) criteria allowing the detection of problems in six main sectors of cows management and 2) criteria identifying potential factors causing those problems. The method used by LAITXPERT VACHES to analyze performance data of the cows can be described by four steps: 1) the establishment of short-term and long-term objectives for each sector, namely, milk production, fat production, protein production, feeding cost, reproduction, and other (herd weight, culling age, and labor efficiency); 2) the detection of problems according to the objectives fixed in the first step; 3) the research

2309

of factors such as herd management, feeding, genetics, health, housing, and other factors (such as season and growth stage) causing the problems detected; and, finally, 4) summary and arrangement of conclusions by sector. Because feeding (including grouping) is a key

TABLE 1. Main criteria used by LAIT-XPERT VACHES to analyze milk production. Criteria Criteria allowing for the detection of problems Long-term critena Milk production (kp/yr per cow) Milk production change (%) BCA' Milk BCA Milk change (a) Short-term criteria Production drops (S) Persistency (76) Criteria to he considered as causes of problems Management Cows in milk (%) Milking times interval 01) Feeding All criteria in feeding verification Health Milk fever (%) Acetonemia (a) Displaced abomasum (S) Dystocia (%) Retained placenta (8) Metritis (9%) Cystic ovaries ('5%) Twins (%) Calves horn dead (%) Feet problems (a) Genetics Average cow weight (kg) Artificial insemination (7%) Culling rate (S) Growth Average heifer weight (kg) Age of first calving (mo) Average 6- to 12-mo-old heifer weight (kg) Reproduction Calving interval (d) Days dry (4 Other criteria Average age of herd QT) Milking equipment condition Presence of stray voltage ]Breed class average. Journal of Dairy Science Vol. 77, No. 8, 1994

2310

PELLERIN ET AL.

TABLE 2. Main criteria used by LAIT-XPERT VACHES to analyze fat production. Criteria allowing for the detection of problems

Criteria to be considered as causes of problems

Long-tern criteria Annual fat (8) Fat change (I) BCA' Fat BCA Fat change (8)

Production Milk (kg/yr per cow)

Short-term criteria Test day fat (%)

Feeding Sequence of meals ADF (%) NDF (9%) NDF from roughage (5%) Roughage ingestion (% of BW) Concentrates (%) Number of concentrate meals CP IndexZ S Index2 Genetics Choice of bulls Other criteria Season (month of the test) Lactation stage (d in milk) Milk sampling on test day Error in laboratory analyses

'Breed class average. 2Ratio of intake over requirement.

factor in dairy herd management, even though it may be a cause of problems identified in each sector, its evaluation is made systematically at the end of the analysis of sectors. Main criteria used by LAIT-XPERT VACHES to analyze milk production are listed

in Table 1. Milk production, milk breed class average, and their change are used to define whether long-term milk production problems exist in a herd. Short-term problems in milk production are identified by production decline (marked decreases in production below the ex-

TABLE 3. Main criteria used by LAIT-XPERT VACHES to analyze protein production. Criteria allowing for the detection of problems


Long-term criteria Annual protein (7%) Protein change ( I ) BCA' Protein BCA Protein change (8) Protein: fat

Production Milk yield (kg/yr per cow)

Short-term criteria Protein on test day (%) Proteiwfat

'Breed class average 2Ratio of intake over requirement Journal of Dairy Science Vol. 77. No. 8. 1994

Feeding Fat in the ration (%) Undegradable protein index2 NEL Index2 Body condition Genetics Choice of bulls Other criteria Season (month of the test) Lactation stage (d in milk) Preservation of milk samples Error in laboratory analyses


231 1

TABLE 4. Main criteria used by LAIT-XPERT VACHES to analyze feed costs. Criteria allowing for the detection of problems Long-term criteria Annual feed cost ($4Il) Feed cost change (9%) Milk-feeding cost value (Vyr) Milk-feeding cost value change (%) Milk from roughage (kg/yr per cow) Milk from roughage change (8) Short-term criteria Feed cost on test day ($411)

Criteria to be considered

as causes of problems Management Cows in milk (%) Feeding Number of groups Roughage quality Roughage ingestion (% of Bw) Milk from concentrate (kgkg) All criteria in feeding verification Health Milk production sector Genetics Average herd BW (kg) Reproduction Dry period length (d) Calving interval (d) Economic criteria Average price of concentrate (%/tonne)

TABLE 5 . Main criteria used by LAIT-XPERT VACHES to analyze reproduction. Criteria allowing for the detection of problems Long-term fertility Calving interval (d) Days open (d)

Herd fertility Breedings per cow (no.) Breedings per cow change Conception rate (8)

Other criteria Culling for reproduction (%) Days dry (d)

Criteria to be considered

as causes of problems Reproductive management Days to first estrus Days to first breeding Other criteria Herd fertility (see following criteria) Feeding at beginning of lactation ENL Index' Body condition CP and DP2 Indexes1 13 and P Indexed Mg and Cu Indexes' Health Milk production sector Genetics Bulls with low fertility Other criteria Detection of estrus Time of insemination Embryos transfer Safety breeding noted as normal

Same criteria than herd fertility Production and reproduction Calving interval (d) Milk yield (kg/yr per cow)

IRatio of intake over requirement. *Degradable protein. Journal of Dairy Science Vol. 77. No. 8. 1994

2312

PELLERIN ET AL

pected lactation curve) and herd average persistency (ratio of actual test production over previous test milk production). If a production problem is detected, the potential causes of this problem are evaluated. Specifically evaluated are management, feeding, health, genetics, growth, reproduction, and other factors (Table 1). Fat production is analyzed mainly with the criteria listed in Table 2. Annual fat percentage, fat BCA, and their change are the criteria used to detect a long-term problem, and test day fat percentage acts as an indicator of shortterm problems. The causes of a fat problem are searched among production, feeding, genetics, and other factors (Table 2). As with the fat production sector, a problem in the protein sector is detected with annual protein percentage, protein BCA, their change, and protein percentage on test day. The protein to fat ratio is also used. Production, feeding, genetics, and other factors are evaluated as potential causes of the problem (Table 3). Main criteria used to analyze feed costs are listed in Table 4. Annual feed costs, milk minus feeding cost value, milk from roughage, and their change are the criteria used for the detection of a problem. Then, if a problem is detected, management, feeding, health, reproduction, and economic factors are considered as potential causes of this problem (Table 4).

The reproduction sector is split into two subsectors: long-term fertility, diagnosed by calving interval and days open, and herd fertility, diagnosed by breedings per cow and conception rate. Reproductive management and herd fertility are evaluated as factors of longterm fertility problems (Table 5). Feeding at the beginning of lactation, herd health, genetics, and other factors are then considered as potential causes of low herd fertility (Table 5). Culling for reproduction and days dry are also evaluated in the reproduction sector. The sixth area, “other sectors”, includes the analysis of some important sectors that cannot be included in previous sectors, such as cow weight, age, and culling rate and labor efficiency (Table 6). Finally, as an important source of problems, feeding is systematically verified (Table 7). Three thresholds were defined for every criterion by the experts: 1) an optimal threshold, herd performance should reach it over the long term; 2) an acceptable threshold, performance is satisfying but can be improved; and 3) a critical threshold, performance is wrong and has to be corrected. As an example of the annual feed cost, experts decided to consider the mean of the top 20% PATLQ herds as optimal threshold, the mean of 20% following as acceptable threshold, and the mean of all herds as critical threshold. Performance data from farms are compared with these

TABLE 6. Main criteria used by LAIT-XPERT VACHES to analyze other sectors Criteria allowing for the detection of problems


Weight Average herd weight (kg)

Growth Average heifer weight (kg) Average height at fmt calving (cm) Reproduction Reproductive management criteria

Age and culling rate Average age of herd (yr) Cows culled (Q)

Work efficiency Cows per labor unit (3000 h) Milk per labor unit (3000 h) Journal of Dairy Science Vol. 77. No. 8. 1994

Health Milk production sector Other criteria Size of stalls Lack of exercise Sale of animals for production (To) Production Milk (kdyr per cow)


2313

TABLE 7. Main criteria used by LAIT-XPERT VACHES to verify feeding Criteria to be considered as causes of problems

Critena allowing for the detection of problems ~~

Grouping Number of groups Cows repartition (9%) Days from calving thresholds Milk thresholds (kg)

Feeding Concentrate and roughage meal (no) Roughage quality Feeding rate (% of BW) Body condition Feeding index’ of 22 nutrients ADF (%) and NDF (76)

‘Ratio of intake over requirement

Figure 1 presents an overview of the transit thresholds. By choosing a coefficient from 1 to 5, depending on priority, users can adjust ob- of information in LAIT-XPERT VACHES. Injectives calculated by LAIT-XPERT VACHES formation obtained from the PATLQ or from for the main sectors. If the coefficient chosen Agri-Lait can be entered manually or eventuis 5 for the sector analyzed, the plausibility of ally captured directly by a data acquisition the conclusions is 0 when performance data are program. Therefore, objectives are calculated greater than optimal thresholds; there are no according to the particulars of the farms. Data problems in this case. If data are lower than and objectives can be printed or sent as an the critical threshold, the plausibility of the ASCII file to be captured and analyzed by the conclusion is 100; this criterion is a real prob- expert system. Finally, conclusions and their lem. Finally, if data are between the acceptable plausibility are listed by sector and can be and the critical threshold, the plausibility of consulted on the screen, printed, or saved in a the conclusion varies between 0 and 100. file. In LAIT-XPERT VACHES, users can However, if the user chooses the coefficient 1, choose which sectors are to be analyzed. It is LAIT-XPERT VACHES compares data with also possible to correct, if necessary, some acceptable, rather than optimal, thresholds. In objectives calculated by the expert system and the case presented previously, if the coefficient judged unadoptable for the farm being tested. for costs is set at 5, the threshold is the top Some criteria not available on some reports 20% of PATLQ herds, but, if it is set at 1, the and useful to data analysis (milk produced threshold is the following 20%. from roughages, days from calving to first The ability of LAIT-XPERT VACHES to breeding, and total amount of concentrate) can analyze dairy herd performance and to issue be estimated with other criteria by the expert conclusions is contained in some 950 in- system. dividual rules. The great number of rules reConclusions can point out low performance; quired that two knowledge bases be created. therefore, they indicate thresholds to be The first base was devised to calculate objecreached. Conclusions can also point out high tives, and the other base was devised to anaperformance if data are greater than the oplyze cow performance data. Because the rules (-50 quasirules) used in the first knowledge timal threshold. If neither low performance nor base were generally forward chaining and high performance occurs in a main sector, without fuzzy factors and plausibility, they “nothing to report” is given. Finally, concluwere programmed in Basic Professional De- sions can point out factors to be considered as velopment System 7. l (Microsoft Corp., Red- potential causes of the problem, but not exammond, WA) to limit the number of rules. The ined by the expert system because of the absecond knowledge base has 905 rules used to sence of criteria evaluating them in data perform data analysis and a dialogue file to reports. Such criteria should be verified using make the conclusions easier to read. LAIT- other sources of information. LAIT-XPERT VACHES was tested using XPERT VACHES has more than 1.5 MB of programming (data acquisition program, rules case data in two experiments. In the first, a monthly report of 10 randomized herds regisbase, and dialogue file). Journal of Dairy Science Vol. 77, No. 8. 1994

23 14

PELLERIN ET AL.

obpctives

Report data and obpcwes

data

EXPERT SYSTEM

Of

data

Production Control (Agn-Lait or PATLQ)

Figure 1. Overview of transit information in LAIT-XPERTVACHES

tered in PATLQ and representing breeds present in the QuCbec cattle population (two Ayrshire, one Canadian, five Holstein, one Jersey, and one Brown Swiss), were analyzed by each of the three experts. Also, the experts noted the time spent to analyze farm data. In another meeting, all of the conclusions of an expert had to be validated by the two others. Then, the experts had to validate the conclusions issued by LAIT-XPERT VACHES using the same reports. The second experiment was carried out to test LAIT-XPERT VACHES for some criteria not included in PATLQ reports (such as protein degradability and microelements). Agri-Lait reports of three farms were then analyzed by LAIT-XPERT VACHES. Reports and conclusions of these analyses were mailed to the experts who had to validate them. RESULTS AND DISCUSSION

The results of the first experiment are presented in Table 8. Experts noted 183 conclusions, about 18 conclusions per farm. Seventy-one conclusions (38.8%) related to the feeding sector of herd management. Production and reproduction sectors followed with 37 (20.2%) and 31 (16.9%) conclusions, respectively. Other sectors and cost sector had 22 (12%) conclusions each. Among the experts’ conclusions, LAIT-XPERT VACHES issued 70.5% of them, missed 27.9% of them, and pointed out 1.6% in opposition to them. The sectors showing the highest number of similar conclusions were reproduction (93.5%) and Journal of Dairy Science Vol. 77, No. 8. 1994

cost (81.8%). Feeding and other sectors appeared to be analyzed with less precision than others, showing 56.3% and 63.6%, respectively, of similar conclusions. Some differences between the conclusions of the expert system and of the experts may be explained. Some information was not retained in the conceptual phase in the expert system development to limit its knowledge sector. After considering the conclusions of the experts issued from information available in the expert system, LAIT-XPERT VACHES reached 82.7% similar conclusions and only 1.9% opposite conclusions. Thus, most of the conclusions reached by the experts were also reached by the expert system prototype. The 27 different conclusions were therefore examined by the experts. The experts revealed they made a mistake for 15 of them, and the expert system made a mistake for 12. According to this evaluation, 92.3% of the conclusions of experts were issued also by LAIT-XPERT VACHES, 7.1% were not, and .6% were opposite. The in ability of the expert system to reach some conclusions may be explained by the analysis method used to develop the expert system; when no problem existed in a sector, the expert system did not examine criteria of this sector, but a human expert could have noted any result out of his or her standards. The human experts drew wrong conclusions, but 10 of these 12 wrong conclusions were drawn for Canadian, Jersey, and Brown Swiss herds. The experts do not usually analyze results for these breeds and tended to use


m m *

rJ

23 15

F.

NF.

d

\49?

z-'O

Eo-

(Ci

9

e-?? m d m m

Z-N

m

. I

m o m

* f :

Journal of Dairy Science Vol. 77, No. 8, 1994

23 16

PELLERIN ET AL

thresholds adaptable for Holstein or Ayrshire herds, which are often too high for other breeds. LAIT-XPERT VACHES listed 433 conclusions, about 43 conclusions by enterprise, but each conclusion contained less information than that of the conclusions of experts. The information included in the 183 conclusions of experts corresponded to 231 conclusions of LAIT-XPERT VACHES. The experts identified 53.3% of the conclusions by LAITXPERT VACHES, but failed to identify 46% of them. The closest agreement between the experts and the expert system was for reproduction (61.9%) and cost (57.7%) sectors. The number of conclusions not noted by the experts was high, but it is difficult for us to estimate the relevance of these conclusions. Perhaps human experts can note only the most important factors. Furthermore, when conclusions contained closely related information, such as conception rate and number of breedings per cow, the experts tended to utilize the most relevant ones. The time required by the experts to analyze the farm data was about 15 min per farm (19.4, 15.0, and 9.0 min). Conversely, LAIT-XPERT VACHES took 6.5 min with a 5-MHz PC, 3.5 min with a IO-MHz 286, and 2.7 min with a 20-MHz 386sx. Thus, the time required by the expert system was about four times less than that required by the experts. However, typing of data on the keyboard can take an additional 15 to 30 min, although the automatic capture of data would shorten time substantially. This first experiment showed that expert system technology could be used to analyze performance data of dairy herds and to obtain equivalent and sometimes better results than those of experts. The systematic analysis of an enterprise’s performance data appears to be of great interest. In the second experiment, using Agri-Lait reports of three farms, almost all 172 conclusions issued by LAIT-XPERT VACHES were validated by the three experts (Table 9). Only one conclusion was accepted by only one expert. Very low plausibility (10%) corresponding to this last conclusion may explain the disagreement of the experts. The data were very close to the acceptable threshold. Furthermore, the analysis was made with an adjustment coefficient of 5 , which is very severe. The results obtained in this second experiment Journal of Dairy Science Vol. 77, No. 8. 1994

TABLE 9. Validation of conclusions of LAIT-XPERT VACHES issued from three Agri-Lait reports by the three experts. Conclusions Sector Production Feeding COSIS Reproduction Other sectors Feeding Total

Expert system

(no.) 34 21 6 9 102 172

Validated by the the three experts

(no.) 34 21 6l 9 102 I72

(%I 100 100 100 100 100

100

1Two experts disagreed for high calving interval issued from famn 2 (plausibility of this conclusion was 10).

leave no doubt as to the reliability of the analysis made by the expert system. From these results and those of Experiment 1, it seems that differences between the expert system and the human experts come mostly from the limited expertise of the expert system. CONCLUSIONS

Some characteristics of the expert system show obvious advantages over human experts. The ability of the expert system to cany out systematic and structured analysis makes it an ideal tool at the service of all dairy herd managers. After the information is gathered on dairy farms, expert system technology allows for more effective use of this information. An expert system such as LAIT-XPERT VACHES may be very useful for herds of less popular breeds or for farms located in countries having deficient expertise in dairy production. The experts have undeniable advantages over expert systems in broad expertise and synthesis ability. The current study was the first of its kind to evaluate the application of expert systems for technical management of dairy herds. Other studies are necessary to extend the expertise of the expert systems (for example, in heifer growth), to improve the pertinence of its conclusions, and to find ways to summarize them. ACKNOWLEDGMENTS

The authors thank the three experts, C. Bachand, B. Farmer, and J. Jalbert, for their


great contribution and S. Rioux for help with manuscript preparation. Financial assistance from Agri-Gestion Lava1 is acknowledged. REFERENCES 1 Benas, B. F. 1986. Les systtmes experts: une application au diagnostic tpidkmiologique des mammites en tlevage bovin laitier. Thtse Doct. Vtt., Ecole Natl. Vtt., Toulouse, France. 2 Blancard, D., A. Bonnet, and A. Coleno. 1985. TOM, un systbme expert en maladies des tomates. P.H.M. Rev. Hortic. 261:7. 3Bourgine. R. T. 1985. Les systtmes experts en agriculture: les difftrentes approches actuelles. Perspect. Agric. 10654. 4 Conlin B. J., J. K. Reneau, R. D. Appleman, and R. J. Famsworth. 1989. Milking system XPERT for diagnosing milking system problems. J. Dairy Sci. 72(Suppl. 1):461.(Abs(r.) 5 Deer, L. A., D. D. Jones, and M. Okos. 1987. Expert system application for soybean oil extraction. Am. Soc. Agric. Eng. Paper 87-5540, Am. Soc. Agric. Eng.. St. Joseph, MI. 6 Doluschitz. R.. and W. E. Schnusseur. 1988. Expert systems: application to agriculture and farm management. Comput. Electron. Agric. 2(3):173. 7 Domecq, J. J., R. L. Nebel, and M. L. McGillard. 1989. An expert system to evaluate reproductive management and the need of an on-farm milk progesterone test. J. Dairy Sci. 72(Suppl. 1): 460.(Abstr.) 8 Fillatre, C. 1987. Interpretation automatique des rtsultats d’entreprises: les systtmes experts ouvrent-ils de nouvelles voies? Mtmoire de tin d’ttudes. Inst. Sup6rieur Agric., Lille, France. 9 Goodrich, P. R.. P. E. Robert, J. Gustafson. and S. N. Kalkar. 1987. The stray voltage adviser. Am. Soc. Agric. Eng. Paper 87-5539, Am. SOC.Agric. Eng., St. Joseph, MI. 10 Hogeveen. H., E. N. Noordhuizen-Stassen, J. F. Schreinemakers, and A. Brand. 1991. Development of an integrated knowledge-based system for management support on dairy farms. J. Dairy Sci. 74:4377. I1 Jones, D. D., and L. A. Hoelscher. 1987. Applications of expert systems in extension engineering. Am. Soc. Agric. Eng. Paper 87-5021, Am. Soc. Agric. Eng., St. Joseph, MI.

2317

12Le Corfec, F. 1987. E A . Bulletin technique d’information-intelligence artificielle et systtmes experts en agriculture. No. 424-425:485. Min. Agric., Paris, France. 13Lemmon. H. 1986. COMAX: an expert system for cotton crop management. Science (Washington, Dc) 233:29. 14 Levallois, R.. and D. Pellerin. 1989. Analyse des rtsultats technico-kconomiques d’une entreprise laititre par un systtme expert Rev. Can. Econ. Rur. 37:681. ISMichalski, R. S.. J. H. Davis, V. S. Bisht, and J. 8. Sinclair. 1983. A computer-based advisory system for diagnosing soybean diseases in Illinois. Plant Dis. 4: 549. 16Morrow. C. T.. R. E. Graves, and P. H. Heinemann. 1987. Expert systems implementation using “Rulemaster” at Penn State. Am. SOC. Agric. Eng. Paper 87-3532, Am. Soc. Agric. Eng., St. Joseph, MI. 17 Patacq, J. P.. 1987. Systtme expert “Aide B la dCcision pour l’ktablissement d’un planning fourrager.” Bull. Tech. Inform.-intelligence artificielle et systbmes experts en agriculture. No. 424425315, Min. Agric., Paris, France. 18 Peart, R. M.. Y. C. Li, R. E. LaTurner. and L. Li. 1987. Integration of expert systems with simulation for agricultural decision support systems. Proc. First Int. Conf. Agnc. Syst. Eng. August 11-14. Chinese SOC.Agric. Eng., Changchun, China. 19Schrmsseur, E., and M. J. Gamroth. 1993. DXMAS: Expert system program providing management advice to dairy operators. 1. Dairy Sci. 76:2039. 20 Schueller, J. K.. R. M. Slusher, and S. M. Morgan. 1985. A speech synthesis grain combine expert system. Am. SOC.Agric. Eng. Paper 85-1580, Am. SOC. Agric. Eng., St. Joseph, MI. 21 Toombs, M. 1988. The development of a silo selection expert system. Agric. Comput. Ext. 7(2):5. 22Tumer. L. W., T. C. Bridges, and D. G. Overhults. 1987. An expert system for troubleshooting livestock ventilation systems. Am. Soc. Agric. Eng. Paper 87-4505, Am. Soc. Agric. Eng., St. Joseph, MI. 23 Wahl,V. 1987. Modtlisation symbolique des conditions spatiales de la production agricole. MCmoire de Diplome d’ktude Apptofondie, Univ. Dijon, Dijon, France. 24 Wain, N., C.D.F. Miller, and R. H. Davis. 1988. A rule-based inference system for animal production management. Comput. Electron. Agric. 2:277.

Journal of Dairy Science Vol. 77, No. 8, 1994