Expert Systems with Applications PERGAMON
Expert Systems with Applications 16 (1999) 73–77
CMEOC—An expert system in the coal mining industry Hong Zhang*, Guanghui Zhao Department of Computer Science and Technology, China University of Mining and Technology, Xuzhou 221008, People’s Republic of China
Abstract This paper introduces an expert system in the mining industry, called the Coal Mining Expert and Optimization Consultation System (CMEOC). The paper mainly describes the architecture of this system, including the knowledge base, inference engine, and relevant optimization techniques. The techniques include multi-objective programming, fuzzy sets, and integer programming. The application results present evidence of the usefulness of the system. 䉷 1999 Elsevier Science Ltd. All rights reserved Keywords: Expert system; Optimization techniques; Inference network
1. Introduction Economic development has greatly impacted industrial processes throughout the world, and the industry of coal mining is no exception. Much effort has been made to promote productivity and quality of products to meet the demands of economic development. The new production process not only requires automation, but also the application of advanced artificial intelligence (AI) technologies and optimization techniques. Optimization decision making techniques have been used to support engineering design and decision making in the coal mining industry, for example the selection of alternative open-pit transportation systems or mining machineries. In recent years, with the development of artificial intelligence, expert systems have become an active application area. Expert systems have been applied in many areas such as medicine, geology, engineering design, education and military (Liebowitz, 1990; Xu and Li, 1993). In the field of mining engineering, expert systems also have important applications. Expert systems provide conclusions based on experts’ experience and knowledge, such conclusions may be qualitative, fuzzy, and imprecise. If expert systems’ outputs can be integrated with relevant optimization techniques, such as multi-objective programming, integer programming, and fuzzy sets, then the conclusions from expert systems can be more quantitative and accurate. As a result, such systems will be more practical in applications. Based on the above ideas, we have developed a system called Coal Mining Expert and Optimization Consultation * Corresponding author.
System (CMEOC). The system aims at integrating a coal mining engineering expert system with optimization techniques to reach the goal of optimal decision making. This paper introduces the CMEOC system and its structure, implementation, as well as optimization techniques. It also provides an application example in coal mining to illustrate its usefulness.
2. CMEOC system architecture The CMEOC system is specially designed for coal mining engineering. It is used to determine the underground mining methods, open-pit mining and transportation systems, etc. In using the CMEOC system, the expert system is used first, then optimization techniques such as multiobjective (MO) programming, fuzzy sets (FS), integer programming (IP), etc. are used to generate the final recommendation, as shown in Fig. 1. From Fig. 1 we can see that the system consists of three components: an expert system, optimization techniques, and a design and drawing (DD) system. 2.1. Expert system The expert system consists of a knowledge base (KB), an inference system (IS), and a control system (CS). 2.1.1. Knowledge base The knowledge base consists of three components. The static database is used mainly to store the data of the current problems. The dynamic database is used to store all kinds of
0957–4174/99/$ - see front matter 䉷 1999 Elsevier Science Ltd. All rights reserved PII: S0 95 7 -4 1 74 ( 98 ) 00 0 32 - 3
74
H. Zhang, G. Zhao/Expert Systems with Applications 16 (1999) 73–77
(geological structure of sample is given) (technical capacity is competitive) THEN (the general composite mining method can be used) Representation of fuzzy knowledge—Experts’ knowledge can be ambiguous and imprecise. Expert knowledge can be assigned a reliability factor b (0 ⬍ b ⬍ 1). There are three ways to determine the value of b. They are: mathematical method, expert judgment method, and similarity comparison method.
Fig. 1. CMEOC system structure.
intermediate information acquired in the inferencing process. The rule base represents the expert knowledge and theoretical knowledge of special fields such as mining engineering. There are three kinds of representation of knowledge in the system. Representation of facts—Predicate calculus is one of the methods used to represent knowledge. It is employed here to represent knowledge. For example, the determination of a mining process is a combination of a set of facts. It can be represented by the conjunction of the predicates as follows: Mining process (method) ∧ Mining process (given conditions) ∧ Mining process (geological structure) ∧ Mining process (technical capacity)
Representation of reasoning knowledge—In the system production rules are used to represent reasoning knowledge (Zhang, 1988). The general format is as follows: IF (fact 1 exists) (fact 2 exists) … … (fact n exists) THEN (conclusion P exists with reliability b) (0 ⱕ b ⱕ 1)As an example: IF (given condition is good)
2.1.2. Inference system and control system The IS and CS consist of a set of programs which controls and coordinates the whole system (Wu, 1991). IS is the key of the expert system. It solves the problem according to certain inference and control strategies using the knowledge base. They work like this: the user provides inputs and the related facts are stored in the knowledge base through CS and IS. Based on the initial inference results, the relevant optimization technique module is invoked through IS, and then the final optimal recommendation is made. Reasoning mode—In developing the inference mechanism for expert systems, one of the inference strategies is to use both forward and backward chaining (Liebowitz, 1990). The system mainly uses forward chaining. In some circumstances we use a double forward reasoning method. The process of matching can form a contextual inference network dynamically. An example of the inference network of the recovery process module is shown in Fig. 2. In the system we also use backward chaining in which a conclusion is proposed first and then proved by the known facts with the help of rules. Reasoning control—The system uses three strategies to control the reasoning. They are depth-first strategy, breadthfirst strategy, and best-first strategy. The adoption of which one depends on the structure of the inference network. In depth-first strategy, when a state is examined, all of its children and their descendants are examined before any of its sibling. Depth-first strategy goes deeper into the search space whenever it is possible. Breadth-first strategy, in contrast, explores the space in a level-by-level fashion. Only when there are no more states to be explored at a given level does the algorithm move on to the next level. Best-first strategy is a search algorithm that is similar to the algorithms for depth- and breadth-first search. It is known that the simplest way to implement a heuristic search is through a procedure called hill climbing. In general, a heuristic search requires a more informed algorithm, this is provided by the best-first strategy. The depth-first strategy is used when the reasoning network is relatively wide. The breadth-first strategy is used when the reasoning network is relatively narrow, but deep. The best-first strategy can be combined with the depth-first and breadth-first methods.
H. Zhang, G. Zhao/Expert Systems with Applications 16 (1999) 73–77
75
Fig. 2. Mining process model inference network.
Inference rule—Currently the system has more than 100 rules. The following are some examples. Rule 1 IF coal sloping angle a ⬍ 15⬚, coal thickness h ¼ 1.3– 3.5 m, geological structure simple, direct ceiling category 1, 2, 3, aging ceiling degree I, II, III THEN composite mining Rule 2 IF coal seam with large sloping angle, h ¼ 1.8–2.5 m, ceiling and floor conditions good, coal quality hard, no special geological structure THEN general mining with high quality equipments Rule 7 IF ceilings without regular pressure, floors stable, very thick coal seams with slow inclination THEN hydraulic frame draw general mining 2.2. Optimization techniques Using the expert system, we can get a set of initial solutions such as A i (i ¼ 1,2,3,…,n). But the user usually wants an optimum solution, so it is necessary to find an optimal solution on the basis of initial solutions. As shown in Fig. 1, IS uses the solutions obtained from the inference process as the intermediate solution and passes that to the intermediate scheme database M 1,M 2,…,M n. Then IS invokes the relevant optimization techniques to make the final
recommendation. The selected optimization techniques adopted in our system are introduced as follows. 2.2.1. Multi-objective programming Using the solution parameters obtained from the expert system, we can get a matrix F as F ¼ (fij )n ⫻ m, where fij is the ith parameter of the jth alternative, n ¼ number of alternatives to be considered, and m ¼ number of parameters to be evaluated. As f ij uses different measurements, a relative deviation method is used for regularization. The composite index k i is calculated as follows: s m X (wj ⫻ bij ) ki ¼ i¼1
where w j is the coefficient of the jth element. The smaller the value of k i, the more preferable the alternative will be. Therefore, the optimal alternative is the one with the smallest k i. This method is usually used to select the optimal coal mining method or coal mining process. 2.2.2. Fuzzy sets This method is used to select the optimal transportation system in open-pit mining or used in determining the optimal underground conveyor signal control system. A simple fuzzy evaluation process is as follows: Initial judgment—Let n be the total number of feasible solutions x i (i ¼ 1,2,…,n), and m the total number of evaluation items of x i. Then the objective set is: x ¼ {x1 , x2 , …, xn }
76
H. Zhang, G. Zhao/Expert Systems with Applications 16 (1999) 73–77
and the evaluation set is: U ¼ {U1 , U2 , …, Un } The evaluation matrix is: R : x·U → [0, 1] rij ¼ R(xi , Uj ) where r ij is the value of the jth evaluation item U j in the ith solution x i. With the variation of u j: 1X r (i ¼ 1, 2, …, m) m i¼1 m
di1 ¼
design, mining machinery design, etc. The programs in the design and mining module together form a toolbox, from which tools can be chosen according to different uses. The design based on the optimal recommendation will be drawn using AutoCAD software, and printed. Since AutoCAD software is not capable of design calculations, it is necessary to set up an interface program to connect AutoCAD to a high level language program. Using such interfaces, the parameters calculated from the high level language programs can be transferred to AutoCAD to invoke the drawing function.
di2 ¼ Max{ri1 , ri2 , …, rin } (i ¼ 1, 2, …, n)
3. CMEOC system in use
di3 ¼ Max{ri1 , ri2 , …, rim } (i ¼ 1, 2, …, m)
At this time the CEMOC system has been used preliminarily in selecting mining methods, open-pit transportation systems, and mining processes. The following is an application example for a working coal face. In this example, the working coal face situations are as follows (Zhang, 1996):
The final evaluation index is: 1 di ¼ (di1 þ di2 þ di3 ) (i ¼ 1, 2, …, n) 3 Finally, the optimum solution is the one with the maximal d j. 2.2.3. Mathematical programming This method is used to determine the types of machinery to be used. For example, 0–1 integer programming is used to determine the types of air-compressor to be used: ! m X X Fk þ Rk ⫻ r1 ⫻ xk (1) Min k僆N
S:T:
X
k¼1
k ⫻ Qk ⫻ xk ⱕ Q
(2)
k僆N
X
xk ¼ 1
(3)
k ⫻ xk ⱕ 6
(4)
k僆N
X k僆N
xk ¼ 0 or 1 The objective function (1) is to minimize the total cost. Constraint (2) meets the requirement on the flow of compressed air. Eq. (3) is used to restrict the selection only for one type of compressor. Eq. (4) meets the requirement on total number of compressors. Since the types of aircompressors are limited, 0–1 integer programming can easily be solved by using an exhaustive enumeration method. 2.3. Design and drawing Usually, the final optimal solution should not only be a qualitative or quantitative one, but also an applicable one. Therefore, a design and drawing (DD) module is built in the system for mining design, open-pit transportation system
Coal average seam thickness 6.61 m Slope angle 2–7⬚ Hiding depth 140–240 m Coal seam containing three layers of waste rock Hardness 1.5–2 Coal quality poor False roof, coal mud, thickness 0–0.3 m Direct roof, mud, thickness 4.53 m Primary span measure 22.4 m, belonging to second degree roof Aging roof, medium arenaceous to second degree roof; aging roof, medium arenaceous rock, thickness 2.09 m Direct floor, arenaceous shale rock, thickness 2.09 m Working coal face length 176 m Mining slope length 1563 m Working coal face length 176 m Mining slope length 1563 m Alignment 1340 m Windgate net section 10 m 2 Conveying net section 13.2 m 2 All in No. 11 ‘I’-like steel trapezoid roof plus metal net support The whole mine is in second degree management. 70% of the team leaders have college degrees. Technical personnel all have a graduate degree. More then 93.7% of the workers are high school graduates. Based on the above data, by means of man–machine interaction, the inference system obtains field data, expert experience and rules from the knowledge base. Then, according to the inference network shown in Fig. 2, using a double inference method, the user can get a number of mining process models. They are: (1) once-through full-lift composite mining in thick seams; (2) hydraulic frame draw coal high-standard composite mining; and (3) hydraulic frame draw coal general composite mining.
H. Zhang, G. Zhao/Expert Systems with Applications 16 (1999) 73–77
Based on the set of alternatives recommended by the expert system, IS invokes the multi-objective optimization technique module (Wang et al., 1994), then finalizes the mining process model as hydraulic frame draw coal highstandard composite mining. From the above, the recommendation made by the CMEOC system is very similar to the mining experts’ recommendation. In other words, the application result has been compared with mining experts’ judgement, presenting evidence of the usefulness of the system.
4. Conclusion The development and application of the CMEOC system in the Chinese coal mining industry is a preliminary attempt to combine expert systems and optimization techniques. The system was developed using AI language Prolog and software engineering principles and methods. The system has integrated AI technology with the optimization techniques. The system generates a set of solutions using the expert system component. Then, such solutions are further refined by using optimization techniques, such as multi-objective programming and integer programming. Based on the system’s final optimal recommendation, the AutoCAD toolbox is used to generate engineering drawing for a variety of design applications.
Appendix A Notation used in Fig. 2
HFDM DEFDM IEFDM OFMTS SMDE SMIE MTDE MTIE TSCCM TSCP HG GM GC
Hydraulic Frame Draw Mining Domestic Equipment for Frame Draw Mining Imported Equipment for Frame Draw Mining Once-Through Full-lift Mining in Thick Seams Stratifying Mining by Domestic Equipment Stratifying Mining by Imported Equipment Medium Thickness Mining by Domestic Equipment Medium Thickness Mining by Imported Equipment Thin Seam Mining by Cylinder-type Coal Machinery Thin Seam by Coal Planer High General General Mining General Composite
HC RPM EC GS PA ESA EX G FG B S FS C H M L E NE CST CSSA CSS CSH S G RV CC FC PD TD WS MC PA MTC GL IA
77
High Composite Recovery Process Model Existing Conditions Geological Structure Personal Ability External Structure Ability Excellent Good Fairly Good Bad Simple Fairly Simple Complicated High Middle Low Enough Not Enough Coal Seam Thickness Coal Seam Slope Angle Coal Seam Structure Coal Seam Hardness Structure Gas Rock Variation Ceiling Condition Floor Condition Psychological Disposition Technical Disposition Working Style Mining Capacity Production Ability Mining Transportation Capability Gateway Layout Investment Ability
References Liebowitz, J. (1990). The dynamics of decision support systems and expert systems. Orlando, FL: The Dryden Press. Wang X. D. et al. (1994). The selection of thick coal seam recovery process by multi-objective decision-making method. Coal Science and Technology, 22 (10), 6–8. Wu K. (1991). A preliminary study on expert systems for mining under structures. Journal of China University of Mining and Technology, 1, 98–101. Xu L., & Li L. (1993). An expert system approach to AIDS intervention and prevention. Expert Systems With Applications, 6, 119–127. Zhang Y. (1988). An expert system for strip mining under the buildings. Journal of China University of Mining and Technology, 4, 44–50. Zhang D. S. (1996). The design and realization of recovery process model selection of expert system. Journal of China University of Mining and Technology, 2, 37–39.