KR100241849B1 - Blackboard-based Mixed Reasoning Expert System - Google Patents

Abstract

Knowledge expression and knowledge base actively deal with simple and complex large-scale systems, and a blackboard-based expert system with independent inference engine. The expert system of the present invention includes a monitoring module for determining symptoms by monitoring data obtained from an object, an event handler for generating symptoms determined by the monitoring module as an event, a knowledge base and facts for storing knowledge in the form of facts and rules. The reasoning engine is operated only by the change of the blackboard without controlling each other between the knowledge source of the inference engine that estimates and extracts new facts from the rule and the rule, and the working memory that stores the information generated in the troubleshooting process. It is composed. Therefore, the present invention provides an independence between the blackboard and the inference engine, and thus can be easily added or deleted, other forms of knowledge integration, and multiple inferences are possible.

Description

Blackboard-based Mixed Reasoning Expert System

The present invention relates to an expert system, and more particularly, to a blackboard-based mixed reasoning expert system.

Knowledge systems, such as expert systems, are computer systems that emulate the inference tasks used by professionals. Such knowledge systems typically use an "inference engine" to translate the knowledge of an encoded expert stored in a "knowledge base." If the area of the knowledge base or the scope of the problem is sufficiently narrow and a sufficiently large amount of knowledge is properly coded in the knowledge base, the expert system can achieve performance comparable to or exceeding that of the expert. As such, the expert system applies the expert's knowledge to solve difficult problems, and uses symbols to represent the concept of thinking as a problem solving act by repetitive methods. The expert system has a knowledge base that stores expertise that is specific to a particular area of the problem area, where knowledge is stored in the form of facts or rules. A key part of the expert system is the reasoning agency, which is responsible for identifying and implementing the appropriate rules for the current situation. However, from the user's point of view, he wants to handle these expert systems more conveniently. To this end, a supporting environment is obtained from the development system of the expert system. Among these support areas, it is important that users input and output (input / output interface or user interface) to handle data more easily and talk to expert system, and to build and change knowledge base from experts in problem areas. The knowledge acquisition part is important. As such, the four parts of the knowledge base, user interface, reasoning agency, and knowledge acquisition are called the basic structure of the expert system. Here, the construction of the knowledge base is shown as rules and facts.

The expert system is not limited to the scope of use such as medical equipment, mechanical or electronic equipment. In the expert system for diagnosing failures in structural equipment such as cranes, methods such as RBR, MBR, and CBR were used independently. Here, RBR (Rule Based Reasoning) is basically a method of deriving a result using a forward reasoning or backward reasoning method. Model Based Reasoning (MBR) is a method of systematically organizing the structure and function of the equipment itself as a model in the knowledge base to make inferences of large or complex contents more accurate, detailed and easier. CBR (Case Based Reasoning) is a method of storing past diagnosis cases in the knowledge base and inferring similar past situations when a new situation is entered and comparing them or performing similar analysis.

As such, the expert system is largely composed of knowledge base, working memory, control part, and inference engine. In detail, it is classified into knowledge base module, inference engine module, knowledge acquisition module, explanation module, and user interface module. It is largely classified into inference engine and knowledge base. Inference is the process of estimating and extracting new facts from known facts and rules. A knowledge base is a database of knowledge on a particular subject used in expert systems, consisting of the facts representing data and the rules that use those facts as the data for decision making. Reasoning engines include forward reasoning, backward reasoning, and hybrid reasoning. The knowledge acquisition module is a system that manages the acquisition of knowledge, and the explanation module infers the process of inferring the user to the conclusions or intermediate results of "how did the system conclude?" And "why special data was needed?" It has a function to explain to the user. The expert system is a knowledge base that expresses the expert's knowledge based on rules, frames, or models, and the inference engine performs diagnosis based on the knowledge base. In order to speed up the execution of this diagnosis, the board structure, which is a working memory, is used. The Blackboard structure is useful for integrated management of various knowledge expressions, reasoning methods, and controls, and is an independent "Loosely Coupled Type" without the interconnection of thematic knowledge groups that form the knowledge base. (Oppertunistic) The reasoning method is easy to apply to relatively complex problem solving domains.

Many systems use a blackboard as a reasoning structure for real-time problem solving. The blackboard is a dynamic system that generates all the dynamically generated diagnostic information such as message names, other symptoms, and possible causes of failures. This is where the kind of information exists. This information on the blackboard dynamically triggers knowledge sources. Trouble shooting knowledge source refers to various rules that can be used for troubleshooting. These rules are composed of various forms, and the knowledge sources are operated only by the change of the blackboard without controlling each other. That is, only a new fault diagnosis state, which is a change of the blackboard, triggers a knowledge source and the knowledge source is not driven by another knowledge source. This independence facilitates the free addition or deletion of knowledge sources. Blackboards, which consist of working memory, knowledge sources, and controls as basic components, have the advantage of integrating knowledge of different forms in one structure and structural features that can be inferred. Complex.

However, such a conventional expert system uses a black board as an inference structure for real-time problem solving. This has the disadvantage that the knowledge base is only applied to complex systems.

Accordingly, an object of the present invention is to provide a blackboard-based mixed reasoning expert system having an independent reasoning engine that can actively use a blackboard for a small system with a simple knowledge expression and a knowledge base and a complex large system.

1 is a block diagram showing a blackboard-based mixed reasoning expert system according to the present invention,

Figure 2 is a detailed view showing the inference engine of the system of Figure 1,

3 is a detailed view illustrating an interpreter of the inference engine of FIG. 2;

4 is a detailed view illustrating an inference controller of the inference engine of FIG. 2;

5 is a conceptual diagram illustrating the operation of the pattern matcher of FIG. 2;

6 is a detailed view showing a blackboard of the system of FIG. 1;

FIG. 7 is a detailed view of the knowledge base of the system of FIG. 1; FIG.

<Description of the symbols for the main parts of the drawings>

10: monitoring module 20: user interface

30: event handler 40: inference engine

41: interpreter 42: pattern matcher

43: scheduler 44: inference controller

50: learning module 60: developer interface

70: knowledge acquisition module 80: knowledge base

81: domain knowledge 82: meta knowledge

90: Blackboard 91: Blackboard

92: Agenda

Blackboard-based mixed inference expert system according to the present invention for achieving the above object, in the expert system for diagnosing the failure of the object, performing matching between the rules obtained for the data and behavior knowledge obtained from the object and A monitoring module for determining a symptom by performing a rule selected according to a ranking among the completed matching rules, an event handler for generating and outputting a symptom determined by the monitoring module as an event, an event and a blackboard generated by the event handler. Pattern matching is performed on domain rules loaded by the department, and the inference engine unit estimates and extracts new facts by performing a rule selected according to the rank among the matching rules, and stores the knowledge in the form of rules and facts. Knowledge of multiple representations of all decisions made by the inference engine. Provide a knowledge base having a domain knowledge and a meta knowledge, and a rule corresponding to an event generated by the event handler to the inference engine unit, and output the inference result corresponding thereto. It includes a blackboard, which is a working memory that manages rules according to symptoms.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Figure 1 shows the configuration of the blackboard-based mixed reasoning expert system according to the present invention. Referring to FIG. 1, the expert system of the present invention is composed of a knowledge base consisting of a knowledge base 80 and an inference engine unit 40 and a blackboard unit 90 as a work memory, and a learning module 50 and monitoring. It consists of a monitoring module 10, an event handler 30, the user interface 20 for. And, it is composed of a knowledge acquisition module 70 for learning new cases obtained by the developer and the learning module 50 with new knowledge through the developer interface 60. The inference engine unit 40 is an interpreter 41 for determining how to apply rules to infer new knowledge, a scheduler 43 for determining in which order to apply the selected rule, a pattern matcher 42, Inference controller 44 is configured. The knowledge base 80 includes domain knowledge 81 and meta knowledge 82 that store facts and rules. The blackboard unit 90 includes a blackboard 91 and an agency 92.

In FIG. 1, the expert system performs a procedure for diagnosing a failure of a target structure between a target structure to be diagnosed and a user who is familiar with it. The expert system largely performs monitoring and diagnostic operations. The monitoring operation determines whether the symptom due to a malfunction or the fact is normal due to the data acquired through the sensor. The diagnosis operation diagnoses the malfunction when it is determined to be a malfunction by the monitoring operation.

First, the monitoring operation is performed in the monitoring module 10 of the expert system. There is a knowledge base in the monitoring module 10 can monitor the data obtained from each component constituting the object through a sensor not shown. The monitoring flows for monitoring and diagnostics are the same, so monitoring is outlined. The obtained data triggers the blackboard 91 of the blackboard unit 90. The blackboard 91 loads behavioral knowledge from the knowledge base present in the monitoring module 10. The monitoring module 10 performs a matching step between the rules for the loaded behavior knowledge and the obtained data. The monitoring module 10 outputs the matching rules to the agency 92 of the blackboard unit 90. Agenda 92 places the matched rules in the triggered agency and then moves them to a viable agent according to the strategy. The scheduler 43 selects a rule according to priority among rules KSRs located in an executable agency, and performs a rule fire on the selected KSAR by the interpreter 41 in order. do. Here, the fire means that the rule that the conditional is satisfied while the generation system is operated, and means the execution of the action unit by the interpreter 41.

On the other hand, if the diagnosis is performed on the symptoms determined through the monitoring procedure as described above or if the user detects a failure by guessing, the diagnosis is performed by inputting a failure diagnosis request through the user interface 20. In this case, the blackboard unit 90 loads the meta knowledge and the corresponding meta knowledge network into the control blackboard in the blackboard 91 instead of the behavior knowledge for monitoring. Here, metaknowledge is referred to as knowledge about one's own thinking of the system, which means knowledge about knowledge. The meta node is searched using the loaded meta knowledge. In order to search the metanode, the pattern matcher 42, the scheduler 43, and the interpreter 41 should be called in this order. Using the discovered metanode, the minimum domain rule set is loaded into the domain blackboard in the blackboard 91 and then the domain network is loaded. The pattern matcher 42 performs pattern matching on the symptoms of the blackboard 91 and the domain rules to locate the matched rule as the triggered agent, and then moves the state to the execution agent according to the degree of matching. The scheduler 43 performs scheduling on the rules loaded in the execution agent to execute the selected rule in the interpreter 41.

2 shows a detailed view of the inference engine unit 40. Referring to FIG. 2, the inference engine unit 40 includes an interpreter 41, a pattern matcher 42, a scheduler 43, and an inference controller 44. Features for these configurations are defined as shown in Table 1 below.

TABLE 1

Component Characteristic Scheduler Determine when and when to apply domain knowledge Interpreter How domain knowledge is determined Matcher Pattern Matching Inference mechanism Decision module

The inference engine unit 40 updates the domain blackboard in the blackboard 91 with the event delivered from the event handler 30, and updates the symptoms included in the event and the acquired symptoms stored in the domain blackboard. Call the pattern matcher 42 to find the matching rule and store it in the trigger-agent in agent 92. In the pattern matching process, matching is performed by dividing the key-symptom considering the priority and the related-symptom related to all symptoms. After the state transition is performed to the performance-agent based on the matching rate for the rules stored in the trigger-agent, the scheduler 43 selects one of them and the interpreter 41 performs the conclusion of the selected rule. If the final result is the final action, after verifying the user through the user interface 20, the decision is made, and in the case of the intermediate action, it is regarded as a new fact and patterned again with the acquired symptom (42). Iterates and repeats the process of selecting matching rules again. If the final conclusion is reached, a message is output to the user, and if the process is new knowledge, it is added to the knowledge base 80.

3 shows a detailed view of the interpreter 41. 3, the interpreter 41 includes an event classifier 411, an execution queue 412, and an action executor 413. The event classifier 411 classifies the event delivered from the event handler 30 to determine execution or waiting for execution. Here, the event includes not only data transmitted from the event handler 30 but also data transmitted from the pattern matcher 42. Thus, the event classifier 411 classifies the event corresponding to the data transmitted from the pattern matcher 42 to determine execution or execution wait. The execution queue 412 stores the events classified through the event classifier 411 and outputs the stored events to the action executor 413. The action executor 413 outputs a message to the user interface 20 so that the user can know the result by executing the event read from the conclusion part of the rule where the pattern matching is completed and the execution queue 412 through the event classifier 411. do. The action executor 413 updates the contents of the domain blackboard through the execution result.

4 shows a detailed view of the inference controller 44. Referring to FIG. 4, the inference controller 44 is composed of an inference strategy determination module 441, an inference strategy progress module 442, and a KSAR compiler 443.

There are three inference strategies, forward chaining, backward chaining, and direct chaining, depending on how the rule is used during inference. The reasoning strategy decision module 441 is selected by the developer during initial system development, and one of the three reasoning strategies is selected according to the purpose of reasoning and the nature of the problem field. The reasoning strategy progress module 442 controls the reasoning to proceed according to the selected reasoning strategy. The KSAR compiler 443 compiles the selected reasoning strategy and delivers it to the scheduler 43.

5 shows a conceptual diagram for explaining the operation of the pattern matcher 42. Referring to FIG. 5, the pattern matcher 42 is largely divided into six stages by using the rule to be a target and the acquired symptoms (facts). It is assumed that the rule to be matched is a rule to be applied to diagnosis, and then it is set as an initial hypothesis. The initial hypothesis is a semi-activate or suspend stage to obtain the acquired symptom based on the degree of matching with the acquired symptom, and an activation to trigger the blackboard. Matching is performed by moving state to a stage, a concluding conclusion, or a reject, which indicates that the hypothesis is useless. That is, if the hypothesis is initially set to suspend (step 501) when performing the set hypothesis, the process proceeds to a semi-activation to bring about the acquired symptoms (step-to-semi-active hypothesis) (step 502). If the state is transferred to the semi-activated hypothesis step through step 502, the acquired symptom is obtained (step 503). In the semi-activate hypothesis stage, go back to suspend (semi-active-to-suspend hypothesis) (step 504) or as an activator to verify that the symptoms are true. -symptoms-expected-on hypothesis (step 505). If the state is moved to the activate hypothesis step 505, it is verified whether the symptom is true (step 506). Depending on the verification result, it can proceed in various states. As a result of the verification, the conclusion is moved to the include hypothesis stage (step 507), and the conclusion obtained is the hypothesis that the acquired symptom is set, that is, how related symptoms are to the applied rule (confirm-fault hypothesis). (Step 508). As a result of the verification, if the symptom is not necessary, it moves to the reject hypothesis stage (step 509) and removes the symptom (remove-symptom-assumed-on hypothesis) (step 510). As a result of the verification, if the acquired symptom satisfies the set hypothesis, the process moves to an in-activate hypothesis stage (step 511). In the in-activate hypothesis state, if it needs to be verified again, it moves to the activate hypothesis state of step 506, and to the reject hypothesis state of step 509 if it does not work. If a new hypothesis is needed, move to suspend hypothesis in step 501; otherwise, go to refinement manager (step 512) to make an improved proposal (suggest refinement). hypothesis) (step 513). In this matching process, the matching degree is formulated and then used for the agent multi-state transformation using this value.

6 shows a detailed view of the blackboard system 90. Referring to FIG. 6, the blackboard system 90 is largely composed of a blackboard 91 and an agency 92. The blackboard 91 is composed of a domain blackboard 912, a control blackboard 913, and a memory 914 for storing facts obtained from the inference engine 40. The input / output is a loader (blackboard manager) 911 and dispatcher 915. The domain blackboard 912 is divided into a pattern network area and a domain rule area, and the pattern network area stores a pattern network for rules loaded in the domain rule area. The control blackboard 913 stores general information about meta knowledge and uses it during the inference process. This information also includes a network about meta knowledge. The agent 92 manages rules according to symptoms, and is composed of a trigger-agent 921 and an execution-agent 922. The trigger-agent 921 stores rules for which pattern matching is equal to or greater than a reference value, selects executable ones among them, and transfers the state to the execution-agent 922. State transitions are performed by Agenda Managers.

7 shows a detailed view of the knowledge base 80. Referring to FIG. 7, the knowledge base 80 uses permanent static memory. This is the most important part of the expert system and provides knowledge of different representations of all the decisions made in the inference engine. The knowledge base 80 manages input and output of knowledge by a loader and dispatcher, which is a knowledge base manager, and includes domain knowledge 81 to be applied to domains and control knowledge that is a higher level of these domain knowledges, that is, meta knowledge 82. do. Domain knowledge is divided into rules and facts, and the knowledge called by the inference engine is created and stored with the pattern network.

As described above, the expert system of the present invention has a structure that performs the advantages of the blackboard and the inference engine, has flexibility in inference control, and enables the use of multiple knowledge base languages. In addition, it is possible to combine each reasoning engine by dividing the knowledge into multiple, separate knowledge expression and application, and various strategies can be applied during inference, so it is easy to implement multiple reasoning structure for multiple knowledge expression.

Claims (6)

In the expert system for diagnosing a failure of an object, a monitoring module that performs a matching between the data acquired from the object and a rule for behavioral knowledge and determines a symptom by performing a rule selected according to a ranking among the completed rules. An event handler which generates and outputs a symptom determined by the monitoring module as an event; An inference engine that performs pattern matching on the events generated by the event handler and the domain rules loaded by the blackboard unit, and estimates and extracts new facts by performing a rule selected according to a rank among the matched rules; Domain knowledge that stores knowledge in the form of rules and facts, provides knowledge of various representations for all decisions made by the inference engine, and stores pattern networks generated for it, and meta-control that controls domain knowledge. Knowledge base with knowledge; And a blackboard unit which is a working memory that outputs a rule corresponding to an event generated by the event handler to the inference engine unit and manages rules corresponding to symptoms through an inference result corresponding thereto. The apparatus of claim 1, wherein the inference engine unit comprises: an interpreter configured to classify and execute an input event; A pattern matcher which initially sets a rule to be matched as a rule to be applied to the diagnosis and obtains a rule to be applied to the diagnosis by matching the acquired symptom with a predetermined rule; A scheduler for determining the order and timing of domain knowledge application; And an inference controller that determines one of three inference strategies: forward connection, backward connection, and direct connection according to the purpose of reasoning and the nature of the problem field, and controls the inference to proceed according to the determined inference strategy. system. The apparatus of claim 2, wherein the interpreter comprises: an event classifier configured to receive an event generated by the event handler and a result of the pattern matcher as an event, classify the event, and determine execution and execution wait; An execution queue configured to store an event determined to be executed by the event classifier and an incomplete action, and output an event to be executed; And an action executor for executing an event determined to be executed in the event classifier and an event output from the execution queue, and outputting an incomplete action to the execution queue. The method of claim 2, wherein the pattern matcher assumes a rule to be matched as a rule to be applied to diagnosis, and sets it as an initial hypothesis; Bringing the acquired symptom based on the degree of matching with the obtained symptom; Verifying that the symptom is true; And if the symptom obtained is determined based on the verification result, the degree of related symptoms to the hypothesis set or if the symptom is not necessary, the symptom is eliminated or if the verification is necessary while satisfying the hypothesis set by the symptom. Performing a step, performing a hypothesis setting step if a new hypothesis is needed, or performing a matching process of suggesting a rule set as the hypothesis as a rule to be applied to diagnosis. The apparatus of claim 1, wherein the blackboard unit receives an event generated by the event handler, outputs corresponding strategy information to the inference engine unit, and updates the information by receiving the inference result of the inference engine unit; And a rule selected from a trigger-agent, a trigger-agent that stores rules whose pattern matching is greater than or equal to a reference value, and selects dual executable rules, an execution-agent that performs rules selected from a trigger-agent, and a trigger-agent. An expert system comprising an agent that manages the rules according to symptom. 6. The apparatus of claim 5, wherein the blackboard comprises: a domain blackboard comprising a domain rule region storing rules for symptoms and a pattern network region storing pattern networks for rules loaded into the domain rule regions; A control blackboard for storing general information about meta knowledge; Memory for storing facts; And a black board manager comprising a loader and a dispatcher for managing the input and output of the components.

Images