JPH01312636A - Inference processor - Google Patents

Abstract

PURPOSE:To easily correct knowledge by providing a macro calling a procedure which exists in correspondence with respective elements of a rule in a knowledge base storage part. CONSTITUTION:The knowledge base storage part 17 has the macro 20 which calls the procedure of a procedure storage part 6 existing in correspondence with respective elements of the rule. For changing knowledge, a knowledge processing can easily by resumed without correcting the procedure on the side of a system only by correcting a file in which knowledge is mentioned. Thus, the trouble of re-compiling whenever knowledge is corrected is eliminated and knowledge can easily be corrected.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of Industrial Application The present invention relates to an inference processing device used for artificial intelligence.

BACKGROUND OF THE INVENTION A conventional inference device (general-purpose expert shell) is composed of an inference engine for performing inference, and rules and working memory that serve as a knowledge base. In particular, as a feature of conventional general-purpose expert shells, when invoking rules to perform inference, it is necessary to have the data necessary for inference in the working memory specified by that expert shell6 for inference to be performed. .

Recently, there have been many cases in which expert systems are constructed using general-purpose expert shells such as those described above in order to make existing systems perform knowledge processing.

In this case, one method is to create an expert system by newly writing all the data that already exists in the system into the working memory of a general-purpose expert shell in order to have the expert shell perform knowledge processing.

Another method is to reduce the working memory of a general-purpose expert shell to the minimum amount of data necessary for inference, and to create an expert system by retrieving data from an existing system from the expert shell side through procedures on the system side. There is. In this case, modifying the expert shell's knowledge means changing the system's procedures for manipulating existing data.

In each of the above cases, some form of interface is required between the existing system and the general-purpose expert shell.

Generally, general-purpose expert shells are rlispJ and rpr.
Because they are often written in a knowledge processing language such as ologJ, the languages used by existing systems and general-purpose expert shells are often different.

In such cases, a special means of communication is used between the two to activate the general-purpose expert shell from the system side, or to return from the general-purpose expert shell to the system side.

Such a conventional inference processing device will be explained using an example in which this inference processing device is incorporated into a CAD system.

First, FIG. 2 shows a configuration diagram of a CAD system without an inference device. 1 is the main body of the CAD system. Reference numeral 2 denotes a man-machine interface section, which controls data input from the keyboard of the input/output device 8 and outputs results to the display of the input/output device 8, etc. A control section 3 controls the CAD data storage section 4 and the procedure management section 5. Reference numeral 4 denotes a CAD data storage section where information on boards, parts, etc. is stored. A procedure management section 5 manages procedures in the procedure storage section 6. 6 is a procedure storage section;
Procedures (algorithms) for placing and routing components
This is where you store things like. Reference numeral 7 denotes a CAD data secondary storage section, which is a magnetic tape, magnetic diskette, or the like that stores CAD data for each board. 8 is an input/output device, such as a keyboard or a display.

The operation of arranging parts using this CAD system will be described below. A signal input from the input/output device 8 passes through the man-machine interface section 2 and then passes to the control section 3.

The signal is analyzed by the control unit 3, and the information on the board and parts is extracted in the CAD data storage unit 4, and the placement procedure is processed in the procedure storage unit 6 via the procedure management unit 5 to place the parts. . When the parts are placed, the placement results are output to the input/output device 8 via the man-machine interface unit 2.

Next, FIGS. 3 and 4 show configuration diagrams when a general-purpose expert shell is used in this existing CAD system.

1 to 8 in FIGS. 3 and 4 are the same as in FIG. 2. 22 is the main body of the general-purpose expert shell, r
LIsPJ and rPr.

It is an independent entity written in a knowledge processing language such as 1ogJ. Reference numeral 9 denotes an inference mechanism section, which controls the knowledge base storage section 10 and the working memory 11, and serves as an interface section 13.9 between the general-purpose expert shell 22 and the CAD main body 1.
This is where communication is performed with the control section of the CAD main body 1 via the CAD main body 1.

10 is a knowledge base storage unit, in which a knowledge source (hereinafter referred to as KS) 19 is stored. 19 is a KS in which rules are described. 11 is a working memory, which temporarily stores data necessary for inference.

Reference numeral 12 denotes a knowledge base secondary storage unit, which is a file for storing the knowledge base for each board.

The operation when automatically placing parts using the expert system shown in FIG. 3 will be explained. A signal input from the input/output device 8 passes through the man-machine interface section 2 and then passes to the control section 3. The signal is analyzed by the control unit 3, and the signal is sent to the general-purpose expert shell 22 for inference through the interface unit 13.14 between the CAD main body 1 and the general-purpose expert shell 22 in order to perform component placement by inference processing on the general-purpose expert shell 22 side. It is passed to the mechanism section 9. The signals from the inference mechanism unit 9 are used to analyze the rules in the KS 19 regarding the arrangement of the knowledge base storage unit 10, and the working memory 1 performs matching of CAD data necessary for inference. In this way, the placement of parts is performed only by the reasoning of this general-purpose expert shell 22. However, before control is transferred to the general-purpose expert shell 22, the data necessary for inference must be stored in the working memory 11 from the CAD data secondary storage section 7. The result of automatic component placement based on inference is passed back to the control section 3 as a signal through the interface sections 13 and 14 between the CAD main body 1 and the general-purpose expert shell 22. The signal from the control section 3 is outputted as a placement result to the input/output device 8 via the man-machine interface section 2.

Next, the operation when arranging parts using the expert system shown in FIG. 4 will be explained.

A signal input from the input/output device 8 passes through the man-machine interface section 2 and then passes to the control section 3. The signal is analyzed by the control unit 3, and the signal is sent to the general-purpose expert shell 22 for inference through the interface unit 13.14 between the CAD main body 1 and the general-purpose expert shell 22 in order to perform component placement by inference processing on the general-purpose expert shell 22 side. It is passed to the mechanism section 9. The signals from the inference mechanism unit 9 sequentially analyze the rules in the KS 19 regarding the arrangement of the knowledge base storage unit 10. First, when the elements of one rule are analyzed, the inference mechanism section 9 understands which procedure management is necessary for the arrangement of the CAD main body 1 described therein. Then, the signal passes through the interface section of 13 and 14, and the C
A signal is passed to the control unit 3 of the AD main body 1 and further to the procedure management unit, and each procedure is processed in the procedure storage unit 6, and arrangement is performed according to the procedure. Further, when referring to data in a procedure, the data in the CAD data storage section 4 is referred to via the control section 3. When the processing in the control section 3 is completed, the signal passes through the interface section 13 and 14 again and returns to the inference mechanism section 9, where the next element of the rule in the knowledge base storage section 10 is analyzed. Next, the process returns to the CAD main body 1 and processing is performed. When the above component placement processing is completed for all rules in the knowledge base storage section 10, the results of the component placement are passed to the control section 3 as a signal. The signal from the control section 3 is outputted as a placement result to the input/output device 8 via the man-machine interface section 2.

Problems to be Solved by the Invention However, when trying to use a general-purpose expert shell in an existing system as described above, in a large system with a lot of data, a huge amount of data must be stored as knowledge in the working memory of the general-purpose expert shell. For example, the expert system shown in FIG. 3 uses an independent general-purpose expert shell 22.
Therefore, all CAD data necessary for inference must be transferred from the CAD data secondary storage unit 7 to the working memory 11 of the general-purpose expert shell 22. Therefore, it requires a large amount of memory, making it difficult to implement this expert system.

Also, even if you use the database of an existing system instead of using all the working memory of a general-purpose expert shell, if you modify the knowledge base, you will also need to change the procedures on the system side, and the system will need to be recompiled. The problem is that it has to be done.

Furthermore, when performing complex inference processing, there is a problem in that the number of communications between the expert shell and the existing system increases, reducing processing efficiency. For example, although the expert system shown in FIG. 4 uses an independent general-purpose expert shell 22, all inferences are not performed in the working memory 11, but between the CAD main body 1 and the general-purpose expert shell 22 for each element of the knowledge base rule. It is an expert system that performs procedures for arranging the CAD main body 1 and refers to CAD data through the interfaces 13 and 14. However, in this case, in order to change the rules of the general-purpose expert shell, it is necessary to modify the placement procedure and procedure management program to refer to one CAD data for one rule element, and the system must be restarted. You will need to compile it. In this way, this expert system has the problem of having to recompile the system every time the knowledge is modified. Also, when performing complex inference processing, CA
The number of communications between the D main body 1 and the general-purpose expert shell 22 increases, reducing processing efficiency.

In view of these points, the present invention eliminates the need to rewrite the data of the existing system into the working memory, eliminates the need to re-pile the knowledge base when modifying it, and improves processing efficiency by reducing the number of communications between the expert shell and the existing system. The purpose is to provide an expert shell that can be embedded into existing systems without crashing.

Means for Solving the Problems The present invention includes an input/output unit that inputs commands and outputs processing results, a control unit that performs overall control, and interprets the commands input to the input/output unit and transmits the contents to the input/output unit. a man-machine interface section that passes the contents to the control section and interprets the contents from the control section and sends them to the input/output section; a data storage section having data to be processed; A procedure storage section that performs processing according to the procedure while referring to it through a control section, and a file in which rules are described and macros that call procedures in the procedure storage section that exist corresponding to each element of the rule are written. a knowledge base storage section having a file in which rules and macros are described in the knowledge base storage section, and an inference mechanism section that interprets the contents of the file in which rules and macros are described in the knowledge base storage section and activates processing according to the procedure in the procedure storage section via the control section. This is an inference processing device characterized by having the following.

According to the above-described configuration, the knowledge base storage unit has a macro that calls a procedure in the procedure storage unit that corresponds to each element of a rule, so that when an attempt is made to change the knowledge, the knowledge is changed. Knowledge processing can now be easily restarted by simply modifying the edges of the description without having to modify the procedures on the system side, eliminating the need to recompile every time knowledge is modified, making it possible to easily modify knowledge. It is.

Embodiment Hereinafter, as an embodiment of the inference processing device of the present invention, the content of CAD data will be calculated based on the content related to the layout described in the knowledge base in an existing printed circuit board placement and wiring CAD system (hereinafter referred to as CAD system). An expert system that performs inference processing will be explained with reference to.

FIG. 1 is a block diagram of a CAD expert system using the present invention. 2.3.4.6.7.8 are the same as FIGS. 2, 3, and 4.

16, 17, and 18 are an inference mechanism section, a knowledge base storage section, and a knowledge base secondary storage section based on the present invention.

15 is the main body of the CAD expert system.

As shown in FIG. 1, the knowledge base storage unit 17 stores a KS19 in which rules are written, a macro 20 in which macro definitions of each element of the KS rules are written, and a frame 21 in which knowledge necessary for inference is written. Divided. These knowledge! ,(
KSl macro, frame) will be referred to herein as a knowledge unit or simply a unit. FIG. 5, FIG. 6, and FIG. 7 show examples of the above-mentioned KSL macro and frame placement knowledge of the CAD system.

The operation when arranging parts according to the embodiment of the present invention will be explained below based on the flowchart shown in FIG.

A signal input from the input/output device 8 passes through the man-machine interface section 2 and then passes to the control section 3. The control section 3 analyzes the signal, and passes the signal to the inference mechanism section 16 in order to perform component placement through inference processing. Inference mechanism part 1
After analyzing the rule elements of the knowledge source (KS) 19 regarding the arrangement of the knowledge base storage unit 17, the signal No. 6 analyzes the macros 20 corresponding to the elements in order. This macro 20
If the inference mechanism unit 16 recognizes which of the procedures regarding placement should be performed during the analysis, the signal is transferred to the procedure in the procedure storage unit 6 via the control unit 3, and the placement process is performed. If CAD data is required for inference, the CAD data in the CAD data storage section 4 is referenced via the control section 3. Further, if the contents of the frame 2 are required, the contents of the frame 21 are referred to by the inference mechanism section 16 via the control section 3. When the processing in the procedure storage section 6 is completed, the signal is transferred to the inference mechanism section 6 via the control section 3 again, and returns to the macro analysis. If the inference mechanism unit 16 recognizes multiple procedures in one macro, the same process as described above will be repeated. Once the macro analysis is complete, the knowledge source (KS) 1
The elements of the 9 rules are analyzed in order, and the same process as above is performed until the end of the rule in the knowledge source (KS) 19.

The result of automatic component placement based on inference is passed to the control unit 3 as a signal. The signal from the control section 3 is outputted as a placement result to the input/output device 8 via the man-machine interface section 2.

As shown in Figure 1, this expert shell
CAD system data can be freely referenced by being embedded in the AD system.

Furthermore, in this CAD expert system, even if the knowledge is modified, if the rules of the knowledge base are to be changed, only the file in which the KS is written needs to be rewritten. Also,
If you want to change the procedures in the knowledge base, as will be explained next, you can write more detailed knowledge about each element of the rule in the macro definition. Bye. Afterwards, you can easily modify your knowledge by simply loading the modified file. Therefore, unlike the expert system shown in FIG. 4, there is no need to directly modify the procedure or modify the program that manages the procedure, so there is no need for recompilation. In the case of the expert system shown in Figure 4, it was necessary to prepare a program to manage procedures for each element of a rule that requires multiple procedures, so if a rule is modified, it is necessary to manage the corresponding procedures and procedures. I often modify programs that I use. However, this first
In the expert system shown in the figure, as mentioned above, one rule element can be further macro-expanded to select and combine more detailed procedures, so it is necessary to prepare in advance a wealth of procedure functions related to placement in the procedure storage unit 6. It is rare for procedural functions to be modified. This requires recompilation in the expert system shown in Figure 4.
This is why the expert system shown in the figure does not require recompilation.

FIG. 9 shows a conceptual diagram showing the relationship between the three knowledge units that are the knowledge base of this expert shell and the application functions, and the relationship between the knowledge units and the application functions will be explained in detail. Here, the procedure function related to placement, which is called by the inference mechanism unit 16 and exists in the procedure storage unit 6 of the CAD expert system 15, is referred to as an application function. Application
The Chiron function corresponds to a function defined in a macro, and is used to perform processing within the CAD system or manipulate data within the CAD system, and is prepared in advance.

As shown in Figure 9, this expert shell can have multiple KSs at once. i for each KS
A production rule is written in the format f-then.... When one KS is executed, the rules within that KS are evaluated in the order in which they are written. As this rule is evaluated, the knowledge necessary for inference can be freely extracted from the frame.

The condition element (if part) and execution element (th
en section) must both have macro definitions. This macro can describe detailed knowledge that would be complicated if expressed in the KS. Specifically, the cal1 statement of the application function is written in the macro.

The application function called in this macro is written in the same language as the existing CAD system and exists in the procedure storage section 6 within the CAD system. When the processing of this application function is finished, the process returns to the macro again and when the macro is finished, the process moves to the next element in the KS.

Next, we will discuss how to actually implement the knowledge processing flow described above. First, Figure 10 shows the data structure in which the knowledge units, KS, macro, and frame, are stored when a file that describes the knowledge base regarding component placement in a CAD system is loaded.

When knowledge processing is started, which KS is to be executed is recognized by the KS name, and control is transferred to the beginning of the data in which the rule element character string of that KS is described (a).

First, the conditional element (if part) character string of the rule is extracted, and a character string that matches the left-hand character string of the macro is searched from the macro left-hand side table (b). (Here, the character strings in the KS element table and the character strings in the macro left side table are matched at runtime, but when loading knowledge, the macro corresponding to the KS element is first searched and the macro left side is added to the KS element table.) A pointer to the table may also be indicated.

) If the left side of the macro is found, control is transferred to the macro right side table to see the contents of the corresponding right side of the macro (C
).

When the call of the applicator function on the CAD system side of the contents of the right side of the macro is recognized, control is transferred to the CAD system side and the applicator function is called (d).

In this process, the CAD database in the CAD system is referred to, and frame data as knowledge is extracted from the frame data structure and used for inference (e).

Since we are currently considering a conditional element, we will return the result of whether the condition is satisfied in this application function to the expert shell. Recognizing this result, the expert shell determines whether to transfer control to the execution element of this rule or to the next rule.

The process of processing the execution element (then part) of the rule is exactly the same as the processing of the condition element above, the only difference is that the application function is executed and no result is returned when returning to the expert shell. be.

Finally, FIG. 11 shows a simple flow when starting the expert shell described above. As shown in the figure, when you start the expert shell, the knowledge base is first loaded. Next, knowledge processing is performed based on this loaded knowledge base. After knowledge processing is performed, if there is a change in knowledge, the knowledge is loaded again and knowledge processing can be performed. Therefore, if knowledge changes, the same knowledge processing can be continued simply by reloading the knowledge.

As explained above, the inference processing device of the present invention can be applied to an existing CA.
By embedding this expert shell in the D system etc., the database for knowledge processing can be easily extracted from the existing CAD database, and high-speed knowledge processing can be realized.

Furthermore, by writing the knowledge base in a file, even if the knowledge is modified, the same knowledge processing can be continued by simply reloading only the modified knowledge unit. Furthermore, the knowledge base is divided into three knowledge units, making it easy for users to freely describe detailed knowledge.

As described in detail, according to the present invention, it is possible to construct an expert system that can easily process knowledge without changing the database of the existing system because of the expert shell embedded in the existing system. , high-speed knowledge processing can be achieved.

Furthermore, since knowledge can be easily changed simply by modifying the file that describes the knowledge, the effect of system knowledge can be easily verified.

Furthermore, since the knowledge base is divided into knowledge units, it is easy for users to freely describe detailed knowledge, which has great practical effects.

[Brief explanation of the drawing]

Fig. 1 is a block diagram of an embodiment in which the inference processing device of the present invention is incorporated into a CAD system, Fig. 2 is a block diagram of an existing CAD system, and Figs. 3 and 4 are a general-purpose expert shell installed in the CAD system. Figure 5 is a knowledge representation diagram showing an example of KS placement knowledge in the knowledge base, and Figure 6 is a knowledge representation diagram showing an example of macro placement knowledge in the knowledge base. , Fig. 7 is a knowledge diagram showing an example of frame placement knowledge in the knowledge base, Fig. 8 is a flowchart of the system when the above embodiment performs automatic placement, and Fig. 9 is a diagram showing the relationship between the three knowledge units of the expert shell. Conceptual diagram showing the relationship between and application functions, Part 1
FIG. 0 is a data storage diagram of the KSl macro and frame in the embodiment, and FIG. 11 is a flowchart showing the flow for starting the expert shell in the embodiment. 2...Man-machine interface engineering department, 3...Control unit, 4...CAD data storage unit, 6...
・Procedure storage unit, 7... CAD data secondary storage unit, 8... Input/output device, 15... CAD expert system main body, 16...
・Inference mechanism section, 17...Knowledge base storage section, 18...
-Knowledge base secondary storage, 19...KS. 20...Macro, 21...Frame. Name of agent Patent attorney Toshio Nakao 1 person Figure 2 1 Main body Figure 5 [KS CPU arrangement if (there is CPU) Q/1.1 * l arrangement] Figure 66 Tuna (Cpu or Roru ) -----------(call
f1 fart-Cpu ) (cpu 翫H iL )
----------(call pttt-cp
tz) ] Fig. 77 Frame CptL 昭版 ------- Riki Kibayashi Chuo 1; Place ] Fig. 8 Fig. 9 1-Fa, □ mark----"

Claims (1)

[Claims] an input/output section that inputs commands and outputs processing results; a control section that performs overall control; and an input/output section that interprets the commands input to the input/output section and passes the contents to the control section, and also receives the contents from the control section. a man-machine interface unit that interprets and sends the data to the input/output unit; a data storage unit that has data to be processed; and a procedure-based process while referring to the contents of the data storage unit via the control unit. a knowledge base storage section having a file in which a rule is written and a file in which a macro that calls a procedure in the procedure storage section that exists corresponding to each element of the rule is written; An inference processing device comprising an inference mechanism section that interprets the contents of a file in which rules and macros are described in a knowledge base storage section and activates processing according to procedures in the procedure storage section via the control section.