JPH11120000A - Design system using case-based reasoning and learning - Google Patents

Abstract

(57) [Summary] [Problem] To increase the range of design application and shorten the design period by accumulating design histories and reusing them at the time of case inference. An analysis of product knowledge, various design knowledge, analysis knowledge, design history knowledge, and the like, and a design case database that stores design cases in an object-oriented manner, and case correction rules, countermeasure rules, and procedure data are stored. Only the necessary knowledge in the general-purpose knowledge base 2 and the design case database 1 is constructed as a case, the actual case data is expanded in space, and the case that is the most similar to the current design among the expanded cases , And a case inference agent 3 for displaying a finally usable case, a simultaneous cooperative design unit 4 for simultaneously coordinating a plurality of design evaluations, and a case correction as cases are stored in the design case database 1 It has a learning agent 5 for accumulating and reinforcing rules and general countermeasure rules.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a product or process design system, and more particularly to a CAE expert system using CAE and knowledge processing.

[0002]

2. Description of the Related Art Conventionally, various systems using case inference have been proposed as design systems for products and processes.

For example, a design specification to be applied to a required design specification is determined in consideration of design constraints and interference relationships between specification items, and a past design having the same design specification is determined from a design case base. Search cases as similar design cases,
There is a similar design case search device (Japanese Patent Laid-Open No. 7-219899) which outputs the search result. However, this similar design case search apparatus can only apply knowledge of similar parts, and cannot completely acquire all knowledge including dissimilar parts.

Therefore, a case-based reasoning system has been proposed in which, when knowledge cannot be completely acquired, a case is modified and applied (JP-A-6-19714).

[0005] This case-based reasoning system is used for solving a problem from a plurality of cases stored in a storage unit in advance.
It is configured to search for the case that is most similar to the problem to be solved, correct the case based on a predetermined rule for a part different from the searched case and the problem, and additionally store the corrected case in the storage unit. I have.

[0006]

However, the above case-based reasoning system has a problem that rules for correction must be accumulated, and the case is limited to the initial design. . In addition, there was no method for evaluating a design solution or a system that used a design history for case inference.

On the other hand, the present inventors previously prepared an initial pipe according to design rules in designing a cooling pipe of an injection molded product, and evaluated a design solution using CAE (flow / cooling analysis for injection molding). Until the target quality (high cycle, demolding failure, etc.) is achieved, measures are extracted from the knowledge base, and an optimization method that automatically executes the measures is proposed. Maintenance of rules and countermeasures rules was necessary, and knowledge acquisition was an important issue.

The present invention has been devised to solve such a problem. The purpose of the present invention is to accumulate design histories and reuse them at the time of case inference to increase the range of design application and increase the design period. It is an object of the present invention to provide a design system using case-based reasoning and learning that aims at shortening the time.

[0009]

In order to solve the above-mentioned problems, a design system using case-based reasoning and learning of the present invention analyzes and designs product knowledge, various design knowledge, analysis knowledge, and design history knowledge. A design case database that stores cases in an object-oriented manner, a general-purpose knowledge base that stores case correction rules, countermeasure rules, and procedure data,
The case base editor that configures only the necessary knowledge in the design case database as a case, a case list creation unit that deploys actual case data in space, and a case list created by the case list creation unit. A case inference agent having a similar case search unit for inferring the case most similar to the design and a result display unit for displaying a finally usable case, a simultaneous cooperative design unit for simultaneously coordinating a plurality of design evaluations, A case modification rule learning unit that accumulates and strengthens the case modification rules as cases are accumulated in the design case database, and accumulates and strengthens general countermeasure rules themselves as cases are accumulated in the design case database. And a learning agent having a countermeasure rule learning unit.

That is, in the conventional optimization method using CAE, a design case database for accumulating cases, a case inference agent for analogizing and correcting similar cases from the design case database, and a case inference agent obtained from the case inference agent. A general-purpose knowledge base that applies general-purpose knowledge common to knowledge and examples, and a learning agent that learns these general-purpose knowledge bases are added.

[0011]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is an explanatory diagram showing the overall configuration of a design system using case-based reasoning and learning according to the present invention.

The design system using the case-based reasoning and learning is roughly composed of a design case database 1, a general-purpose knowledge base 2, a case reasoning agent 3, a simultaneous cooperative design unit 4, and a learning agent 5. .

The design case database 1 stores product knowledge, various design knowledge (design design knowledge, structural design knowledge, etc.), analysis knowledge, design history knowledge, etc., and design cases in an object-oriented manner.

The general-purpose knowledge base 2 has a case modification rule 2
1. The countermeasure rule 22 and the procedure 23 for shape creation, design, and correction are stored. The general-purpose knowledge base 2 applies general-purpose knowledge common to the case and the knowledge obtained from the case inference agent 3.

The case inference agent 3 includes a case base editor 31 for constructing only necessary knowledge in the design case database 1 as a case, a case list creation unit 32 for developing actual case data in a space, and this case list creation. Part 32
A similar case search unit 33 for estimating the case most similar to the current design from the cases developed in the above, a result display unit 34 for displaying a finally usable case, and a correlation / weight definition unit 35
And a similarity calculation formula definition unit 36.

The simultaneous cooperative design section 4 is a block for simultaneously and cooperatively performing a plurality of design evaluations. The request input section 41 for inputting data such as required functions and qualities. Condition creation unit 4 for setting
2. A tool execution unit (design simulation, functional simulation, CAE (structure, molding), design based on rules, etc.) 43 for evaluating an evaluation item, a result evaluation unit 44 for evaluating an execution result, and a result evaluation unit 44 A measure solution extractor 45 that extracts a measure from the general-purpose knowledge base 2 based on the evaluation result, a measure implementer 46 that implements the extracted measure, and a design solution display that displays the evaluation result (design solution) of the result evaluator 44. It comprises a unit 47 and a case storage unit 48.

The learning agent 5 includes a case modification rule learning unit 51 that accumulates and reinforces the case modification rules 21 of the general-purpose knowledge base 2 as the cases are accumulated in the design case database 1. It is configured by a countermeasure rule learning unit 52 that stores and strengthens the general countermeasure rules 22 of the general-purpose knowledge base 2 as they are stored.

Next, the overall operation in the case of actually designing a new case using the design system using the case-based reasoning and learning having the above configuration will be described with reference to the flowchart shown in FIG.

The data of the requested function, quality, and the like input to the request input unit 41 are input to the case inference agent 3 (step S1). The case inference agent 3 creates a list of cases based on the input data of required functions, quality, and the like (step S2), and infers past cases satisfying the required functions, quality, and the like from the design case database 1 (step S2). Step S3), and display the result on the result display unit 34 (Step S4).

Thereafter, in order to realize all the obtained knowledge, procedural rules are obtained from the general-purpose knowledge base 2 and input to the simultaneous cooperative design unit 4 (step S5). The simultaneous cooperative design unit 4 performs an initial design based on the rules obtained here (step S6), and evaluates the initial design (steps S7 to S10). The evaluation here is an evaluation using CG and VR such as design simulation and functional simulation (step S8), an evaluation using CAE such as structural analysis and formability simulation (step S9),
A plurality of evaluations, such as an evaluation using an empirical formula (step S10), are performed simultaneously. Then, when a defect is found in the evaluation result, a measure is extracted from the general-purpose knowledge base 2 (steps S11 and S12). Here, when there is no countermeasure rule in the countermeasure rule 22 of the general-purpose knowledge base 2,
A request is made to the case inference agent 3 to extract a case in which a similar measure has been taken in the past. Then, the obtained countermeasure is implemented (step S13), and the evaluation is performed again (step S7). This work (Step S7 to Step S13)
Is repeated, and when the design solution becomes optimal, the optimized design solution is displayed on the design solution display section 47 (step S14).

Finally, these series of operations are stored as cases in the design case database 1 (step S15).
The learning agent manages the state of accumulation as needed, and discovers general rules, strengthens a specific rule group, and the like (step S16).

The overall processing flow has been described above. Specific processing procedures in the case inference agent 3, the simultaneous cooperative design unit 4, and the learning agent 5 will be individually described below. (1) The processing in the case inference agent 3 will be described with reference to the flowchart shown in FIG.

When the case inference agent 3 receives the obtained required function and quality, and product knowledge such as shape characteristics and material information, it expands the design case database 1 and sets them in the problem part (step S21, step S21). S22).
In the problem part, for example, “product group”, “product name”, “volume”, “surface area”, “boundary condition”, and the like are set. Thereafter, the number of cases M in which the same item is set in the problem part is calculated while comparing with the problem part of the past case stored in the design case database 1 (step S).
23) Then, the number N of problem parts in each case is calculated (step S24), and the inter-case distance is calculated for each case (step S25).

That is, one case is extracted from the calculated cases (step S26), and the distance between the problem part An and the problem part Bn of the extracted case is calculated. In this case, when the item set in the problem part is a numerical value X, and the numerical value X indicates an error, An and Bn are calculated using the distance calculation formula 1.
Is calculated (steps S27, S28, S2).
9, S30).

[0026]

(Equation 1)

[Distance calculation formula 1] y = 1.0 (where -αX <x <αX) On the other hand, the item set in the problem part is a numerical value X, and the distance calculation formula is set as distance calculation formula 2. If you have
The distance y between An and Bn is calculated using the distance calculation formula 2 (steps S27, S28, S29, S31).

[0028]

(Equation 2)

[Distance calculation formula 2] y = 0.0 (where x <
−αX) y = (1 / αX) · x + 1 / (− α) (However, −α
X <x <X) y = 1 (where x =
X) y = (1 / −αX) · x + 1 / α (where X <
x <αX) y = 0.0 (However, αX
<X) On the other hand, if the item set in the problem part is a character, the distance y between An and Bn is calculated using the distance calculation method 3 (steps S27, S28, S29, S32). The distance calculation method 3 here is, for example, a perfect matching method. That is, the character (name) of the problem part An is, for example, “flange”.
When the character (name) of the problem part Bn is “flange”, y = 1.0, the character (name) of the problem part An is “flange”, for example, and the character (name) of the problem part Bn is set. Is “flange”, y = 0.0. In addition, a distance calculation method based on an ontology using a thesaurus is possible.

The entire distance between An and Bn calculated in this way is stored as the distance Y (n) of the case (step S33). Such processing is performed in step S2.
After repeating the process up to the number of problem copies N calculated in Step 4 (Step S34), the inter-case distance obtained by comprehensively evaluating the extracted cases as a whole is calculated (Step S35). At this time, in addition to numerical values and text,
Natural languages, semantic distances, images, shapes, etc. may also be handled.

That is, in the calculation of the inter-case distance, if the individual knowledge is not weighted, the inter-case distance is calculated using the inter-case distance calculation method 1 (step S36,
S37).

[0032]

(Equation 3)

In the calculation of the inter-case distance, if the individual knowledge is weighted, the inter-case distance is calculated using the inter-case distance calculation method 2 (steps S36 and S3).
8, S39). That is, the weight changes in each region even when calculating the entire inter-case distance.

[0034]

(Equation 4)

The distance L between the cases calculated as described above
(M) is stored (step S40).

Such steps S26 to S4
0 is repeatedly performed up to the number M of cases calculated in step S23 (step S41), and the calculation result is
The cases are displayed in ascending order of the inter-case distance (step S42), the applied knowledge is displayed (step S43), and the process in the case inference agent 3 ends.

(2) Next, the processing in the procedure execution unit will be described with reference to FIG.
This will be described with reference to the flowchart shown in FIG.

In the procedure execution unit, the case inference agent 3
Based on the knowledge obtained in (1), all the knowledge required for performing the initial design is set (step S61). If the knowledge is insufficient during the execution, the case inference agent 3 reconfigures the problem part and the answer part again, executes the part again, and acquires the knowledge. The procedure execution unit creates an initial design solution based on the design procedure obtained from the case (steps S62 and S6).
3). In this case, shape creation, mesh creation for analysis, and addition of boundary conditions are performed (steps S64, S65, S6).
6). In the shape creation, the history of CAD commands for executing the procedure and the base shape thereof are managed in a database.

(3) Next, the processing in the simultaneous cooperative design unit 4 will be described with reference to the flowchart shown in FIG.

In the simultaneous cooperative design unit 4, first, an evaluation item and a criterion are set in order to simultaneously perform a plurality of evaluations (step S71). Then, select a tool that can evaluate those evaluation items (for example, CG software for evaluating designability, structural analysis software for evaluating structure, etc.),
Information that satisfies these analysis conditions is set from the application knowledge of the case inference agent 3 (steps S73 to S7).
6). Then, these tools are simultaneously executed according to the set analysis conditions.

That is, a design simulation (step S77) and a function simulation (step S7)
8), structural analysis (step S79), and molding simulation (step S80) are simultaneously executed, and the evaluation results are output individually for each (steps S81 to S8).
4). Then, the result evaluation unit 43 aggregates the individual evaluation results, extracts and executes a measure until the total evaluation result satisfies the required quality (steps S86, S87, S8).
8).

FIG. 6 shows a specific structure of the countermeasure rule 22 stored in the general-purpose knowledge base 2, in which each defect is associated with each countermeasure with a weight. The weights change from time to time, and new items and new countermeasures for defects also increase from time to time. From the contents of the countermeasure rule 22, for example, when it is desired to reduce the stress concentration at the corner without increasing the volume, it is effective to execute the countermeasure 1 (large R). In addition, when it is desired to suppress the deflection without increasing the volume, it is effective to execute the measures 1 (large R) and the measure 5.

In addition, the general-purpose knowledge base 2
If the countermeasure rule 22 has no countermeasure, the case inference agent 3 is inquired again to infer the past countermeasure.

The extraction and execution of such a measure are repeated, and when the total evaluation result satisfies the required quality, the result is displayed (step S89).

FIGS. 7A and 7B show examples of the individual evaluation results and the aggregated evaluation results thus obtained. That is, in measure 1, the evaluation value of the designability is 5, the evaluation value of the functionality is 7, the evaluation value of the product strength is 6, the evaluation value of the moldability is 3, and the total evaluation value is 21.
It has become. Similarly, in the case of the measure 2, the total evaluation value is 23, and in the case of the measure 3, the total evaluation value is 1
6. In measure 4, the total evaluation value is 24 (see FIG. 7A).

Here, assuming that the evaluation value that satisfies the required quality is, for example, 30, the simultaneous cooperative design unit 4 first sets the result of the measure 4 having a high total evaluation value as an initial measure. Evaluation value 4 of formability in the countermeasure plan,
The evaluation value 8 in the newly extracted countermeasure plan is replaced (measure 1), and the evaluation value 3 of the product strength is then replaced with the evaluation value 7 in the newly extracted countermeasure plan (measure 2) to obtain a total. Is obtained (see FIG. 7B). The evaluation value 32 is an evaluation value that satisfies the required quality.

(4) Next, the processing in the learning agent 5 will be described with reference to the flowchart shown in FIG.

Here, case accumulation is performed to accumulate as design cases when the design is actually completed (step S9).
1) The learning agent 5 is executed according to the accumulation status (step S92). For example, learning of the case modification rule (step S93) is executed when the number of cases has increased by 50, but learning of the countermeasure rule (step S94).
The learning of the procedure (step S95) is executed every time.

Learning of case modification rules (step S93)
Then, all past cases are analyzed (step S96), and those having a high track record of being executed by correlation between the problem part and the answer part are converted into rules (step S97) and added to the case modification rule 21 of the general-purpose knowledge base 2. Register (step S98).

In the learning of the countermeasure rule (step S94), the weight of the countermeasure for the one having a large effect is strengthened from the history of the countermeasures performed this time (step S99, step S99).
(S100) If the measure is not in the past, the sheet of the measure rule is additionally updated to the measure rule 22 of the general-purpose knowledge base 2 (step S101).

Regarding the learning of the procedure (step S95), if the procedure does not exist in the past, the procedure name and the CAD command history are additionally registered in the procedure base 21 of the general-purpose knowledge base 2 as a general-purpose one (step S95). Step S10
2, S103).

[0052]

According to the design system using case-based reasoning and learning of the present invention, a design case database for accumulating cases and a similar case from the design case database are analogized to a conventional optimization method using CAE. A configuration in which a case inference agent that modifies by adding a general knowledge base that applies general knowledge common to the knowledge obtained from the case inference agent and the case, and a learning agent that learns these general knowledge bases, is added. As a result, not only maintenance of design rules and countermeasure rules, which were required in the past, becomes unnecessary, but also necessary rules can be secured.
Therefore, it is possible to quickly and surely respond to design measures.

[Brief description of the drawings]

FIG. 1 is an explanatory diagram showing the overall configuration of a design system using case-based reasoning and learning according to the present invention.

FIG. 2 is a flowchart for explaining an overall operation when a new case is actually designed using a design system using case-based reasoning and learning of the present invention.

FIG. 3 is a flowchart for explaining a processing operation in a case inference agent.

FIG. 4 is a flowchart illustrating a processing operation in a procedure execution unit.

FIG. 5 is a flowchart for explaining a processing operation in a simultaneous cooperative design unit.

FIG. 6 is a table showing a specific structure of a countermeasure rule stored in a general-purpose knowledge base.

7A is a chart showing an example of an individual evaluation result and an aggregated evaluation result, and FIG. 7B is a chart showing a result of optimizing a collaborative design.

FIG. 8 is a flowchart for explaining a processing operation in the learning agent 5.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Design case database 2 General-purpose knowledge base 3 Case inference agent 4 Simultaneous collaborative design part 4 5 Learning agent 21 Case modification rule 22 Countermeasure rule 23 Procedure data 31 Case base editor 32 Case list creation part 33 Similar case search part 34 Result display part 35 Correlation / weight definition unit 36 Similarity calculation expression definition unit 41 Request input unit 42 Condition creation unit 43 Tool execution unit 44 Result evaluation unit 45 Countermeasure extraction unit 46 Countermeasure execution unit 47 Design solution display unit 48 Case accumulation unit 51 Case modification rule Learning unit 52 Countermeasure rule learning unit

Claims (1)

[Claims] 1. A design case database that stores product knowledge, various design knowledge, analysis knowledge, design history knowledge, and the like, and design cases in an object-oriented manner, and stores case correction rules, countermeasure rules, and procedure data. A general-purpose knowledge base, a case base editor that configures only necessary knowledge in the design case database as cases, a case list creation unit that deploys actual case data in space, and a case developed by this case list creation unit A case inference agent that has a similar case search unit that infers the case most similar to the current design from among the cases and a result display unit that displays the finally usable cases. A collaborative design unit, and a case modification that accumulates and strengthens the case modification rule as cases are accumulated in the design case database. Case-based reasoning, comprising: a rule learning unit; and a learning agent having a countermeasure rule learning unit that stores and strengthens general countermeasure rules themselves as cases are stored in the design case database. And a design system using learning.