JP7301547B2 - Design support device - Google Patents

Description

The present disclosure relates to a design support device for supporting system design, particularly vehicle design.

In a system consisting of many elements, such as a vehicle, each element such as a frame (member), an engine, wheels, etc., are interrelated and cooperate with each other to drive. Therefore, if the design of one of the elements is changed, the other elements will be affected, and it is not easy to change the design of the system. Therefore, a design support system for designing a vehicle automatically changes the specification values of other related elements based on the input specification values that are input with the design change of one element. (For example, Patent Document 1).

JP-A-2009-37509

In a system with many elements such as a vehicle, each element is interrelated, so it is desirable to base the design on a wider range of information about more elements. Therefore, the design support apparatus is required to be able to provide appropriate information to the user based on a wider range of information on design.

In view of the above background, it is an object of the present invention to enable a design support apparatus to provide appropriate information to a user based on a wider range of information regarding design.

In order to solve the above problems, one aspect of the present invention is a design support device (1), comprising: a data storage unit (3) for classifying and storing design-related data into a plurality of different abstract levels; an information block (5) that stores as graphs (31, 33, 35) defining relationships; a user interface (7) that accepts queries from users in text form; a conversation engine (9) for processing to extract question intent; a knowledge engine (11) for deciding how to trace the graph based on the question intent extracted by the conversation engine; and the knowledge engine. an expert frame (13) for extracting said data corresponding to answers to said queries from said data store by tracing said graph according to said tracing method determined by said The answer is constructed based on the data extracted by the expert frame, and the answer is output to the user interface, and the user interface notifies the user of the answer.

According to this configuration, an answer is constructed by tracing a graph that defines data relationships at multiple levels of abstraction. This provides the user with more pertinent information because the answer is built on a wider range of data than when tracing only one level of abstraction.

In the above aspect, the data storage unit stores parameters relating to members constituting a past design target, constraints on the members, goals set at the time of development of the design target, or strategies toward the goals, and The graphs are a first graph (31) containing blocks (S11-S41) associated with said parameters or said constraints and a second graph containing elements (G11-G43) associated with said goals or said strategies. (33) and a knowledge graph (35) showing the relationship between the first graph and the second graph, wherein the knowledge graph includes nodes (s11 to s41, g11 to g41), an edge (e1) connecting the nodes (s11 to s41) corresponding to the two blocks related to each other, and the nodes corresponding to the two elements (g11 to g41) related to each other connecting edges (e2) and at least one edge (e3) connecting said nodes (s31, g32) respectively corresponding to said block (S31) and said element (G32) related to said block .

According to this configuration, a knowledge graph is constructed based on the first graph and the second graph. As a result, the design support apparatus constructs an answer based on a wide range of data compared to the case where the knowledge graph is constructed based on either one of the first graph and the second graph. Able to provide appropriate information.

In the above aspect, the knowledge engine selects the elements of one of the second graphs based on the intent of the question extracted by the conversation engine, and traces the nodes corresponding to the selected elements. and the expert frame extracts the constraints most relevant to the intent of the question extracted by the conversation engine by tracing the knowledge graph from the starting node. and output to the conversation engine.

According to this configuration, the expert frame extracts constraints on structures and parts included in the query, and constructs answers based on the extracted constraints. This allows the user to easily acquire constraints that should be taken into consideration when designing the structures and parts included in the query.

In the above aspect, the data store includes any one of requirements for how the system under design should function, laws of physics that describe the properties of the system under design, and the process of design development. A first layer (21) and a second layer containing system component parameters designed based on the requirements and physical laws described in the first layer information and functions required of the components. (22), a third hierarchy (23) containing data obtained by computer simulation, and a fourth hierarchy (24) containing data obtained by experiments with real objects, and hold the data. , the expert frame may be capable of extracting the constraints classified into at least one of the first hierarchy or the second hierarchy.

According to this configuration, the design support device can output an answer based on the constraints included in the first hierarchy or the second hierarchy. As a result, the design support device can provide the user with more useful design information including design processes and functions, as well as simulation results and experimental results using the actual product.

In the above aspect, the information block holds, for each of the edges, a weight indicating the strength of the relationship between the two nodes connected by the edge, and the expert frame generates the conversation engine based on the weight. and the conversation engine may construct the answer to the user based on the extracted constraints.

According to this configuration, by using the weight, it is possible to easily acquire a constraint closely related to the input element. In addition, experienced design developers determine the weights based on their experience, so that users can be notified of answers based on the experience of the experienced design developers.

In the above aspect, when a plurality of said answers have been constructed based on said constraints extracted by said expert frame, said conversation engine may further construct a question for selecting said answers.

In the above aspect, by selecting the answer based on the question, the user can be quickly notified of the appropriate answer.

In the above aspect, there may be provided a knowledge input system (73) for updating the data and the graph based on input from the user interface.

According to this configuration, the data stored in the data storage unit and the graph stored in the information block are changed according to the input from the user, so it is possible to accurately provide information to the user.

According to the above configuration, the design support device can provide appropriate information to the user based on a wider range of information regarding design.

1 is a configuration diagram of a design support device according to the present invention; (A) Explanatory diagram of the first graph, and (B) Explanatory diagram of the second graph Knowledge Graph illustration Diagram showing an example of a method selection model A diagram showing an example of a thesaurus and an intention understanding model Sequence diagram for explaining the processing performed by the design support device

An embodiment in which a design support device according to the present invention is applied to vehicle design support will be described below with reference to the drawings.

Design support here means receiving questions (queries) from users involved in design, data on existing products (existing vehicles) and their parts, goals and strategies created in the design process of existing products, etc. is to derive an answer based on the data, etc., and notify the user of the answer.

The design support device 1 is composed of a computer having a central processing unit (CPU), a memory, and a hard disk 2 . As shown in FIG. 1, the design support device 1 includes a data storage unit 3, an information block 5, a user interface 7, a knowledge engine 11, and an expert frame 13. FIG.

The data storage unit 3 stores in the hard disk 2 the structure of an existing product and the design data including the goals and processes during the development of the product, classified into a plurality of different levels of abstraction. More specifically, the data storage unit 3 divides the data into four hierarchies, first to fourth, in descending order of abstraction level, and stores the data. The data 21 of the first layer defines the concept, parameters of the system to be designed (vehicle in this embodiment), requirements of how the system should function, physical laws describing them, equations , and/or data relating to the design process. The data 23 of the second hierarchy includes the parameters of the members constituting the system designed based on the requirements described in the information of the first hierarchy and the laws of physics, and the data related to the functions required of the members. Data 25 of the third layer includes data obtained by computer simulation. The fourth tier of data 27 represents the physical space of the actual materials, test samples, and includes data, drawings, photographs, reports and other physical evidence obtained from real world experiments.

The information block 5 stores in memory a graph defining data relationships. The graphs stored in the information block 5 are the first graph 31 associated with the data on the structure of the existing product in the data storage unit 3 and the data on the goals and processes during development of the product in the data storage unit 3. and a knowledge graph 35 defining the relationship between the first graph 31 and the second graph 33 .

In other words, the data held in the information block 5 of the design support device 1 includes a first graph 31, a second graph 33, and a knowledge graph 35 that defines the relationship between the two graphs 31 and 33. It has a data structure X. The first graph 31, the second graph 33, and the knowledge graph 35 are databases each having a graph structure including vertices and directed lines (arrows) connecting the vertices based on relevance. database.

The first graph 31 is a graph database showing the system structure of existing products described by SysML as shown in FIG. 2(A), and includes a plurality of blocks as vertices. Each block includes a table of contents, a summary, a file name, etc. of the data stored in the data storage unit 3, and is associated with the data stored in the data storage unit 3. FIG. The first graph 31 includes, as blocks, a General Constructor (hereinafter, general block 37) and a Constraint block (hereinafter, constraint block 39). The general block 37 is associated with data containing parameters (load capacity, dimensions, weight, etc.) relating to existing products (i.e., past design objects) as a whole, or data containing parameters relating to members that make up past design objects. is a block. The constraint block 39 is a block associated with data concerning constraints on the entire existing product (entire past design objects) or data concerning constraints on members constituting the existing product (past design objects).

FIG. 2A shows part of the first graph 31 according to this embodiment. The first graph 31 includes blocks S11, S21-S27, S31, and S41 (hereinafter referred to as S11-S41). S11 is a general block 37 associated with data relating to parameters such as dimensions, weight and load capacity of the vehicle as a whole. S21 is a general block 37 associated with data on parameters such as dimensions and weights of members forming the frame of the vehicle. S22 is a general block 37 associated with data relating to parameters such as engine room dimensions. S23 is a general block 37 associated with data on parameters such as dimensions of the side members that make up the members, S24 is a general block 37 associated with data on parameters such as dimensions of the cross members that make up the members, S25 is a general block 37 associated with data relating to parameters such as dimensions of side members forming the engine room. S23 and S25 may be configured to be associated with the same data. S26 is a constraint block 39 associated with data describing functions generally required of members, including side members and cross members. S27 is a constraint block 39 associated with data describing the functions required of the side members forming the engine room. S31 is a general block 37 associated with load bearing simulation data for a single side member, and S41 is a general block 37 associated with load bearing test data for a side member.

The general blocks 37 are connected to each other by solid line arrows based on the inclusion relationship of the corresponding members. However, the direction of the arrow is set so as to point from the general block 37 corresponding to the included member to the general block 37 corresponding to the included member. More specifically, S21 and S22 are each connected by a solid arrow pointing to S11. This means that the vehicle (S11) contains the member (S21) and the engine room (S22). S23 and S24 are each connected by a solid arrow pointing to S21. This means that the members (S21) include side members (S23) and cross members (S24). S25 is connected by a solid arrow to S22. This means that the side member (S25) is included in the engine room (S22). S31 is connected by a solid arrow to S23. This means that the load bearing simulation data (S31) is included as the information on the side member (S23). S41 is connected to S23. This means that the load bearing test (experiment) data (S41) for the side member is included in the information on the side member (S23).

Constraint block 39 is connected by an arrow to general block 37 corresponding to the member to be constrained. However, the arrow is configured to point from the constraint to the general block 37 corresponding to the member to which the constraint is applied. More specifically, S26 is connected by dashed arrows to S23, S24 and S25. This means that the requirements stated in S26 are restrictions on the side members (S23, S25) and the cross member (S24). Similarly, S27 is connected by a dashed arrow to S25. This means that the request described in S27 is a constraint on the side member (S25) that constitutes the engine room (S22). However, in FIG. 2A, the inclusion relationship and the constraint relationship are indicated by arrows for the sake of convenience, but the present invention is not limited to this aspect. For example, the arrowheads of the arrows in FIG. may be

The first graph 31 shows the blocks associated with the data 21 of the first hierarchy, the blocks associated with the data 23 of the second hierarchy, the blocks associated with the data 25 of the third hierarchy, and the blocks associated with the data 25 of the third hierarchy. Blocks associated with data 27 of 4 layers. More specifically, as shown in FIG. 2A, S11 is a block associated with the data 21 of the first hierarchy, and S21 to S25 are blocks associated with the data 23 of the second hierarchy. , S31 is a block associated with the data 25 of the third hierarchy, and S41 is a block associated with the data 27 of the fourth hierarchy.

The second graph 33 is a graph database showing the target structure during development described by GSN as shown in FIG. 2(B), and includes multiple elements as vertices. Each element is associated with the data stored by the data storage unit 3, and each element includes a table of contents and a summary of the associated data. Elements are categorized into three types, goals, strategies, and evidence, depending on the type of data associated with them. Elements classified as goals are associated with past design objects, that is, data indicating goals set during the development of existing products. Elements classified as strategies are associated with data that indicate strategies for achieving goals set during the development of existing products. Elements classified as evidence are associated with data acquired through experiments, computer simulations, etc., in the past development process of design objects. Hereinafter, for simplification, each element in the second graph 33 will be described using the name of its type (goal, strategy, evidence) as necessary.

FIG. 2B shows part of the second graph 33 of this embodiment. The second graph 33 includes elements G11, G21, G22, G31-G33, and G41-G43 (hereinafter referred to as G11-G43). In FIG. 2B, goals are indicated by rectangles, strategies by parallelograms, and evidence by ellipses. The second graph 33 shows the goal (G11) described as improving the safety performance of the vehicle, the strategy (G21) of improving the load capacity of the vehicle body for the goal (G11), and the side member design for the strategy (G21). Goal (G22), strategy (G31) of simulation for goal (G22) of side member design, goal (G32) of side member load capacity calculation for strategy (G31), and load capacity calculation result for goal (G32) and evidence (G33) that In this embodiment, the second graph 33 further includes a strategy (G41) of load bearing test for the goal (G22) of designing the side member, a goal (G42) of executing the actual test for the strategy (G41), and a goal Evidence (G43) which is the actual test result for (G42).

As shown in FIG. 2B, goals and strategies for achieving the goals are connected to each other by solid-line arrows pointing from the goals to the strategies. A strategy and a goal for realizing that strategy are connected to each other by solid-line arrows pointing from the strategy to the goal. More specifically, G11 is connected to G21 by a solid arrow pointing to G21, G21 is connected to G22 by a solid arrow pointing to G22, G22 is connected to G31 by a solid arrow pointing to G31, and to G41 by a solid arrow pointing to G41. are respectively connected to G41 by arrows. G31 is connected to G32 by a solid arrow pointing to G32, and G32 is connected to G33 by an arrow pointing to G33. G41 is connected to G42 by a solid arrow pointing to G42, and G42 is connected to G43 by an arrow pointing to G43. These arrows correspond to the design process.

The second graph 33, like the first graph 31, includes elements associated with data corresponding to the first to fourth hierarchies. In this embodiment, G11 is associated with the data 21 of the first hierarchy, G21 and G22 are associated with the data 23 of the second hierarchy, G31 to G33 are associated with the third hierarchy, and G41 to G43 are associated with the third hierarchy. It is associated with the data 27 of the fourth layer.

As shown in FIG. 3, the knowledge graph 35 has a first graph 31 and a second graph 33, nodes (S11 to S41, G11 to G43) as vertices, and arrows (edges) connecting the nodes. It was obtained by converting to a graph database.

More specifically, the knowledge graph 35 is obtained as follows. First, the blocks S11 to S41 of the first graph 31 are converted into nodes s11 to s41, respectively, and connected by an edge (first edge e1) indicating inclusion and constraint, as in FIG. 2(A). Next, the elements G11 to G43 of the second graph 33 are converted to nodes g11 to g43, respectively, and the nodes g11 to g43 are connected by edges (second edge e2), as in FIG. 2B. However, the nodes s11 to s41 corresponding to the blocks S11 to S41 of the first graph 31 are connected by arrows pointing in the opposite direction to the first graph 31, respectively.

Furthermore, the blocks S11 to S41 of the first graph 31 and the elements G11 to G43 of the second graph 33 that have a high degree of commonality are selected from the nodes corresponding to the blocks of the second graph 33 to the first The elements of the graph 31 of 1 are connected by an edge (third edge e3) directed to a node corresponding to the element. In this embodiment, a node g32 corresponding to the element G32 (withstanding load calculation) of the second graph 33 and a node s31 corresponding to the block S31 (withstanding load simulation data of the side member) of the first graph 31 are , are connected by an edge going from node g32 to node s31. The commonality between a block and an element may be determined based on the commonality of words and images included in each content. With this configuration, each node is connected by an edge based on the design development process and design considerations. As a result, the knowledge graph 35 holds the relationship of the data stored in the data storage unit 3 .

Each of the first edge e1, the second edge e2, and the third edge e3 is given a weight, which is a numerical value indicating the strength of the relationship between the two nodes connected by the edge. The weights are determined based on the past experiences of experienced design developers (experts, experts), questions asked by users, and the like. For example, the weights may be determined by supervised machine learning in which experts input questions into the design support device 1 and evaluate answers derived based on their own experiences. In this embodiment, the weight is set to increase as the relationship between two nodes connected by an edge becomes stronger. As shown in FIG. 3, the weight is represented by a numerical value between 0 and 1.

The user interface 7 is a device for accepting a question (query) from a user, outputting it in a text format, and notifying the user of an answer. Here, the text format means data composed only of characters that can be understood by the user.

In this embodiment, the user interface 7 includes a microphone 41 into which voice from the user is input. When there is a voice input from the user to the microphone 41, the user interface 7 accepts it as a query by converting the voice into a text format. The user interface 7 then outputs the query to the conversation engine 9 in text form. The user interface 7 may further include a keyboard 43 and an input/output port 45 (USB port or LAN port) for storing numerical data such as drawings, photographs, and simulation results in the hard disk 2 .

The user interface 7 includes a speaker 47 that accepts textual data from the conversation engine 9 and outputs it as voice to the user. The user interface 7 converts the text-format data from the conversation engine 9 into voice and notifies the user of the text-format data by voice. In this embodiment, the user interface 7 includes a monitor 49 for displaying drawings, experimental data, photographs, images, and textual data.

The conversation engine 9 consists of software executed by the CPU. The conversation engine 9 holds a dictionary 51 for parsing textual data and an intent understanding model 52 . In this embodiment, the dictionary 51 and the intent understanding model 52 are each stored in memory.

The dictionary 51 includes a synonym dictionary 51a. The thesaurus 51a classifies words based on their meanings, and as shown in FIG. 5, holds a plurality of semantics and synonyms corresponding to the semantics. A sememe indicates the sense associated with a word, and a synonym includes multiple words with the same sense.

As shown in FIG. 5, the intention understanding model 52 associates an intention 52a of a question with a query corresponding to the intention 52a of the question, which is assumed to be uttered by the user (hereinafter referred to as assumed text 52b). database. In the intention understanding model 52, for example, hypothetical texts 52b corresponding to the intention 52a of the question "Increase load capacity" include "I want to increase the load capacity of the side frame", "I want to increase the load capacity of the side frame", and the like. It is included. Further, the intent understanding model 52 may include an intent 52a for the question "specification construction" and an assumed text 52b corresponding to it "I want to construct a specification that satisfies performance." Further, the intention understanding model 52 may include an intention 52a of the question "factor analysis" and an assumed text 52b corresponding to it "want to analyze the cause of failure". As shown in FIG. 5, intention identifiers 52c serving as IDs are assigned to intentions 52a of questions.

The conversation engine 9 executes natural language processing on queries input from the user interface 7 , extracts keywords from the queries, and outputs them to the knowledge engine 11 . The keyword here is an index word for outputting an answer based on the data held by the data storage unit 3 and the information block 5 . The keywords extracted by the conversation engine 9 are preferably set based on the table of contents and summaries described in the blocks of the first graph 31 and the elements of the second graph 33 .

After that, the conversation engine 9 refers to the intent understanding model 52 to understand the intent of the question indicated by the query and outputs it to the knowledge engine 11 . More specifically, the conversation engine 9 searches the hypothetical text 52b of the intent understanding model 52 for any that correspond to the query. If there is a corresponding query, the intention 52a of the corresponding question is output to the knowledge engine 11. If not, the synonym dictionary 51a is referred to and the words included in the query are replaced with synonyms. For example, when receiving a query "I want to increase the load bearing capacity of the side frame", the conversation engine 9 refers to the thesaurus 51a and replaces "increase" with "increase" to obtain "increase the load bearing capacity of the side frame". I want to increase it" (hereinafter referred to as conversion query). Next, the conversation engine 9 searches the assumed text 52b of the intent understanding model 52 for the same as the conversion query. If there is the same question, the intention 52a of the corresponding question is output to the knowledge engine 11. If not, the thesaurus 51a is referred again to replace the query, and the assumed text 52b of the intention understanding model 52 is retrieved. do. In this way, by replacing the words included in the query with synonyms, the user's question intention is appropriately output even if the query is ambiguous or the assumed text 52b does not match the query. can do.

When the conversation engine 9 receives an input from the expert frame 13 or the knowledge engine 11, the conversation engine 9 performs natural language processing on the input data to construct a text format answer and outputs it to the user interface 7 as text format data. . However, when the input from the expert frame 13 constructs a plurality of answers, the conversation engine 9 constructs a question for selecting an answer and outputs it to the user interface 7 as data in text format. In addition, the conversation engine 9 causes the monitor 49 to display drawings, experimental data, photographs, and images stored in the data storage unit 3 as necessary.

The knowledge engine 11 is configured as software executed by the CPU. The knowledge engine 11 determines the tracing method of the knowledge graph 35 for obtaining the answer based on the output result of the conversation engine 9 and outputs it to the expert frame 13 .

For example, when the intent of the question is "specification construction," the knowledge engine 11 uses a tracing method of tracing from a node associated with data in a higher hierarchy to a node associated with data in a lower hierarchy, and the question If the intention of is "factor analysis", it is good to output a trace method of "going back to find the wrong design method from the failed goal".

In this embodiment, the knowledge engine 11 holds a method selection model 61 . The method selection model 61, as shown in FIG. 4, is a database containing a question intent 61a and a tracing method 61b corresponding to the question intent 61a. The method selection model 61 may include intent identifiers 61c corresponding to each intent of the question. In this embodiment, the method selection model 61 is stored in memory.

The intent 61a of the question includes "increase load capacity". "Increase in withstand load" is associated with a constraint trace as the trace method 61b. A constraint trace is a constraint associated with data belonging to the first layer or the second layer, which is connected only on the arrowhead side of the arrow by following (tracing) the knowledge graph 35 from the start node to the edge. is a method of extracting a node corresponding to .

The intention 61a of the question includes "display simulation results". "Display of simulation result" is associated with the calculation result trace as the trace method 61b. Calculation result tracing is a method of extracting nodes corresponding to data belonging to the third layer by tracing edges from a start node.

The intent 61a of the question included in the method selection model 61 may include "display of experimental results". The "display of experimental results" may preferably be associated with a tracing method of extracting nodes corresponding to data belonging to the fourth layer by tracing edges from the start node.

In this embodiment, the knowledge engine 11 outputs to the expert frame 13 a start node serving as a starting point for tracing and a tracing method based on the output result of the conversation engine 9 . More specifically, first, one element of the second graph 33 is selected based on the keyword extracted by the conversation engine 9, and the node corresponding to the selected element is extracted as the starting node. For example, the knowledge engine 11 may be configured to extract the node corresponding to the element containing the most input keywords (eg, load bearing, side member, etc.) as the starting node.

After that, the knowledge engine 11 refers to the method selection model 61 and outputs a tracing method corresponding to the intent of the question input from the conversation engine 9 . More specifically, when the knowledge engine 11 receives the input of the question intent of "increase load capacity", it selects the question intent 61a of the method selection model 61 that is the same as the question intent input from the conversation engine 9. Extract things. At this time, the knowledge engine 11 may extract from the intention identifier 61 c of the method selection model 61 the same intention identifier as the intention identifier corresponding to the intention of the question input from the conversation engine 9 . After that, the knowledge engine 11 refers to the method selection model 61 to obtain a tracing method, ie, a constraint trace, corresponding to the intent or intent identifier of the extracted question. After that, the knowledge engine 11 outputs to the expert frame 13 an instruction to execute constraint tracing with the extracted node as the starting node.

When the knowledge engine 11 receives the intention of the question "display of simulation results", it refers to the method selection model 61 and obtains the corresponding calculation result trace as the trace method. After that, the knowledge engine 11 outputs to the expert frame 13 an instruction to execute the calculation result trace with the extracted node as the start node.

In this embodiment, the knowledge engine 11 has an inference engine 62 that makes inferences based on the intent of the question output by the conversation engine 9 . The inference engine 62 includes the existing vehicle structure, past simulation data, past test data, and past failure analysis stored in the data storage unit 3, that is, information of the third hierarchy and information of the fourth hierarchy. It refers to the data, makes inferences using known algorithms, and outputs appropriate countermeasures. The inference engine 62 outputs data indicating a countermeasure to the conversation engine 9 as necessary.

The expert frame 13 is configured as software executed by the CPU. The expert frame 13 traces the knowledge graph 35 stored in the information block 5 according to the tracing method determined by the knowledge engine 11, and extracts the node corresponding to the answer to the query from the user as an answer node.

However, when the expert frame 13 extracts a plurality of nodes that can be answer nodes (hereinafter referred to as answer node candidates), the node most closely related to the keyword extracted by the conversation engine 9 is extracted from the answer node candidates as the answer node. do. At this time, the expert frame 13 may use weights to determine the strength of the relationship. In this embodiment, when the expert frame 13 extracts a plurality of answer node candidates, the knowledge graph 35 is used to extract the shortest route from the start node to each node, and the average value of the weights of the nodes provided on that route is calculated. is calculated, and the node most closely related to the start node is extracted as the answer node. The expert frame 13 then extracts the data associated with the answer node from the data storage unit 3 and outputs it to the conversation engine 9 as data corresponding to the answer to the query. After outputting the data corresponding to the answer to the conversation engine 9, the expert frame 13 increases the weight of the edge on the shortest route.

At this time, the expert frame 13 may output the answer node to the inference engine 62 when the answer node is a node corresponding to the strategy. The inference engine 62 makes inferences according to the design method described in the strategy corresponding to the answer node, and appropriate countermeasures etc.) and output to the conversation engine 9. Thereby, the conversation engine 9 notifies the user of the coping method derived by the inference engine 62 from the user interface 7 .

Also, when the expert frame 13 extracts a plurality of nodes that can be answer nodes, it may output information on all extracted nodes to the inference engine 62 . The inference engine 62 may rank the nodes in order of suitability as an answer and output the data associated with the nodes to the conversation engine 9 along the rank order. The inference engine 62 may perform the ordering by calculating the average weight on the path from the starting node to each node and sorting the nodes in descending order of the average weight. Also, the inference engine 62 may output the average value of the calculated weights as edge strength to the conversation engine 9 together with the data. At this time, the conversation engine 9 may notify the user from the user interface 7 of the data received from the inference engine 62 and the strength of the corresponding edge.

In this embodiment, the design support device 1 further comprises an analysis platform 71 and a knowledge input system 73 . Both the analysis platform 71 and the knowledge input system 73 are configured as software executed by a CPU.

The analysis platform 71 provides an environment in which known algorithms required for inference by the inference engine 62 can be executed. In this embodiment, the analysis platform 71 can execute algorithms such as Bayesian models, neural networks, decision trees, and image recognition. The analysis platform 71 may also be configured to provide an environment for executing natural language processing algorithms performed in the conversation engine 9 .

The knowledge input system 73 accepts an input from the user interface 7 and uses a natural language processing algorithm to convert the data stored in the data storage unit 3 into the first graph 31 and the second graph held in the information block 5. The graph 33 and the knowledge graph 35 are updated respectively. More specifically, when the user interface 7 receives data converted into text format data, the knowledge input system 73 analyzes the relationship between words contained in the data in order to understand the sentence structure. After that, the knowledge input system 73 determines the abstraction level of the data based on the sentence structure and word relationships, and stores the data in the data storage unit 3 . Furthermore, the knowledge input system 73 adds blocks of the first graph 31 and elements of the second graph 33 based on sentence structures and word relationships, and associates them with the data stored in the data storage unit 3 . Furthermore, the knowledge input system 73 adds nodes and edges of the knowledge graph 35 according to the added blocks of the first graph 31 and elements of the second graph 33 . The knowledge input system 73 also stores numerical data such as drawings, photographs, and simulation data input via the input/output port 45 in the data storage unit 3 and in the information block 5 as necessary. Blocks of the first graph 31 , elements of the second graph 33 , and nodes corresponding to the knowledge graph 35 can be added. As a result, the data stored in the data storage unit 3 and the graph stored in the information block 5 are changed according to the input from the user, so that information can be accurately provided to the user.

Next, the operation of the design support device 1 according to this embodiment will be described. As shown in FIG. 6, when the user inputs into the microphone 41 a voice saying "I want to increase the load capacity of the side frame", the user interface 7 converts the input voice into a text format query (for example, "I want to give you something to eat") and output it to the conversation engine 9.

The conversation engine 9 uses the dictionary 51 to perform natural language processing on the query converted by the user interface 7, and converts the query into a sentence "I want to increase the load capacity of the side frame." Next, the conversation engine 9 extracts the four keywords "side frame", "load capacity", "lift" and "tai" from the query.

Furthermore, the conversation engine 9 refers to the intent understanding model 52 to extract the intent of the question corresponding to the query. More specifically, the conversation engine 9 refers to the intent understanding model 52 to extract hypothetical text containing the same as the query, and obtains the corresponding question intent “increase load capacity”. After that, the conversation engine 9 converts the four keywords corresponding to the query (“side frame”, “load capacity”, “raise”, and “want”) and the intention of the question (“increase load capacity”) into the knowledge engine. 11.

The knowledge engine 11 extracts the node corresponding to the element containing the most keywords output from the conversation engine 9 from the second graph 33 as the start node. In the example of FIG. 2B, the knowledge engine 11 extracts a node g32 including "side member" and "bearing load" as starting nodes. After that, the knowledge engine 11 refers to the method selection model 61 and acquires the constraint trace corresponding to the intent of the question "increase load capacity" input from the conversation engine 9 as the trace method. Next, the knowledge engine 11 outputs to the expert frame 13 an instruction to execute constraint tracing with node g32 as the starting node.

When the expert frame 13 receives a constraint trace execution instruction with the start node g32, as shown in FIG. 3, the knowledge graph 35 stored in the information block 5 is traced from the node g32 according to the constraint trace method. That is, the expert frame 13 follows the edges of the knowledge graph 35 from the starting node g32, the nodes s26, to which only the arrowheads are connected, and corresponding to constraints associated with data belonging to the first or second hierarchy, and s27 as answer node candidates.

Next, the expert frame 13 uses the weights to extract, from the answer node candidates s26 and s27, the node corresponding to the constraint most closely related to the keyword extracted by the conversation engine 9 as an answer node. More specifically, the expert frame 13 extracts the shortest route R1 (see dashed arrow in FIG. 3) from the starting node g32 to the node s26. After that, the expert frame 13 calculates the average value of the weights (0.8+0.4+0.3)/3=0.50 by taking the sum of the weights of the nodes listed in the root R1 and dividing by the number of nodes. . Next, the expert frame 13 extracts the shortest route R2 from the start node g32 to the node s27 (see the two-dot chain line arrow in FIG. 3), and calculates the average weight of the route R2 (0 . 8+0.4+0.3+0.3+0.8)/5=0.52. The expert frame 13 extracts s27, which has a large average weight value, from among the answer node candidates s26 and S27 as an answer node. It extracts data indicating constraints such as maximum energy absorption and deformation so as not to affect the other vehicle, and outputs the data to the conversation engine 9 . By using the weights in this way, it is possible to easily extract and obtain the constraints most closely related to the keywords extracted by the conversation engine 9 .

After outputting the data associated with the answer node s27 to the conversation engine 9, the expert frame 13 increases the weight of each edge located on the path of the shortest route R1 by 0.01.

As shown in FIG. 6, the knowledge engine 11 further uses a well-known algorithm to perform inference by the inference engine 62 based on the keywords output from the conversation engine 9, and appropriate countermeasures (for example, cross-sectional area increase, etc.) is output to the conversation engine 9.

When the coping method is input from the inference engine 62, the conversation engine 9 converts the coping method input from the inference engine 62 into appropriate text format data (for example, "Would you like to increase the cross-sectional area of the side frame?"). , and output to the user interface 7. The user interface 7 generates sound from the speaker 47 in accordance with the text format data indicating the coping method input from the conversation engine 9, and notifies the user of the coping method.

Furthermore, the conversation engine 9 responds to two inputs from the knowledge engine 11, ``Absorb the most energy in the vehicle during a frontal collision'' and ``Transform so as not to affect the other vehicle''. Two corresponding answers, more specifically, "Make sure that the maximum energy absorption in the car body in the event of a frontal collision" and "Make it deform so as not to affect the other vehicle." construct one answer. After that, a question (for example, "Does the side member absorb the most energy in the car body at the time of a frontal collision?") is constructed to ask the user to identify one of the multiple answers, and the user interface is constructed. 7 to output as voice from the speaker 47 . After that, the conversation engine 9 receives the user's answer to the question (for example, “yes”) from the user interface 7 , identifies the appropriate answer from multiple answers, and outputs it to the user interface 7 . The user interface 7 notifies the user of the answer input from the conversation engine 9 by generating a voice, for example, ``Please change the shape so as not to affect the other vehicle'' from the speaker 47. - 特許庁

When the user vocally inputs "Please display the simulation result of the side frame" through the microphone 41, the conversation engine 9 extracts the three keywords "side frame", "simulation" and "result" from the query. Extract. Also, the conversation engine 9 refers to the intention understanding model 52 and acquires "display of simulation results" as the intention of the question corresponding to the query. The conversation engine 9 outputs three keywords (“sideframe”, “simulation”, and “result”) corresponding to the query and the intention of the question (“display simulation results”) to the knowledge engine 11 .

Knowledge engine 11 selects g32 as the starting node based on the keyword. Next, the knowledge engine 11 refers to the method selection model 61 and acquires the calculation result trace corresponding to the intent of the question "display simulation results" input from the conversation engine 9 as the trace method. After that, the knowledge engine 11 instructs the expert frame 13 to execute a calculation result trace with g32 as the start node. As a result, the expert frame 13 extracts the node g33 associated with the third layer data 25 as an answer node, and the conversation engine 9 causes the monitor 49 of the user interface 7 to display the simulation result. After that, the expert frame 13 increases the weight of the edge connecting g32 and g33 by 0.01.

Next, the effects of the design support device 1 according to this embodiment will be described. In the design and development of new vehicles, execution of simulations and experiments with actual vehicles are indispensable. However, the execution of simulations and experiments with real objects often require time and money, and it is preferable to omit unnecessary experiments and simulations by using data on past knowledge, experience, and failures.

Data such as knowledge, experience, and failures in the existing product design and development process are often stored in two types, a first graph 31 and a second graph 33 . The first graph 31 is mainly created after the end of development, and corresponds to a specification sheet or the like in which a lot of knowledge such as parameters such as dimensions of an existing product is described. On the other hand, the second graph 33 is mainly created at the time of product development, and many descriptions of goals in the design/development process and processes including failures are described. It is desirable that the design support device 1 constructs an answer based on the first graph 31 and the second graph 33 of the existing product for the user.

However, if the data associated with the first graph 31 and the second graph 33 are decomposed into words to form a graph database, relationships between blocks such as elements and constraints, and between elements such as goals and strategies can be obtained. It is difficult to inform the user of the appropriate answer because the relationship is lost.

A data structure X including a first graph 31, a second graph 33, and a knowledge graph 35 stores relationships between blocks and elements. Therefore, by following (tracing) the knowledge graph 35, the design support device 1 can notify the user of the answer according to the relationship between blocks such as elements and constraints and the relationship between elements such as goals and strategies. As a result, it is possible to notify the user of a more appropriate answer based on the knowledge, experience, and failure-related data so far, and it is possible to omit unnecessary experiments and simulations and experiments.

In addition, since the answer is constructed by tracing the knowledge graph 35, the process for obtaining the answer is simpler than when an algorithm such as inference is used.

The knowledge graph 35 is associated with data belonging to four abstract levels of the first to fourth hierarchies. This allows the design support device 1 to construct an answer based on data of a wide range of abstract levels, compared to tracing the knowledge graph 35 associated with one abstract level. This makes it possible to provide more appropriate information to the user.

Also, the knowledge graph 35 is constructed based on the first graph 31 and the second graph 33 . As a result, the design support device 1 provides an answer based on a wider range of data than when the knowledge graph 35 is configured based on only one of the first graph 31 and the second graph 33. can be output. This makes it possible to provide more appropriate information to the user.

Inexperienced design developers often refer to the second graph 33 described by GSN to proceed with the design. On the other hand, experts often refer to the first graph 31 described by SysML and the second graph 33 described by GSN to proceed with the design. By using the knowledge graph 35, the design support device 1 can output answers based on a wider range of data, and is particularly useful for inexperienced design developers.

It is conceivable to use an expert system as the design support device 1 . In an expert system, an expert constructs a knowledge base in the form of "IF...THEN..." and inferences are made based on that data. Therefore, it is not suitable for implementing cases where there is more than one answer, and when there are multiple points to be noted, it is difficult to configure to notify the user of multiple answers.

When multiple pieces of data are associated with an answer node, the computer aided design apparatus 1 receives multiple answers ("Make the side members absorb the largest amount of energy in the car body at the time of a frontal collision", (Answer "Also, configure the vehicle so as not to affect the other vehicle.") and a question that prompts the user to select an appropriate answer. After that, the design support device 1 selects an answer according to the user's answer to the question. In this way, the design support device 1 can construct a plurality of answers. In addition, the design support apparatus 1 asks the user a question that allows the user to select an answer from among a plurality of answers, so that the user can be notified of an appropriate answer more quickly.

If the intent of the query is design change, impact evaluation, or the like, it is preferable to notify the user of answers other than simulation results and actual experiment results. In the design support device 1, if the intent of the query question is "increase load capacity", an answer is constructed based on data indicating constraints associated with the data 23 of the first or second hierarchy. be. As a result, the user is notified of answers other than the results of simulations and the results of experiments using real objects. As described above, in the design support device 1, since the data is classified into a plurality of abstract levels in the data storage unit 3, it is possible to construct an answer based on the data of the appropriate abstract level and notify the user of the answer. can.

In the design support device 1, the user can input a query by voice and receive an answer by voice. In this manner, since communication between the design support device 1 and the user is performed based on voice, the user does not need to use the keyboard 43 to input queries.

Further, the weight given to the edge by the expert frame 13 is updated by voice communication between the design support device 1 and the user, and the data stored in the data storage unit 3 by the knowledge input system 73, the first Graph 31, second graph 33 and knowledge graph 35 are updated. By updating the weights given to the edges, the weights between the nodes positioned on the route traced by the expert frame 13 increase each time the design support device 1 is used. As a result, the relationship between nodes is changed, and the expert frame 13 can more efficiently extract data of high importance to the user. Furthermore, the weights, the data stored in the data storage unit 3, the first graph 31, the second graph 33, and the knowledge graph 35 are updated not only by input from the keyboard 43 but also by voice interaction. , the labor of inputting data using the keyboard 43 can be saved. As a result, the design support device 1 becomes easier for the user to use, and the data from the user can be stored in the data storage unit 3 more quickly and efficiently.

Weights are determined based on expert evaluation. Thereby, the user can be notified of the answer based on the expert's evaluation. When a plurality of nodes are extracted by the expert frame 13 and a plurality of answers are notified with edge strengths, the user can understand the expert's evaluation of the answers based on the edge strengths.

Although the specific embodiments have been described above, the present invention is not limited to the above embodiments and can be widely modified. In the above embodiment, the design support device 1 outputs the answer by tracing the knowledge graph 35 corresponding to one existing product, but it is not limited to this aspect. The design support device 1 may be configured to hold a plurality of knowledge graphs 35, extract an appropriate knowledge graph 35 from the plurality of knowledge graphs 35, and output an answer by tracing the extracted knowledge graph 35. .

In the above embodiment, the first hierarchy in the first graph 31 includes only the parameters of the system to be designed, but the present invention is not limited to this aspect. A first layer of the first graph 31 may include blocks corresponding to the laws of physics describing the properties of the system to be designed and the process of design development.

In the above embodiment, the first graph 31 was described by SysML, and the second graph 33 was described by GSN, but they are not limited to this aspect. Each of the first graph 31 and the second graph 33 may be a graph database, and may be, for example, a graph database written in UML or a graph database obtained by the GQM (Goal Question Metric) method. However, the first graph 31 is preferably a graph database describing parameters and constraints, and the second graph 33 is preferably a graph database describing goals and strategies. Also, in the above embodiment, the second graph 33 includes elements of goal, strategy, and evidence, but may also include context.

In the above embodiment, the dictionary 51 is configured to be included in the software executed by the CPU, but is not limited to this aspect. may refer to the dictionary 51 held in the data storage unit 3 as appropriate.

In the above embodiment, an example was shown in which a plurality of nodes are extracted by the expert frame 13 and a plurality of answers are notified to the user along with edge strengths, but the embodiment is not limited to this. For example, instead of notifying the user of the edge strength itself, the edge strength may be converted into a monotonically increasing numerical value or a ratio judged useful by an expert and then notified to the user.

In the above embodiment, the design support device 1 is configured by one computer, but it is not limited to this aspect. The design support apparatus 1 may be configured by cooperation of a plurality of computers connected by a network such as the Internet. The design support device 1 may be configured by a so-called cloud computing system including a plurality of servers on a network.

In the above embodiment, the design support device 1 is used for vehicle design support, but it is not limited to this aspect. For example, the design support apparatus 1 may be applied to support the design of transportation equipment such as ships and aircraft, and machines.

1: Design support device 3: Data storage unit 5: Information block 7: User interface 9: Conversation engine 11: Knowledge engine 13: Expert frame 21: First layer data 23: Second layer data 25: Third layer data 27 in the fourth layer : data in the fourth layer 31 : first graph 33 : second graph 35 : knowledge graph 37 : general block 39 : restriction block 41 : microphone 43 : keyboard 45 : input/output port 47 : speaker 49: Monitor 51: Dictionary 61: Inference Engine 71: Analysis Platform 73: Knowledge Input System S11-S41: Blocks G11-G42: Elements s11-s41, g11-g42: Nodes e1, e2, e3: Edge

Claims (7)

A design support device,
a data storage unit that classifies and stores design-related data into a plurality of different levels of abstraction;
an information block stored as a graph that defines the relationship of the data;
a user interface that outputs a query from a user in text format;
a conversation engine that performs natural language processing on the query output on the user interface to extract the intent of the question;
a knowledge engine that determines how to trace the graph based on the intent of the question extracted by the conversation engine;
an expert frame for extracting the data corresponding to the answer to the query from the data store by tracing the graph according to the tracing method determined by the knowledge engine;
the conversation engine builds the answer based on the data extracted by the expert frame and outputs the answer to the user interface;
The design support apparatus, wherein the user interface notifies the user of the answer. The data storage unit stores parameters relating to members constituting past design objects, constraints on the members, goals set at the time of development of the design objects or strategies toward the goals,
The graph includes a first graph including blocks associated with the parameter or the constraint, a second graph including elements associated with the goal or the strategy, and the first graph and the second graph. a knowledge graph showing relationships in the graph;
The knowledge graph includes a node corresponding to each of the blocks and the elements, an edge connecting the nodes corresponding to the two blocks related to each other, and the nodes corresponding to the two elements related to each other. and at least one edge connecting said nodes respectively corresponding to said blocks and said elements related to said blocks. The knowledge engine selects one of the elements of the second graph based on the intent of the question extracted by the conversation engine, and uses the node corresponding to the selected element as a starting node of the trace. Output to expert frame,
The expert frame extracts the constraint most closely related to the intent of the question extracted by the conversation engine by tracing the knowledge graph from the start node, and outputs the constraint to the conversation engine. 3. The design support device according to claim 2. The data storage unit stores any one of requirements for how the system to be designed should function, laws of physics describing properties of the system to be designed, and a design development process. a hierarchy, a second hierarchy including the parameters of the members constituting a system designed based on the requirements described in the first hierarchy and the laws of physics, and the functions required of the members; and a computer. Classify the data into a third hierarchy containing the data obtained by simulation and a fourth hierarchy containing the data obtained by experimentation with the real thing and hold the data,
4. The design support apparatus according to claim 3, wherein said expert frame can extract said constraints classified into at least one of said first hierarchy and said second hierarchy. the information block holds, for each edge , a weight indicating the strength of the relationship between the two nodes connected by the edge ;
the expert frame extracts the constraints most relevant to the intent of the question extracted by the conversation engine based on the weights;
5. The design support apparatus of claim 4, wherein the conversation engine constructs the answer to the user based on the extracted constraints. 6. The design of claim 5, wherein when multiple said answers are constructed based on said constraints extracted by said expert frame, said conversation engine further constructs a question for selecting said answer. support equipment. 7. The design support apparatus according to claim 1, further comprising a knowledge input system that updates said data and said graph based on an input from said user interface.

Images