Information processing device, method for creating a design support database, and program

Abstract

This system organizes a large volume of design documents in a wide range of formats as structured data, enabling efficient access to design knowledge. [Solution] The information processing device 100 includes a storage unit 140 that stores a first prompt 141 for extracting design rules from a design document and a second prompt 142 for structuring the design rules extracted from the design document, and a processing unit 130 that extracts design rules by incorporating the design document as context into the first prompt and providing it to the generating AI, and extracts design rule attribute data by incorporating the extracted design rules as context into the second prompt and providing it to the generating AI.

Description

[[Technical Field]]

[0001] The present invention relates to extracting knowledge in the form of rules defined in a prompt from documents related to design deliverables, manufacturing rules, and the like, and to technology for accumulating and utilizing large-scale learning data. [[Background Art]]

[0002] In the design field of the manufacturing industry, most of the information held by companies and organizations is managed in the form of documents. These documents cover a wide range of types, including design specifications, requirement specifications, technical reports, parts lists, and design change instructions, and their contents are often unstructured and voluminous. These documents contain important knowledge that forms the foundation of design, and it is recognized that acquiring necessary information accurately and efficiently is essential for improving the efficiency and quality of design operations.

[0003] However, there are many problems in document management in the design field. Design documents are described in natural language and diagrams, and their expressions are ambiguous, which makes it difficult in some cases to extract and utilize information. Furthermore, along with the increasing complexity of products and the diversification of customization, it is common for relevant information to be spread across multiple documents. As a result, quickly finding necessary information, associating them and organizing them in a usable form requires specialized knowledge and a great deal of man-hours.

[0004] Furthermore, even if necessary information can be extracted from design documents, the amount of such information is huge, and it is not easy to further narrow down the information or access required information therefrom. For example, when a design change is made, in order to grasp how the influence spreads to related specifications and parts, it is necessary to carefully examine documents and check related information one by one. This is a factor that hinders the improvement of the efficiency of the design process and rapid checking of deliverables.

[0005] Patent Document 1 describes a text mining apparatus that performs text mining based on attribute value conditions for positive and negative examples specified by the user, and extracts effective features. Specifically, it describes a text mining apparatus equipped with a data processing device that classifies text into new positive and negative examples based on features selected by the user, and generates attribute value conditions effective for classification (see Claim 1 of Patent Document 1).

[0006] Patent Document 2 describes a method for generating a model of structured data in a database, identifying and comparing candidate reference entities in text data, and calculating similarity. It also describes a method for identifying reference entities and corresponding objects and supplementing text data based on them (see Claim 1 of Patent Document 2). [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2010-61176 [Patent Document 2] Japanese Patent Publication No. 2008-33931 [Overview of the project] [Problems that the invention aims to solve]

[0008] As disclosed in Patent Document 1, the binary conditions of positive and negative examples cannot handle documents that can be classified into multiple categories such as design, manufacturing, inspection, and maintenance, making it difficult to perform appropriate classification. Furthermore, while Patent Document 2 extracts words corresponding to references from documents and adds related information to the documents, actual documents contain a mixture of different formats and technical terms, making it difficult to accurately extract information corresponding to references from text data.

[0009] In recent years, advances in natural language processing (NLP) and machine learning (Machine Learning) technologies have made it possible to automatically extract knowledge from documents. Therefore, the inventors explored using generative AI technology to collect documents related to the design of products that are managed in a distributed manner and involve multiple processes, and to integrally manage the design rules, which constitute the knowledge contained within those documents. These technologies are expected to improve the efficiency of utilizing design information by resolving the ambiguity of natural language and generating structured data from unstructured data. [Means for solving the problem]

[0010] An information processing device according to one embodiment of the present invention is an information processing device that creates a design support database which structures design rules contained in design documents related to the design of a product, and comprises a storage unit which stores a first prompt for extracting design rules from a design document and a second prompt for structuring the design rules extracted from the design document, and a processing unit which extracts design rules by incorporating the design document as context into the first prompt and providing it to a generation AI, and extracts design rule attribute data by incorporating the extracted design rules as context into the second prompt and providing it to the generation AI. [Effects of the Invention]

[0011] Organizing a large volume of design documents in a wide range of formats as structured data enables efficient access to design knowledge. Other challenges and novel features will become apparent from the description and accompanying drawings in this specification. [Brief explanation of the drawing]

[0012] [Figure 1] This is a functional block diagram of the computer 100 in Example 1. [Figure 2] This is a functional block diagram of the computer 100 in Example 1. [Figure 3A] This is an example of a design support system configuration. [Figure 3B] This is an example of the hardware configuration for computer 100. [Figure 4] This is a flowchart showing the processing of computer 100 in Example 1. [Figure 5] This is an example of a category for classifying design rules. [Figure 6] This is an example of the process according to Example 1. [Figure 7] This is a functional block diagram of the computer 100 in Example 2. [Figure 8] This is a flowchart showing the processing performed by computer 100 in Example 2. [Modes for carrying out the invention]

[0013] The embodiments of the present invention will be described below with reference to the drawings. While the drawings show specific embodiments in accordance with the principles of the present invention, they are for the purpose of understanding the present invention and are not intended to be used to restrict its interpretation.

[0014] FIG. 3A shows a design support system of the present embodiment. A computer (information processing apparatus) 100 is connected to a generative AI server 10 and a design document storage 20 via a network 11. The computer 100 extracts design rules by generative AI from documents related to product design (referred to as "design documents") stored in the design document storage 20, and structures the extracted design rules to construct a design support database. Here, the design documents include a wide variety of documents related to product design, such as design specifications, requirement specifications, technical reports, parts lists, and design change instructions. Design documents may include documents related to processes such as product manufacturing and inspection, and documents related to procurement of parts and materials used in products. As such, the scope of design documents broadly covers documents containing information that may be fed back to product design, and is not limited by specific names. There are also various formats and data types for design documents. Design documents are generally managed by respective relevant departments such as design, development, and inspection, and servers, data storage, and the like stored by these respective departments are collectively referred to as the design document storage 20.

[0015] Note that design documents subject to design rule extraction are not limited to documents obtained via the network 11. Design documents read by connecting an external storage device to the computer 100, or design documents obtained by digitizing paper design documents, can also be targeted.

[0016] Further, although FIG. 3A shows a configuration in which the computer 100 performs processing using the external generative AI server 10 via an API, a local LLM (large language model) may be constructed on the computer 100 or on a computer managed by a user to perform processing by generative AI. Since the processing of generative AI is the same in any configuration, the following embodiments will be described focusing on processing in which the computer 100 utilizes generative AI.

[0017] FIG. 3B shows an example hardware configuration of the computer 100. The computer 100 includes a processor (CPU) 21, a memory 22, a storage device 23, an input interface 24, an output interface 25, and a communication interface 26, which are coupled via a bus 27. A GUI (Graphical User Interface) is implemented by an input device that is a keyboard or a pointing device and a display that is an output device, and a user can interactively use the system via the GUI. The input device is connected to the input interface 24, and the output device is connected to the output interface 25. The communication interface 26 is an interface for connecting to a network 11.

[0018] The storage device 23 is normally constituted by an HDD (Hard Disk Drive), an SSD (Solid State Drive), or the like, and stores programs executed by the computer 100, data to be processed by the programs, or data obtained as a result of processing by the programs. The memory 22 is constituted by a RAM (Random Access Memory), and temporarily stores programs, data necessary for executing the programs, and the like in accordance with instructions from the processor 21. The processor 21 functions as a functional unit (functional block) that provides predetermined functions by executing a program loaded from the storage device 23 into the memory 22.

[0019] Note that the system does not need to be implemented by a single computer, and may be implemented by a plurality of computers. Furthermore, some or all of the functions of the design support system may be implemented as an application on the cloud. [Example]

[0020] Figure 1 is a functional block diagram of the computer 100 in Embodiment 1. The input unit 110 inputs the design document into the computer 100. The processing unit 130 includes an extraction unit 131 and a classification unit 132. The extraction unit 131 provides the generation AI with extraction prompts stored in the storage unit 140 to extract design rules from the input design document. The classification unit 132 provides the generation AI with classification prompts stored in the storage unit 140 to structure the extracted design rules into predetermined specifications. The output unit 120 outputs the design rule attribute data extracted from the design document by the processing unit 130.

[0021] Furthermore, as shown in Figure 2, it is also possible to configure the extraction and classification unit 133 as a functional unit that executes the processing of the extraction unit 131 and the processing of the classification unit 132 as a series of processes. In this case, the memory unit 140 stores a set of prompts combined into one as an extraction and classification prompt, and the extraction and classification unit 133 extracts design rule attribute data from the input design document by providing the extraction and classification prompt 143 to the generation AI 150.

[0022] Figure 4 shows the processing flow of the computer 100. First, the input unit 110 receives a design document (S01). The extraction unit 131 takes the received design document as context into the extraction prompt 141, and provides the extraction prompt 141 with the design document to the generation AI 150 to extract design rules from the design document (S02). If the generation AI 150 determines through analysis that the design document does not contain design rules (No in step S03), the processing flow ends. If the generation AI 150 extracts design rules from the design document through analysis (Yes in step S03), the classification unit 132 takes the extracted design rules as context into the classification prompt 142, and provides the classification prompt 142 with the design rules to the generation AI 150 to structure the design rules (S04). As a result, the design rules contained in the design document are systematically organized, and for example, the structured design rules can be stored in a database and centrally managed to improve their usability. The output unit 120 outputs structured design rules in a format usable by the user (S05). The output format is arbitrary and includes registration in a database, display on a screen, etc. The output design rules can be used for review and approval of design deliverables, risk prediction in the manufacturing process, etc.

[0023] Figure 5 shows an example of the items used by the classification unit 132 to classify design rules in order to structure them. Here, an example is shown of assigning attribute information to the following items: rule type 151, shape feature 152, part 153, measurement target 154, and threshold 155. Rule type 151 indicates which process the design rule relates to, such as design, manufacturing, or inspection. Shape feature 152 indicates which shape information of the product the design rule is based on, such as material information or wall thickness. Part 153 indicates which part the design rule relates to. Measurement target 154 indicates what the design rule is judged for. Threshold 155 indicates the threshold set in the design rule. Each of these items is automatically acquired and identified from the design rule by the generating AI. By extracting and structuring attribute information for predetermined items from the extracted design rules in this way, it becomes possible to easily use this for, for example, an automated check to see if a CAD model conforms to the design rules.

[0024] Figure 6 shows an example of processing according to Example 1. Assume that the design document, design guideline 161, states, "In sheet metal processing, if the distance between the hole and the bend is too close, forming defects may occur. Therefore, ensure that the distance between the center of the hole and the bend line is 3 mm or more." The extraction unit 131 extracts the design rule 162, "The distance between the hole and the bend should be 3 mm or more," from the design guideline 161. Furthermore, the classification unit 132 extracts design rule attribute data for predetermined items from the design rule 162. Here, design rule attribute data 163 is output with rule type "design", measurement target "distance between hole and bend", and threshold "3 mm or more". In this way, by extracting design rules from design documents and classifying them by predetermined items, unstructured documents can be processed into structured attribute data, making it possible to quickly access the know-how described in the design documents. [Examples]

[0025] Figure 7 is a functional block diagram of the computer 100 in Example 2. The functional blocks added to the configuration example of Example 1 will be explained in detail. The acquisition unit 211 automatically crawls a folder on a specific server (corresponding to the design document storage 20 in Figure 3A) and accepts the design document when it detects, for example, the addition of a new design document. The input analysis unit 212 analyzes the accepted design document and selects an appropriate extraction prompt from the prompt database 221 for rule extraction. The splitting unit 213 splits the design document into page units if the design document spans multiple pages. This function enables input of multiple documents with multiple pages and allows for automatic continuous rule extraction. The aggregation unit 214 links source information such as the document name, page, and file path of the storage location of the design document with the extracted design rules. The linked source information is added as one item of the design rule attribute data extracted by the classification unit 132 and stored in the design support database 222.

[0026] The search unit 215 searches the design rule attribute data stored in the design support database 222 using a word of the user's choosing, and displays the corresponding design rules on the display. The checklist creation unit 216 automatically creates a checklist listing the rules found by the search unit 215.

[0027] Figure 8 shows the processing flow of computer 100. First, the acquisition unit 211 crawls a folder (design document storage) on a specific server, acquires the target design document, and accepts it (S11). Next, the input analysis unit 212 selects an appropriate prompt from the prompt database 221 for the accepted design document. For example, depending on whether the accepted design document is in Excel format or PDF format, the prompt database 221 contains multiple prompts that perform the same processing. An appropriate prompt for processing the design document is selected from among them and used in the processing using the generation AI in steps S14 and S16. If no appropriate prompt exists in the prompt database 221 (No in step S12), the processing flow ends. The splitting unit 213 divides the entire accepted design document into pages (S13), and subsequent processing is performed individually for each page. Note that page division is a simple example, and in some design documents, the content may be connected across multiple pages. In such cases, the splitting unit 213 may analyze the structure of the design document and adjust the range of division based on the unit of content. In that case as well, subsequent processing is carried out individually for each design document divided by the division unit 213.

[0028] The processing unit 130 takes the design document (page by page) as context and inputs it to the extraction prompt selected by the input analysis unit 212, provides the extraction prompt containing the design document (page by page) to the generation AI 150, and extracts design rules from the design document (page by page) (S14). The aggregation unit 214 associates source information with the extracted design rules (S15). This makes it easy to identify the source of the extracted rules and improves the reliability of the design support database. Next, the processing unit 130 takes the extracted design rules as context and inputs them to the classification prompt selected by the input analysis unit 212, provides the classification prompt containing the design rules to the generation AI 150, and structures the design rules (S16). The aggregation unit 214 adds the source information as an item in the design rule attribute data obtained in step S16 and stores it in the design support database 222. If the design document is divided into multiple pages, the processing in steps S14 to S17 is performed for each page.

[0029] The subsequent processing involves utilizing the design support database 222. The search unit 215 takes a search query using any word and performs the search (S18). The design support database 222 stores attribute data of setting rules, including design rules and source information, linked together, allowing for quick retrieval of specific design rules or information. The checklist creation unit 216 creates a checklist by listing the design rules extracted by the search unit 215 (S19). The checklist can be used for subsequent work and verification.

[0030] Steps S18 and S19 can be performed independently of the creation of the design support database, but by setting a specific search key in step S18 in advance and automatically executing steps S18 and S19 each time the contents of the design support database 222 are updated, the checklist can be kept up to date at all times.

[0031] The present invention is not limited to the embodiments described above, and includes various modifications. For example, the embodiments and modifications described above are explained in detail to make the present invention easier to understand, and are not necessarily limited to those having all the configurations described. Furthermore, it is possible to replace parts of the configuration of one embodiment or modification with the configuration of another embodiment or modification, and it is also possible to add the configuration of another embodiment or modification to the configuration of one embodiment or modification. In addition, it is possible to add, delete, or replace parts of the configuration of each embodiment or modification with other configurations. [Explanation of symbols]

[0032] 10: Generation AI server, 11: Network, 20: Design document storage, 21: Processor, 22: Memory, 23: Storage device, 24: Input interface, 25: Output interface, 26: Communication interface, 27: Bus, 100: Computer, 110: Input unit, 120: Output unit, 130: Processing unit, 131: Extraction unit, 132: Classification unit, 133: Extraction and classification unit, 140: Memory unit, 141: Extraction prompt, 142: Minutes Classification prompts: 143: Prompts for extraction and classification, 151: Rule type, 152: Shape features, 153: Parts, 154: Measured object, 155: Threshold, 161: Design guidelines, 162: Design rules, 163: Design rule attribute data, 211: Acquisition unit, 212: Input analysis unit, 213: Splitting unit, 214: Aggregation unit, 215: Search unit, 216: Checklist creation unit, 221: Prompt database, 222: Design support database.

Claims

1. An information processing device that creates a design support database that structures design rules included in design documents related to product design, A storage unit that stores a first prompt for extracting design rules from the design document and a second prompt for structuring the design rules extracted from the design document, An information processing device having a processing unit that extracts design rules by incorporating the design document as context into the first prompt and providing it to the generating AI, and extracts design rule attribute data by incorporating the extracted design rules as context into the second prompt and providing it to the generating AI.

2. In claim 1, An information processing device in which the first prompt and the second prompt are recorded as a single prompt in the storage unit.

3. In claim 1, Connected via the network to design document storage where design documents are stored, An information processing device further comprising an acquisition unit that crawls the aforementioned design document storage and, when it detects the addition of a new design document, accepts it as a design document to be processed by the processing unit.

4. In claim 3, The storage unit stores multiple instances of the first prompt and the second prompt, An information processing device further comprising an input analysis unit that selects one of the first and second prompts to be used by the processing unit from among a plurality of the first and second prompts stored in the storage unit, in accordance with the design document received by the acquisition unit.

5. In claim 3, A division unit that divides the design document received by the acquisition unit, An information processing device further comprising: an aggregation unit that processes the divided design documents as units and adds source information of the divided design documents to the design rule attribute data extracted by the processing unit.

6. In claim 5, An information processing device in which the design rules extracted by the processing unit and the design rule attribute data extracted from the design rules and to which the source information has been added are stored in the design support database.

7. In claim 6, An information processing device further comprising a search unit that searches the design support database using any word.

8. A method for creating a design support database, which uses a computer to create a design support database that structures design rules included in design documents related to product design, The computer is pre-stored a first prompt for extracting design rules from the design document and a second prompt for structuring the design rules extracted from the design document. The extraction process involves incorporating the design document as context into the first prompt and providing it to the generating AI to extract design rules, A method for creating a design support database, which involves a classification process that extracts design rule attribute data by incorporating the design rules extracted by the extraction process as context into the second prompt and providing it to the generating AI.

9. In claim 8, The computer is connected via a network to a design document storage where design documents are stored. A method for creating a design support database that crawls the aforementioned design document storage and, when it detects the addition of a new design document, performs an acquisition process to accept it as a design document to be extracted.

10. In claim 9, The above acquisition process involves a splitting process that divides the design documents received, A method for creating a design support database, comprising: performing the extraction process and the classification process on the divided design documents as units; and performing an aggregation process that adds source information of the divided design documents to the design rule attribute data extracted by the classification process.

11. In claim 10, A method for creating a design support database, wherein the design rules extracted by the extraction process and the design rule attribute data extracted from the design rules and to which the source information has been added are stored in the design support database.

12. A program that causes a computer to create a design support database that structures the design rules contained in design documents related to the design of a product, The computer has in advance stored a first prompt for extracting design rules from the design document and a second prompt for structuring the design rules extracted from the design document. The first prompt incorporates the design document as context and provides it to the generating AI to extract design rules; A program for implementing a classification function that extracts design rule attribute data by incorporating the design rules extracted by the extraction function as context into the second prompt and providing it to the generating AI.

13. In claim 12, The computer is connected via a network to a design document storage where design documents are stored. A program for implementing an acquisition function that crawls the aforementioned design document storage and, when it detects the addition of a new design document, accepts it as a design document to be processed by the extraction function.

14. In claim 13, A splitting function for splitting the design documents received by the aforementioned acquisition function, A program for implementing the extraction function and the classification function using the divided design documents as units, and an aggregation function for adding source information of the divided design documents to the design rule attribute data extracted by the classification function.

15. In claim 14, A program that stores the design rules extracted by the extraction function and the design rule attribute data extracted from the design rules and to which the source information has been added, in the design support database.

Images