Device fault natural language question answering method and system based on cue word optimization

Abstract

The invention discloses an equipment fault natural language question and answer method and system based on cue word optimization. The method comprises the steps of obtaining an equipment fault problem which is input by a user and described by a natural language; performing natural language understanding and semantic representation on the question, and identifying core words and dimensions; generating an optimal query prompt word based on the core word and the dimension; combining the optimal cue word with a pre-constructed equipment fault knowledge base, and constructing query and retrieving related knowledge; generating a natural language reply based on the retrieval knowledge and presenting the natural language reply to the user; obtaining user feedback information; and based on the feedback information, optimizing the cue word generation strategy and updating the knowledge base to form a closed-loop learning process. The system further comprises innovation modules such as cross-modal fusion and understanding, intelligent labeling and knowledge extension. According to the method, innovative mechanisms such as cue word optimization generation, knowledge base closed-loop optimization, multi-source heterogeneous data fusion utilization and low-cost knowledge extension are provided, and highly-intelligent fault diagnosis and guidance are realized.

Description

Technical Field

[0001] The present invention relates to the field of equipment fault diagnosis and maintenance, and more specifically to a natural language question-answering method and system for equipment fault based on prompt word optimization. Background Art

[0002] Industrial equipment will inevitably encounter various faults during operation. Timely and accurate diagnosis of the cause of the fault and providing effective maintenance guidance are crucial to ensure stable operation of the equipment and reduce maintenance costs. Traditional fault diagnosis and maintenance methods mostly rely on manual operation, which is not only inefficient but also prone to misjudgment, making it difficult to meet the requirements of modern industrial production for the efficiency and accuracy of equipment operation and maintenance.

[0003] In order to improve the intelligent level of fault diagnosis and maintenance, some attempts based on natural language processing and knowledge base technology have emerged in recent years. Users can describe the fault problems they encounter in natural language, and the system retrieves the knowledge base and gives solutions based on the description. However, the existing technology has obvious shortcomings in the following aspects:

[0004] 1) The constructed knowledge base is usually static, difficult to update and expand in time, and has a limited scope of application;

[0005] 2) Lack of effective semantic understanding ability, unable to accurately identify the core content of the problem, and low pertinence and accuracy of knowledge retrieval;

[0006] 3) It only supports a single modality (such as text) and cannot fully utilize multi-source heterogeneous data (such as images, videos, etc.), which limits its understanding and guidance capabilities;

[0007] 4) A large amount of manually annotated data is required as the knowledge source of the knowledge base, and the cost of knowledge expansion is high.

[0008] In view of the above-mentioned defects of the prior art, the present invention proposes an innovative natural language question-answering method for equipment faults based on prompt word optimization, aiming to achieve highly intelligent fault diagnosis and guidance. Summary of the invention

[0009] The purpose of the present invention is to provide a natural language question-answering method and system for equipment faults based on prompt word optimization, aiming to overcome the defects existing in the prior art and realize intelligent fault diagnosis and guidance.

[0010] On the one hand, a natural language question-answering method for equipment failure based on prompt word optimization is provided, comprising the following steps:

[0011] S1, obtaining a device failure problem described in natural language input by a user.

[0012] S2, performing natural language understanding and semantic representation on the equipment failure problem, and identifying the core words and dimensions of the failure problem.

[0013] S3, based on the core words and dimensions, generating optimal query prompt words for the fault problem.

[0014] S4, combining the optimal query prompt word with a pre-built equipment failure knowledge base to construct a structured or semantic query, and accurately retrieving knowledge related to the failure problem from the equipment failure knowledge base.

[0015] S5, based on the relevant knowledge retrieved from the equipment fault knowledge base, generate a corresponding natural language response and present it to the user, wherein the natural language response includes an explanation, analysis and maintenance guidance of the fault problem.

[0016] S6, obtaining user feedback information regarding the presented natural language response.

[0017] S7, based on the feedback information, optimizing the generation strategy of the optimal query prompt word and updating the equipment fault knowledge base to form a closed-loop learning process.

[0018] Furthermore, the step of performing natural language understanding and semantic representation on the equipment failure problem in S2 includes: using named entity recognition and dependency semantic analysis technology to analyze the equipment failure problem, and identify equipment component words, failure phenomenon words and possible cause words in the equipment failure problem.

[0019] Furthermore, the step of generating the optimal query prompt word in S3 includes: using a sequence-to-sequence model based on an attention mechanism to encode the equipment failure problem and generate the optimal query prompt word, and the sequence-to-sequence model is trained on a large amount of annotated corpus.

[0020] Furthermore, the equipment failure knowledge base in the method of the present invention is constructed through the following steps: automatically collecting structured data and unstructured data from channels such as equipment manuals, maintenance records, forums, etc.; pre-processing the collected data such as cleaning, formatting, classification and labeling; based on knowledge graph technology, constructing the pre-processed data into a knowledge base representing the principles of equipment failures and maintenance steps.

[0021] Furthermore, the step of generating a natural language response in S5 includes: inputting relevant knowledge retrieved from the equipment fault knowledge base into a trained knowledge distillation model, which generates a corresponding natural language response; the natural language response is presented in text form, or in other modal forms such as fused images and videos.

[0022] Furthermore, the steps of obtaining feedback information and optimizing in S6 and S7 include: obtaining user feedback on the presented natural language response, the feedback including a satisfaction score and supplementary explanation for the response; inputting the satisfaction score and supplementary explanation into a reinforcement learning-based model, which outputs adjustments to the optimal query prompt word generation strategy and updates the equipment failure knowledge base.

[0023] Furthermore, the method of the present invention also includes: based on an active learning strategy, automatically identifying key corpus samples that require manual annotation; using a knowledge distillation network to learn and expand knowledge from a small amount of manually annotated high-quality data; and adding new knowledge learned from the knowledge expansion network to the equipment failure knowledge base.

[0024] Furthermore, the method of the present invention also includes: cross-modal fusion of the text form of the equipment failure problem and the related image and video data to generate a multimodal fusion feature representation; based on the multimodal fusion feature representation, the semantic and visual clues are integrated to comprehensively understand and analyze the failure problem.

[0025] Furthermore, the equipment fault knowledge base includes the following data: equipment parameter data, equipment fault code data, fault principle and repair instruction data in text form, and fault display and repair guidance data in image or video form.

[0026] On the other hand, a system for implementing the above method based on a computer is also provided, comprising:

[0027] Data acquisition module, used to automatically collect equipment-related data;

[0028] A knowledge base construction module is used to construct a device fault knowledge base based on the collected data;

[0029] The semantic understanding module is used to understand and semantically represent the natural language fault problems input by users;

[0030] A prompt word optimization module, used to generate the optimal query prompt word for the fault problem;

[0031] A knowledge retrieval module, used for retrieving relevant knowledge from the knowledge base according to the optimal query prompt word;

[0032] a response generation module for generating a natural language response based on the retrieved relevant knowledge;

[0033] The feedback optimization module is used to optimize the prompt word generation strategy and update the knowledge base according to user feedback.

[0034] The technical solution provided by the embodiment of the present invention has the following beneficial effects:

[0035] 1. Improved the intelligence level of fault diagnosis and maintenance

[0036] The method of the present invention uses innovative technologies such as prompt word optimization mechanism and cross-modal fusion understanding to more accurately search the knowledge base and understand the fault problem more comprehensively, thereby generating high-quality natural language responses, providing users with accurate fault diagnosis results and maintenance guidance, and greatly improving the level of intelligent fault handling.

[0037] 2. Achieved closed-loop learning and self-iteration of fault knowledge

[0038] The present invention designs a feedback optimization mechanism based on reinforcement learning, which can continuously optimize the prompt word generation strategy and update the knowledge base content according to user feedback, forming a self-iterative closed-loop learning process, and continuously strengthening the system's fault knowledge and question-answering capabilities.

[0039] 3. Reduce the cost of knowledge base construction and expansion

[0040] The present invention introduces an intelligent labeling and knowledge expansion module, and uses technologies such as active learning and knowledge distillation to efficiently expand a large amount of new knowledge from a small number of manually labeled samples, reducing the manpower and time costs of knowledge base construction and expansion.

[0041] 4. Support the integration and utilization of multi-source heterogeneous data

[0042] The knowledge base of the present invention integrates multiple heterogeneous data such as structured data, unstructured text data, images and videos, and comprehensively utilizes these data through a cross-modal fusion and understanding module to make fault diagnosis and guidance more accurate and comprehensive.

[0043] 5. Improve equipment operation and maintenance efficiency and reduce operating costs

[0044] The present invention can automatically and efficiently complete fault diagnosis and maintenance guidance work, greatly improving the efficiency of equipment operation and maintenance, while reducing the cost expenditure of manual diagnosis and maintenance, and reducing the operating costs of the enterprise.

[0045] In summary, the natural language question-answering method and system for equipment failure based on prompt word optimization of the present invention are not only innovative in theory, but also have important economic value and social significance in application. They can greatly promote the intelligent upgrading of industrial production and are worthy of wide promotion and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 It is a flowchart of a natural language question-answering method for equipment failure based on prompt word optimization according to an embodiment of the present invention;

[0048] Figure 2 It is a schematic diagram of the module architecture of a natural language question-answering system for equipment failure based on prompt word optimization in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0050] See also Figure 1 , which shows the process of a natural language question-answering method for equipment failure based on prompt word optimization of the present invention. The method mainly includes the following steps:

[0051] S1, obtaining a device failure problem described in natural language input by a user.

[0052] Specifically, users can describe the device failure they encounter in natural language through a variety of methods such as mobile phone apps, web interfaces or voice input.

[0053] S2, performs natural language understanding and semantic representation of equipment failure problems, and identifies the core words and dimensions of the failure problems.

[0054] This embodiment uses natural language processing technologies such as named entity recognition and dependency semantic analysis to analyze the natural language questions input by users, and identifies equipment component words (such as "spindle", "cooling pipeline", etc.), fault phenomenon words (such as "vibration", "overtemperature", etc.) and possible causes of faults (such as "bearing wear", "blockage", etc.). In this way, the core content of the problem can be accurately grasped.

[0055] S3, based on core words and dimensions, generates the optimal query prompt words for fault problems.

[0056] This embodiment uses a sequence-to-sequence model such as Transformer based on the attention mechanism to encode the core words and dimension information identified in S2 into vectors, and then generates the optimal query prompt words after inputting them into the model. The model is trained on a large amount of annotated corpus and can generate highly relevant query prompt words to guide subsequent more accurate retrieval of the knowledge base.

[0057] S4, structure or semantically query the optimal query prompt word with the pre-built equipment fault knowledge base to accurately retrieve knowledge related to the fault problem from the knowledge base.

[0058] This embodiment first constructs a device failure knowledge base in the following manner:

[0059] 1) Automatically collect structured data (such as equipment parameters, fault codes, etc.) and unstructured data (such as text, images, videos, etc.) from channels such as equipment manuals, maintenance records, and forums;

[0060] 2) Preprocess the collected data by cleaning, formatting, classifying and labeling;

[0061] 3) Based on knowledge graph technology, the preprocessed data is constructed into a knowledge base representing the equipment failure principles and maintenance steps.

[0062] Then, the optimal query prompt words generated by S3 are combined with the knowledge base, and the knowledge related to the user's fault problem is accurately retrieved based on the query language or semantic query method.

[0063] S5, based on the relevant knowledge retrieved from the knowledge base, generates a natural language response and presents it to the user. The response includes the explanation, analysis and maintenance guidance of the fault problem.

[0064] In this embodiment, the relevant knowledge retrieved from the knowledge base is input into a trained knowledge distillation model, which generates a corresponding natural language response. The response can be in plain text form, or it can be integrated with other modal data such as images and videos to form a multi-modal presentation.

[0065] S6, obtaining user feedback information on the presented natural language response.

[0066] S7, based on the feedback information, optimizes the generation strategy of the optimal query prompt word and updates the equipment fault knowledge base to form a closed-loop learning process.

[0067] This embodiment obtains user feedback on the natural language reply generated by S5, which can be a satisfaction score for the reply or a supplementary explanation of the reply content. Then, the score and supplementary explanation are input into the reinforcement learning-based model, which outputs the adjustment of the optimal query prompt word generation strategy and updates the knowledge base according to the supplementary explanation to make up for the deficiencies of the knowledge base. In this way, a closed-loop learning process is formed, and the system can continuously optimize the prompt word strategy and enrich the knowledge base content, and continuously improve the accuracy of the reply.

[0068] On the basis of the above basic process, the embodiment of the present invention also includes the following innovative technologies:

[0069] 1) Intelligent labeling and knowledge expansion module

[0070] Based on active learning strategies, key corpus samples that need manual annotation are automatically identified;

[0071] Use knowledge distillation networks to learn and expand knowledge from a small amount of high-quality manually annotated data;

[0072] The new knowledge learned from the knowledge expansion network is added to the equipment failure knowledge base to expand the knowledge scope.

[0073] This greatly reduces the manual annotation cost of knowledge expansion.

[0074] 2) Cross-modal fusion and understanding module

[0075] Perform multimodal fusion on the text form of the equipment failure problem and the related image and video data to generate a fused feature representation;

[0076] Based on fused feature representation, semantic and visual clues are integrated to fully understand and analyze fault problems.

[0077] This can fully utilize multi-source heterogeneous data and improve the comprehensiveness of fault understanding.

[0078] See also Figure 2 , which shows the module architecture of a natural language question-answering system for equipment failure based on prompt word optimization of the present invention, which is used to implement the above method. The system includes:

[0079] Data acquisition module, used to automatically collect equipment-related data;

[0080] A knowledge base construction module is used to construct a device fault knowledge base based on the collected data;

[0081] The semantic understanding module is used to understand and semantically represent the natural language fault problems input by users;

[0082] The prompt word optimization module is used to generate the optimal query prompt word for the fault problem;

[0083] A knowledge retrieval module is used to retrieve relevant knowledge from the knowledge base according to the optimal query prompt word;

[0084] a response generation module for generating a natural language response based on the retrieved relevant knowledge;

[0085] Feedback optimization module, used to optimize prompt word generation strategy and update knowledge base according to user feedback;

[0086] Intelligent annotation and knowledge expansion module, which is used to expand new knowledge from a small amount of annotated data and add it to the knowledge base based on active learning and knowledge distillation;

[0087] The cross-modal fusion and understanding module is used to fuse and comprehensively understand multimodal data such as text, images, and videos to provide support for response generation.

[0088] The above system modules work together through efficient data interaction and model calling processes to achieve intelligent fault diagnosis and maintenance guidance.

[0089] In summary, the device fault natural language question-answering method based on prompt word optimization proposed in the present invention has the following innovations compared with the prior art:

[0090] 1) A new prompt word optimization generation mechanism is proposed to generate highly relevant prompt words based on core words and dimensions to guide knowledge retrieval, greatly improving the accuracy and efficiency of retrieval.

[0091] 2) A closed-loop optimization process for the knowledge base based on reinforcement learning was designed, which can continuously optimize the prompt word strategy and update the knowledge base content according to user feedback, form self-iterative learning, and continuously enhance system capabilities.

[0092] 3) Innovatively integrates multi-source heterogeneous data from multiple channels, including structured data, unstructured text data, images and videos, and comprehensively utilizes these data through cross-modal fusion technology to make fault understanding and response generation more accurate and comprehensive.

[0093] 4) The introduction of strategies such as active learning and knowledge distillation has greatly reduced the cost of manual annotation for knowledge base expansion, enabling the system to have the ability of continuous learning and continuously expanding its scope of application.

[0094] Through the above innovative technology, the present invention can significantly improve the intelligence level of equipment fault diagnosis and maintenance, improve work efficiency, and reduce maintenance costs. It has important theoretical value and broad application prospects.

[0095] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A natural language question-answering method for equipment failure based on prompt word optimization, It is characterized in that include: Obtain the device failure problem described in natural language input by the user; Perform natural language understanding and semantic representation on the equipment failure problem to identify the core words and dimensions of the failure problem; Based on the core words and dimensions, generate optimal query prompt words for the fault problem; Combining the optimal query prompt word with a pre-built equipment failure knowledge base to construct a structured or semantic query, and accurately retrieving knowledge related to the failure problem from the equipment failure knowledge base; Based on the relevant knowledge retrieved from the equipment fault knowledge base, a corresponding natural language response is generated and presented to the user, wherein the natural language response includes an explanation, analysis and maintenance guidance of the fault problem; Obtaining user feedback on the presented natural language responses; Based on the feedback information, the generation strategy of the optimal query prompt word is optimized and the equipment fault knowledge base is updated to form a closed-loop learning process.

2. The method according to claim 1, It is characterized in that The steps of performing natural language understanding and semantic representation on the equipment failure problem include: The equipment failure problem is analyzed by using named entity recognition and dependency semantic analysis technology to identify equipment component words, failure phenomenon words and possible cause words in the equipment failure problem.

3. The method according to claim 1, It is characterized in that The step of generating the optimal query prompt word for the fault problem comprises: The equipment failure problem is encoded by using a sequence-to-sequence model based on an attention mechanism to generate the optimal query prompt word, and the sequence-to-sequence model is trained on a large amount of annotated corpus.

4. The method according to claim 1, It is characterized in that The equipment failure knowledge base is constructed by the following steps: Automatically collect structured and unstructured data from equipment manuals, maintenance records, and forum channels; Pre-process the collected data by cleaning, formatting, classifying and labeling; Based on the knowledge graph technology, the preprocessed data is constructed into a knowledge base representing the equipment failure principles and maintenance steps.

5. The method according to claim 1, It is characterized in that The step of generating a natural language response comprises: Inputting relevant knowledge retrieved from the equipment failure knowledge base into a trained knowledge distillation model, the knowledge distillation model generating a corresponding natural language response; The natural language response is presented in the form of text, or in the form of fused images or videos.

6. The method according to claim 1, It is characterized in that The steps of obtaining feedback information and optimizing include: Obtaining user feedback on the presented natural language answer, the feedback including a satisfaction rating and additional explanations for the answer; The satisfaction score and supplementary description are input into a reinforcement learning-based model, and the model outputs adjustments to the optimal query prompt word generation strategy and updates the equipment failure knowledge base.

7. The method according to claim 1, It is characterized in that Also includes: Based on active learning strategies, key corpus samples that need manual annotation are automatically identified; Use knowledge distillation networks to learn and expand knowledge from a small amount of high-quality manually annotated data; The new knowledge learned from the knowledge expansion network is added to the equipment fault knowledge base.

8. The method according to claim 1, It is characterized in that Also includes: Perform cross-modal fusion on the text form of the equipment failure problem and related image and video data to generate a multi-modal fused feature representation; Based on the multimodal fusion feature representation, the fault problem is fully understood and analyzed by integrating semantic and visual clues.

9. The method according to claim 1, It is characterized in that The equipment failure knowledge base includes the following data: Equipment parameter data, equipment fault code data, fault principle and repair instruction data in text form, fault display and repair guidance data in image or video form.

10. A computer-based system for implementing the method according to any one of claims 1 to 9, comprising: Data acquisition module, used to automatically collect equipment-related data; A knowledge base construction module is used to construct a device fault knowledge base based on the collected data; The semantic understanding module is used to understand and semantically represent the natural language fault problems input by users; A prompt word optimization module, used to generate the optimal query prompt word for the fault problem; A knowledge retrieval module, used for retrieving relevant knowledge from the knowledge base according to the optimal query prompt word; a response generation module for generating a natural language response based on the retrieved relevant knowledge; The feedback optimization module is used to optimize the prompt word generation strategy and update the knowledge base according to user feedback.

Images