Expert online question-answering system and method for industrial equipment fault diagnosis based on vector matching

Abstract

The application discloses an online question and answer system and method for industrial equipment fault diagnosis based on vector matching, wherein a user inputs a problem description on a client, the system converts the problem description into a vector representation locally or remotely, matches the vector representation with stored standard problem descriptions, determines a problem item closest to the problem description, and recommends an expert ID skilled in processing the closest problem item; the user sends a remote communication request to the recommended expert ID on the client; and the expert accepts the remote communication request on the client and establishes remote online communication based on websocket and webRTC technology. The application aims to provide efficient, accurate and convenient fault diagnosis services through intelligent expert matching mechanism, flexible communication mode and perfect record management, so as to improve the overall performance and user experience of industrial equipment fault diagnosis.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of artificial intelligence, and particularly relates to an online question and answer system and method for industrial equipment fault diagnosis based on vector matching. BACKGROUND

[0002] The technologies used in industrial equipment fault diagnosis include traditional expert systems, rule-based diagnosis methods, and machine learning and artificial intelligence technologies. Traditional expert systems mainly rely on the experience and knowledge base of domain experts, and conduct fault diagnosis through pre-set rules and logic. The basic principle is to match the fault phenomenon with possible fault causes, and then provide corresponding diagnosis suggestions according to the matching results. The limitation of this method is that it depends on the experience level of experts, and it is difficult to handle complex and newly emerging fault patterns, and the diagnosis accuracy and efficiency are limited.

[0003] Rule-based diagnosis methods usually use fixed diagnosis rules and logic to analyze equipment faults. These rules are usually pre-defined according to the operating characteristics and fault patterns of the equipment, and the system matches the equipment state parameters to find the fault causes that meet the conditions. For example, a series of fault conditions and corresponding handling measures can be pre-set, and when the equipment state parameters meet a certain fault condition, the system will automatically trigger the corresponding handling measures. Rule-based diagnosis methods also have certain limitations, and the construction and maintenance of the rule base require a lot of manual intervention, and it is difficult to respond to dynamic changes in equipment operating environment and fault patterns.

[0004] With the development of artificial intelligence and big data technologies, machine learning-based fault diagnosis methods have gradually become a research hotspot. This method trains a large number of historical fault data to build a fault diagnosis model, achieving automatic identification and prediction of equipment faults. Machine learning algorithms can handle complex nonlinear relationships and have strong generalization ability, and can maintain high diagnosis accuracy when the equipment operating environment and fault patterns change. For example, neural networks and support vector machines are widely used in equipment fault diagnosis, which trains the equipment operating data to build a fault diagnosis model, thereby achieving efficient and accurate fault diagnosis, but requires enough reliable fault data or reports to be sorted out.

[0005] In the prior art, some remote diagnosis systems have been applied to the field of industrial equipment fault management. These systems usually have basic fault reporting, expert matching and online communication functions, but have limitations in intelligent matching mechanisms, real-time communication technologies, diversified communication methods, record management and feedback mechanisms. For example, in terms of communication, general systems support text and picture communication between users and experts, and some systems may support audio and video calls, but the communication technology of these systems may be based on traditional HTTP polling or SIP protocol, which has the problem of high delay and poor real-time performance.

[0006] These problems can be summarized as follows:

[0007] 1. Low diagnostic efficiency: Traditional fault diagnosis methods usually require on-site experts to analyze and judge, which is time-consuming and inefficient. Even if remote diagnosis technology is used, due to the low matching efficiency of experts, it is difficult to find suitable experts to solve problems in time, affecting the speed of diagnosis.

[0008] 2. Low accuracy: Existing remote diagnosis systems often rely on fixed rules or simple keyword matching when matching experts, making it difficult to accurately understand user-described problems, resulting in the matched experts may not have the most appropriate expertise, affecting the accuracy and effectiveness of diagnosis.

[0009] 3. Poor flexibility: Existing remote diagnosis systems lack flexibility in communication methods, usually only supporting text or voice communication, which cannot meet the diverse needs of users in different situations. In addition, the system is rough in managing communication records, and cannot achieve efficient historical query and management.

[0010] 4. Poor user experience: In existing technical solutions, users cannot easily view expert professional information and historical evaluations during the diagnosis process, and lack reference when selecting experts. At the same time, user evaluation information of experts is updated laggingly, which cannot effectively promote experts to improve service quality. SUMMARY

[0011] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present application is to provide an online question and answer system and method for industrial equipment fault diagnosis experts based on vector matching, aiming to provide efficient, accurate and convenient fault diagnosis services through intelligent expert matching mechanism, flexible communication method and perfect record management, so as to improve the overall performance and user experience of industrial equipment fault diagnosis.

[0012] In order to achieve the above-mentioned purpose, the technical scheme adopted by the present application is:

[0013] An online question and answer system for industrial equipment fault diagnosis experts based on vector matching, comprising a system matching expert module and a communication module;

[0014] The system matching expert module comprises a problem description library, an expert database, a matching unit and a recommendation unit; the problem description library stores problem items and problem standard descriptions; the expert database stores expert data, the expert data comprising an expert ID and a problem item that the expert is good at handling; the matching unit is configured to convert an input problem description into a vector representation and match the vector representation with the problem standard descriptions to determine a problem item closest to the input problem description; the recommendation unit is configured to recommend an expert ID that is good at handling the closest problem item from the expert database; the problem item is a failure mode or a fault mode of an industrial equipment; the problem standard description is a text description of the problem item;

[0015] The communication module is configured to realize online communication with the expert corresponding to the recommended expert ID.

[0016] In one embodiment, the input problem description is a text, an image or a video, the matching unit converts the input problem description into a vector representation and constructs a latent Dirichlet allocation (LDA) model, and matches the vector representation with each problem standard description in the following manner:

[0017] First, the text data of historical diagnosis reports is preprocessed;

[0018] Then, the text data of historical diagnosis reports is trained, and the text of each historical diagnosis report is represented as a probability distribution of a set of topics, each topic being a probability distribution of a word corresponding to the problem item;

[0019] The vector representation is input into the trained model to obtain a topic distribution, i.e. a probability distribution of a set of topics, reflecting possible relevant topics, i.e. corresponding problem items.

[0020] In one embodiment, the preprocessing comprises text data cleaning (removing punctuation marks, etc.), removing stop words (using TF-IDF), and dividing into words or word groups, and constructing a document-term matrix (document-term frequency matrix).

[0021] In one embodiment, the representation of the text of each historical diagnosis report as a probability distribution of a set of topics is implemented in the following manner:

[0022] Gibbs sampling is used to calculate the probability of generating each word in a document under each topic, and then the topic is redistributed according to the probability, and the distribution of each document and topic, including document-topic distribution and topic-word distribution, is updated according to the new topic distribution, and the process is repeated until the change in model parameters is less than a predetermined threshold or a specified number of iterations is reached.

[0023] In one embodiment, the subject is matched with an expert who is good at handling the corresponding problem item, so that the matching unit obtains the expert ID who is good at handling the closest problem item and provides it to the recommendation unit.

[0024] In one embodiment, the system further comprises a record management module that saves all communication records, generates a unique session ID for each communication, and introduces automatic tagging and classification technology to automatically organize and tag historical communication content through an AI model, and the saved communication records are used to train the latent Dirichlet allocation (LDA) model.

[0025] In one embodiment, the expert database introduces an expert knowledge base management based on a graph database, and constructs the professional field, historical diagnosis cases, and success rate of an expert into a complete graph structure.

[0026] In one embodiment, the communication module is based on websocket and webRTC technology, and realizes instant communication and real-time video call between the user and the expert, and adopts a communication optimization method based on edge computing in the communication process, and deploys a real-time communication module on an edge device close to the user and the expert.

[0027] In one embodiment, the system further comprises a virtual assistant and a solution database, the solution database stores solutions to failure modes or fault modes of industrial equipment, and the virtual assistant is a chat robot for an industrial scene, pre-processes a question submitted by a user, finds a corresponding solution from the solution database according to the closest problem item, and shows it to the user if there is one.

[0028] In one embodiment, the system further comprises a feedback module based on behavior analysis, which is used to score and evaluate the communication after the user finishes the communication, intelligently analyzes the behavior mode (such as response speed and communication quality) of the expert and the user in the communication process, provides objective scoring suggestion reference for the user, and records the evaluation information of the user in the database to update the scoring information of the expert.

[0029] The application also provides an online question and answer method based on the expert online question and answer system, comprising the following steps:

[0030] The user inputs a problem description on the client;

[0031] The system converts the problem description into a vector representation locally or remotely, matches it with the stored problem standard description, determines the closest problem item to the problem description, and recommends the expert ID who is good at handling the closest problem item;

[0032] The user sends a remote communication request to the expert corresponding to the recommended expert ID on the client side;

[0033] The expert accepts the remote communication request on the client side and establishes remote online communication.

[0034] Compared with the prior art, the present application has the following advantages:

[0035] (1) The present application can convert the user's problem description into a vector representation and match it with the pre-set problem standard description. This intelligent matching mechanism can dynamically adjust the matching result according to the actual situation, improving the accuracy and efficiency of fault diagnosis and reducing the dependence on single expert experience.

[0036] (2) The present application introduces WebSocket and WebRTC technologies to realize instant communication and real-time video call between users and experts. WebSocket provides low-latency full-duplex communication capability to ensure the real-time nature of problem submission and feedback. WebRTC supports point-to-point media transmission, enabling high-quality audio and video communication under low-bandwidth conditions and improving communication efficiency and user experience.

[0037] (3) The present application system supports multiple message forms such as text, pictures, and videos, and also supports real-time video calls. Users can choose the most suitable way to describe problems and communicate with experts, thereby obtaining more accurate diagnosis and guidance and optimizing user experience. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 is a system structure diagram of the present application.

[0039] Figure 2 is a diagnostic system diagram of an embodiment of the present application.

[0040] Figure 3 is a diagnostic flowchart of an embodiment of the present application. DETAILED DESCRIPTION

[0041] The embodiments of the present application will be described in detail below in conjunction with the drawings and examples.

[0042] In the field of industrial equipment fault diagnosis, experts still mainly rely on on-site analysis and processing, which cannot achieve real-time response and fault handling, and there is no good online system to remotely solve these problems in real time. Even some remote systems only stay at the level of text and picture exchange, lacking more intuitive communication means such as real-time video call.

[0043] In view of the above-mentioned defects existing in the prior art, the present application designs an online expert question and answer system for industrial equipment fault diagnosis based on vector matching, and introduces websocket and webRTC technologies to realize remote diagnosis, aiming to solve the defects in the prior art through an intelligent expert matching mechanism, a flexible communication mode and perfect record management.

[0044] As shown in Figure 1 The system of the present application mainly consists of two parts of a system matching expert module and a communication module.

[0045] Expert matching is an indispensable part of the system of the present application, through which the user can find suitable experts according to his own needs, and the system is more flexible and effective.

[0046] In the present application, the system matching expert module mainly includes a problem description library, an expert database, a matching unit and a recommendation unit. The problem description library stores problem items and problem standard descriptions corresponding to the problem items, wherein the problem items are failure modes or fault modes of industrial equipment, and the problem standard descriptions are text descriptions of the corresponding problem items. The expert database stores expert data, mainly including expert IDs and problem items that the experts are good at handling. The expert and the problem item are generally in a many-to-many correspondence, that is, one expert may correspond to multiple problem items, and one problem item may correspond to multiple experts.

[0047] The matching unit and the recommendation unit are important parts of the system matching expert module of the present application, wherein the matching unit is mainly used for converting the problem description input by the user into a vector representation and matching it with the problem standard descriptions in the problem description library to determine the problem item closest to the problem description of the user. The recommendation unit recommends expert IDs good at handling the closest problem item from the expert database, and the recommended expert IDs can be multiple.

[0048] Communication is a basic component of the system of the present application, and the communication parties can send messages in the form of text, pictures and videos. In addition, real-time video calls can also be made, and the parties can effectively communicate to exchange information to solve problems. The communication module of the present application is used to realize online communication with the expert corresponding to the recommended expert ID.

[0049] In hardware, the system of the present application mainly includes a user terminal, an expert terminal, a server and the like, and the components are connected through a network. The user and the expert access the system through their respective terminals, the user submits a problem, the system matches experts, and then the user and the expert communicate in real time, while the system performs record management.

[0050] Reference Figure 2In the expert matching stage, the user input question description can be in various forms such as text, image or video, and the matching unit can convert the input question description into a vector representation according to the keywords in the question description library. For example, when the input is an image or a video, a text vector representation can be obtained by recognizing the image or the video through a picture-text model. Then, the cosine similarity between the text vector representation and the question standard description vector can be calculated to determine the most matched failure mode or fault mode, i.e. the question item. The system then retrieves experts skilled in handling the matched failure mode or fault mode from the expert database, recommends them to the user and displays them on the terminal interface of the user. This intelligent matching mechanism can dynamically adjust the matching result according to the actual situation, improve the accuracy and efficiency of fault diagnosis, and reduce the dependence on the experience of a single expert.

[0051] Further, in order to improve the matching effect and efficiency, the present application constructs a latent Dirichlet allocation (LDA) model and trains it, and uses the trained model to match the vector representation with each question standard description. Still referring to Figure 2 The specific implementation method can be described as follows:

[0052] First, the text data of the historical diagnosis report is preprocessed, including text data cleaning (removing punctuation marks, etc.), removing stop words (using TF-IDF), and dividing into words or phrases, constructing a document-term matrix (document-term frequency matrix), etc.

[0053] Then, the text data of the historical diagnosis report is trained, and the text of each historical diagnosis report is represented as a probability distribution of a set of topics, and each topic is a probability distribution of a word, corresponding to a question item, i.e. a failure mode or a fault mode. Specifically, Gibbs sampling can be used to calculate the probability of generating each word under each topic, and then the topic is redistributed according to the probability. According to the new topic distribution result, the distribution of each document and topic is updated, including the document-topic distribution and the topic-word distribution. The process is repeated until the change of the model parameters is less than a preset threshold, or a specified number of iterations is reached.

[0054] The vector representation of the user question description is input into the trained model to obtain its topic distribution, i.e. a probability distribution of a set of topics, which reflects the possible related topics, i.e. the corresponding question items. The topics can be matched with experts skilled in handling the corresponding question items, so that the expert ID skilled in handling the closest question item is obtained in the matching unit and provided to the recommendation unit.

[0055] Based on this, the present application can directly recommend suitable experts by inputting the topic distribution of the text, thereby directly omitting the part of calculating the cosine similarity and simplifying the matching process.

[0056] According to the above scheme, the present application determines a more suitable failure mode by training a potential Dirichlet distribution model to obtain the topic distribution thereof from the user-submitted problem description. In addition, an expert knowledge base management based on a graph database can be introduced, and information such as the professional field, historical diagnosis cases, and success rate of an expert can be constructed into a complete graph structure, and the expert base based on the graph database is combined for recommendation. This form of expert database can quickly retrieve experts related to the user problem description, and has better scalability in a large-scale data environment. Finally, according to the matched failure mode, an expert skilled in handling the failure mode is retrieved from the database and recommended to the user. This intelligent matching mechanism can dynamically adjust the matching result according to the actual situation, improve the accuracy and efficiency of fault diagnosis, and reduce the dependence on the experience of a single expert.

[0057] The communication module of the present application is based on websocket and webRTC technology. WebSocket is a protocol for full-duplex communication over a single TCP connection, which can realize real-time data transmission between the server and the client. When communicating, the WebSocket technology is used to realize instant communication between the user and the system, and between the user and the expert, so that the user can submit problems and receive system feedback in real time, ensuring the real-time nature of problem submission and feedback. WebSocket provides full-duplex communication capability, and the low-latency feature ensures the efficiency of the fault diagnosis process. WebRTC is an API that supports real-time voice or video conversations between web browsers, which provides point-to-point media transmission capability. In the present application, it is used to realize real-time video calls and multimedia information transmission between the user and the expert. Through WebRTC, the user and the expert can communicate face-to-face, and the expert can view the running state and failure phenomenon of the device in real time, providing more accurate diagnosis and guidance. In addition, WebRTC supports point-to-point media transmission, ensuring high-quality audio and video communication under low-bandwidth conditions, improving communication efficiency and user experience.

[0058] Further, the present application can use a communication optimization method based on edge computing in the communication process, which reduces delay and network congestion by deploying real-time communication modules on edge devices close to the user and the expert, ensuring the real-time nature and stability of communication, which is especially suitable for some industrial sites with poor network conditions.

[0059] Further, the system of the present application further comprises a virtual assistant and a solution database, wherein the solution database stores solutions to failure modes or fault modes of industrial equipment, which are basic and common solutions. The virtual assistant is a chat robot for industrial scenarios, which can preprocess the question submitted by the user, find a corresponding solution from the solution database according to the question item closest to the question submitted by the user, and show the user the corresponding solution if there is a corresponding solution. Through the virtual assistant, the question described by the user can be preprocessed, and automated basic diagnostic opinions can be provided to assist the work of experts and improve the user experience.

[0060] Reference Figure 3 The online question and answer method using the system of the present application mainly includes the following aspects:

[0061] 1. The user and the expert enter the online question and answer system respectively.

[0062] 2. The user inputs a question description in the form of text, picture or video on the client, and the system matches an expert for the question description of the user. Specifically, the following steps are performed:

[0063] The question description submitted by the user is converted into a vector representation through a question description library, and then cosine similarity or an LDA model is used for expert matching. This intelligent matching mechanism improves the accuracy and efficiency of fault diagnosis and reduces the dependence on the experience of a single expert. This process can be completed locally or remotely.

[0064] 3. The user sends a remote communication request, i.e., a chat application, to the expert corresponding to the recommended expert ID on the client.

[0065] 4. The expert accepts the remote communication request on the client and establishes remote online communication.

[0066] 5. The two parties chat.

[0067] The WebSocket and WebRTC technologies are introduced in the online communication of the present application to realize instant communication and real-time video call between the user and the expert. The WebSocket provides low-latency full-duplex communication capability to ensure the real-time nature of question submission and feedback. The WebRTC supports point-to-point media transmission and can perform high-quality audio and video communication under low-bandwidth conditions, thereby improving the communication efficiency and user experience.

[0068] The present application supports various communication modes such as text, picture and video, and the user can select the most suitable mode for question description and expert communication. The real-time video call function enables the expert to intuitively view the equipment state and fault phenomenon, thereby providing more accurate diagnosis and guidance and optimizing the user experience. In addition, the virtual assistant (a chat robot for industrial scenarios) is combined, which can preprocess the question submitted by the user and provide automated basic diagnostic opinions to reduce the burden on the expert.

[0069] 6. The user ends the chat and leaves a review.

[0070] This invention establishes a systematic record management mechanism capable of storing all communication records, including text, images, and videos. Each communication generates a unique session ID, ensuring the integrity and traceability of communication records. It incorporates automatic tagging and classification technology, using an AI model to automatically organize and tag historical communication content for easy subsequent retrieval and analysis. Users and experts can view historical communication records at any time for convenient reference and problem-solving, improving system reliability and user satisfaction.

[0071] Based on the system's comprehensive recording mechanism, in addition to existing historical diagnostic reports in document form, chat logs generated during remote diagnosis (including problem descriptions and diagnostic results) can also be used as data sources to further optimize the LDA model.

[0072] This invention employs a behavior-based feedback system. After a communication session, users can rate and evaluate the interaction. The system intelligently analyzes the behavioral patterns (such as response speed and communication quality) between experts and users, providing objective rating suggestions rather than relying solely on subjective user feedback. The system then records user feedback in a database and updates the expert ratings accordingly. This mechanism helps the system continuously optimize expert matching and service quality, incentivizing experts to provide better service and improving the overall user experience.

[0073] As described above, the system of this invention has excellent cross-domain application potential. It is not only suitable for fault diagnosis of industrial equipment, but can also be applied to other fields such as online consultation in medical systems and equipment maintenance in manufacturing. Its versatility and flexibility give the system broad application prospects and market value.

Claims

1. An online question-and-answer system for industrial equipment fault diagnosis based on vector matching, characterized in that, This includes a system matching expert module and a communication module; The system's expert matching module includes a problem description library, an expert database, a matching unit, and a recommendation unit; the problem description library stores problem items and standard problem descriptions. The expert database stores expert data, including expert IDs and the types of problems each expert specializes in. It incorporates a graph-based expert knowledge base management system, constructing a complete graph structure from expert expertise, historical diagnostic cases, and success rates. The matching unit converts the input problem description into a vector representation and matches it against the standard problem description to determine the problem item closest to the description. The recommendation unit recommends expert IDs from the expert database that specialize in handling the closest problem item. The problem item represents a failure mode or fault mode of industrial equipment. The standard problem description is a textual description of the problem item. The communication module is used to enable online communication with the experts corresponding to the recommended expert IDs; The input question description is text, image, or video. The matching unit converts the input question description into a vector representation, constructs a latent Dirichlet assignment model, and matches the vector representation with the standard question descriptions, as follows: First, the text data of historical diagnostic reports is preprocessed; Then, the text data of historical diagnostic reports are used for training, and the text of each historical diagnostic report is represented as a probability distribution of a set of topics. Each topic is a probability distribution of a word, which corresponds to the question item. The vector representation is input into the trained model to obtain its topic distribution, which is a probability distribution of a set of topics, reflecting the possible related topics, i.e., the corresponding question items.

2. The online question-and-answer system for industrial equipment fault diagnosis based on vector matching according to claim 1, characterized in that, The method for representing the text of each historical diagnostic report as a probability distribution of a set of topics is as follows: Using Gibbs sampling, for each word in a document, the probability of generating that word under each topic is calculated. Then, topics are reassigned based on this probability. Based on the new topic assignment results, the distribution of each document and topic, including document-topic distribution and topic-word distribution, is updated. This process is repeated until the change in model parameters is less than a preset threshold or the specified number of iterations is reached.

3. The online question-and-answer system for industrial equipment fault diagnosis based on vector matching according to claim 1, characterized in that, The topic is matched with experts who are good at handling the corresponding problem item, so that the matching unit obtains the expert ID of the expert who is good at handling the closest problem item and provides it to the recommendation unit.

4. The online question-and-answer system for industrial equipment fault diagnosis based on vector matching according to claim 1, characterized in that, The system also includes a record management module, which saves all communication records. Each communication generates a unique session ID and introduces automatic tagging and classification technology. The system uses an AI model to automatically organize and tag historical communication content. The saved communication records are used to train the potential Dirichlet assignment model.

5. The online question-and-answer system for industrial equipment fault diagnosis based on vector matching according to claim 1, characterized in that, The communication module is based on WebSocket and WebRTC technologies to enable instant communication and real-time video calls between users and experts. During the communication process, an edge computing-based communication optimization method is adopted, and the real-time communication module is deployed on edge devices close to users and experts.

6. The online question-and-answer system for industrial equipment fault diagnosis based on vector matching according to claim 1, characterized in that, The system also includes a virtual assistant and a solution database. The solution database stores solutions for failure modes or fault modes of industrial equipment. The virtual assistant is a chatbot designed for industrial scenarios. It preprocesses user-submitted questions and searches the solution database for corresponding solutions based on the closest question. If a solution is found, it is displayed to the user.

7. The online question-and-answer system for industrial equipment fault diagnosis based on vector matching according to claim 1, characterized in that, The system also includes a feedback module based on behavior analysis. This feedback module is used to score and evaluate the communication after it ends, and to intelligently analyze the behavioral patterns of experts and users during the communication process. It provides users with objective scoring suggestions and records the user's evaluation information in the database to update the expert's scoring information.

8. The online question-and-answer method of the online question-and-answer system for industrial equipment fault diagnosis based on vector matching as described in claim 1, characterized in that, Includes the following steps: The user enters a description of the problem in the client; The system, either locally or remotely, converts the problem description into a vector representation and matches it with the stored standard problem description to determine the problem item that is closest to the problem description, and recommends expert IDs that are good at handling the closest problem item. The user sends a remote communication request to the expert corresponding to the recommended expert ID in the client. The expert accepts the remote communication request on the client side and establishes remote online communication.

Images