Method and electronic equipment for checking and processing faults of industrial personal computer in field operation and maintenance

Abstract

The invention discloses a method for troubleshooting and processing faults of an industrial personal computer in field operation and maintenance and electronic equipment, relates to the field of industrial personal computer equipment, and aims to improve the efficiency and accuracy of fault processing. According to the method, the running state of the industrial personal computer is monitored in real time, fault data are collected, unstructured documents and historical fault data are converted into structured information through the natural language processing technology, and a knowledge graph is established. Through a vectorization technology and an enhanced search algorithm, the system can quickly retrieve cases similar to current fault symptoms from a knowledge base, and fault processing suggestions described by natural languages are generated through LLM. In addition, the invention further comprises a computer program executed by the central processing unit, data of the industrial personal computer are read through the communication module, and an automatic fault diagnosis process is realized. According to the method and the equipment, the dependence on artificial experience is effectively reduced, the reliability of the industrial personal computer is improved, and the downtime is reduced.

Description

Technical Field

[0001] The present invention relates to the field of industrial computer equipment, and in particular to a method and electronic equipment for on-site operation and maintenance troubleshooting and processing of industrial computer faults. Background Art

[0002] Industrial Control Computer (ICC) is a key device in modern industrial automation systems. It is widely used in manufacturing, energy, transportation, chemical industry and other industries. It is responsible for real-time monitoring and control of the production process to ensure the safe, stable and efficient operation of the equipment. With the continuous improvement of industrial automation and intelligence, the operation and maintenance of IPCs has become more and more complicated. In the face of increasingly complex systems, traditional on-site operation and maintenance methods of IPCs have gradually exposed some technical shortcomings.

[0003] Industrial computers have various types of failures, including hardware failures, software failures, communication failures, etc. Traditional operation and maintenance methods mainly rely on the experience and step-by-step troubleshooting of operation and maintenance personnel, which makes it difficult to quickly locate problems. Especially when facing complex interactive failures, manual judgment alone cannot quickly find the root cause of the problem and provide targeted solutions. Due to the low efficiency of troubleshooting, equipment often shuts down for a long time, increasing the operation and maintenance costs of enterprises.

[0004] Therefore, some automatic troubleshooting solutions gradually emerged.

[0005] Rule-based fault diagnosis system: This type of system relies on a pre-set rule base to determine the fault type by matching it with the device status data. The limitation is that the construction and maintenance of the rule base requires a lot of manual intervention, and it is difficult to cover all possible fault types. In addition, as the complexity of industrial computer systems continues to increase, the maintenance and expansion costs of the rule base have also increased dramatically. The system has poor scalability and is difficult to cope with the high complexity of modern industrial automation systems.

[0006] Expert system: Expert system diagnoses faults by simulating the reasoning process of human experts. It relies on the expert knowledge base and uses experience and rules to solve problems. However, the expert knowledge base is updated slowly, making it difficult to cope with the emergence of new equipment and new fault types. Secondly, the reasoning process of the expert system is complex and the response time is long, which cannot meet the real-time requirements. In addition, the operation cost of the expert system is high, and it relies on experienced experts, making it difficult to promote and apply on a large scale.

[0007] Traditional data analysis methods: Using statistical analysis or machine learning techniques, traditional data analysis methods rely on modeling and training of historical data to predict possible failures. To a certain extent, they can reduce the workload of operation and maintenance personnel, but when faced with massive, multi-dimensional, and real-time data, they have problems such as high computational complexity and long response time. Summary of the invention

[0008] In view of the above-mentioned shortcomings of the prior art, an object of the present invention is to provide a method for on-site operation and maintenance troubleshooting and processing of industrial computer failures, so as to solve one or more problems in the prior art.

[0009] To achieve the above object, the technical solution of the present invention is as follows: A method for on-site operation and maintenance troubleshooting and processing of industrial computer failures comprises the following steps: Receive externally imported support documents and collect historical fault data.

[0010] Structural processing is performed on supporting documents and historical fault data, and structured data is output.

[0011] Receive structured data, build a knowledge graph, and output entity and relationship data as search text.

[0012] Receive the search text, convert the search text into a high-dimensional vector, establish a search library, and store the converted search text into the search library.

[0013] After receiving the event triggered by the operation and maintenance personnel's inquiry, the real-time operation data of the industrial control equipment is obtained, the real-time operation data and the operation and maintenance personnel's question are combined, and the query prompt word of the question is output.

[0014] The query prompt words are input into the LLM, and the diagnosis results generated by the LLM after calling the search library through the enhanced search algorithm are received.

[0015] Furthermore, LLM calls the search library through the enhanced algorithm including the following steps: The query prompt words are converted into high-dimensional vectors, and the converted query prompt words are input into the search library.

[0016] By calculating the cosine similarity, the vectors in the search library are sorted from low to high according to the similarity.

[0017] After the search database executes the query task through the search formula, multiple most similar vectors and their corresponding document IDs are returned.

[0018] Get the document ID and the knowledge fragment corresponding to the most similar vector, and output them to LLM.

[0019] Furthermore, the method of converting the search text into a high-dimensional vector is one or more of sentence embedding and paragraph embedding. The method of converting the query prompt word into a high-dimensional vector is an author topic model.

[0020] Furthermore, the diagnosis result includes the fault handling method and steps output in natural language, and the fault-free conclusion output in natural language.

[0021] Furthermore, the structural processing of the supporting document data includes: Remove text formatting from supported documents, delete duplicate paragraphs and irrelevant comments, and get plain text.

[0022] Through the natural language processing library, key information such as equipment descriptions, maintenance procedures, and troubleshooting methods can be extracted from plain text.

[0023] Through document conversion, the extracted content is converted into Markdown format.

[0024] Get structured data.

[0025] Furthermore, the structured processing of historical fault data includes: Delete incomplete, irrelevant, and abnormal historical fault data.

[0026] Fill missing values ​​with the mean value for the same period.

[0027] Reduce the deviation of random values ​​through average smoothing algorithm.

[0028] Get the cleaned data.

[0029] Through the natural language processing library, key entities and relationships are extracted from the cleaned data, and the extracted content is saved in Markdown format.

[0030] Get structured data.

[0031] Furthermore, collecting historical fault data includes: monitoring the operating status of the industrial computer in real time through communication and automatically recording abnormal events to obtain historical fault data.

[0032] Furthermore, when calculating the cosine similarity, the vectors have different weighting factors.

[0033] To achieve the above-mentioned purpose, the second set of technical solutions adopted by the present invention is: an electronic device, based on a method for on-site operation and maintenance troubleshooting and processing of industrial computer failures, including a central processing unit, a storage medium, and a detachable communication module, and the communication module is used to connect to the industrial computer.

[0034] Specifically, the storage medium stores a computer program, and when the computer program is executed by the central processing unit, the central processing unit executes a method for on-site operation and maintenance to troubleshoot and handle industrial computer failures.

[0035] Compared with the prior art, the beneficial technical effects of the present invention are as follows (I) Application of vectorization: Through vectorization, non-quantifiable text-based support documents are converted into quantifiable vector data, which greatly expands the capacity of the knowledge base. Compared with the expert system and rule-based fault diagnosis methods that are difficult to expand, the scale of the knowledge graph depends on the volume of imported support documents, which can be increased, decreased and modified as needed, greatly reducing the possibility of no matching results in diagnosis and improving the reliability of diagnosis.

[0036] (II) High-dimensional vector conversion and cosine similarity calculation: The key information in the question and supporting documents is converted into high-dimensional vectors. After quantifying the knowledge base, a mathematical algorithm is used to find the supporting document content closest to the question to obtain the most appropriate diagnostic opinion. The cosine similarity calculation also includes a weighting factor to consider the importance and reliability of different documents.

[0037] (III) Access to the big model (LLM): Access to the big model (LLM) outputs easy-to-understand diagnostic results and processing methods in natural language through the big model, and improves the accuracy of diagnostic results.

[0038] (IV) Deployment and adaptation of electronic equipment: A complete investigation and processing method is deployed in the electronic equipment, and the electronic equipment can complete data exchange with the industrial computer through communication methods. It can actively adapt to the different interfaces and different models of industrial computers. The adaptation difficulty is lower than the software actively adapting to industrial computers with different hardware. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a logical flow diagram of a method for on-site operation and maintenance troubleshooting and processing of industrial computer failures in the present invention.

[0040] Figure 2 It is a schematic diagram of the operation logic of an electronic device in the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present invention more clear, the following is a further detailed description of a method for on-site operation and maintenance troubleshooting and processing of industrial computer failures proposed by the present invention in combination with the accompanying drawings and embodiments. It should be noted that the structure, proportion, size, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the limiting conditions for the implementation of the present invention, so they have no technical substantive significance. Any modification of the structure, change in the proportional relationship or adjustment of the size should still fall within the scope of the technical content disclosed by the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0042] Application Overview In the field of industrial automation, the stable operation of industrial computers is crucial. However, due to the complexity of industrial computers and the variability of their operating environment, troubleshooting and troubleshooting often requires a lot of time and human resources. Common processing methods in the industry mainly include the following aspects: Document review: When troubleshooting, technicians usually refer to documents such as equipment manuals and maintenance manuals, which contain detailed information about the equipment and solutions to common problems. However, these documents are usually unstructured and difficult to retrieve and apply quickly.

[0043] Historical failure data analysis: Some companies collect and analyze historical failure data in order to discover the patterns and causes of failures. However, this approach is often limited by the integrity of the data and the capabilities of the analysis tools, and it is difficult to respond to new failure modes in real time.

[0044] Expert systems: Some companies will establish expert systems to aggregate the knowledge and experience of industry experts to assist in fault diagnosis. However, these systems often lack flexibility and the ability to update in real time, making it difficult to adapt to the rapidly changing industrial environment.

[0045] To address these issues, this solution significantly improves the efficiency and accuracy of industrial computer troubleshooting through automated structured data processing and enhanced search algorithms. Using LLM (Large Language Model) and search libraries, this solution can quickly and accurately diagnose faults and provide intuitive fault handling guidance in natural language, thereby improving the efficiency and quality of on-site operation and maintenance.

[0046] Exemplary Methods Please refer to Figure 1 , A method for on-site operation and maintenance troubleshooting and processing of industrial computer failures described in the present invention includes: Data collection and preprocessing: Data collection consists of two parts: document data and equipment operation data. Document data covers equipment manuals, maintenance manuals, operation and maintenance records, etc., while equipment operation data is collected in real time through sensors and monitoring systems, including key parameters such as temperature, pressure, flow, fault codes, operating time, and downtime.

[0047] The operating status of the industrial computer is monitored in real time through communication, and abnormal events are automatically recorded to obtain historical fault data. This process can use existing industrial communication protocols, such as Modbus, OPC, etc., to achieve real-time monitoring of the industrial computer. In this way, it can ensure that when an abnormality occurs in the industrial computer, the relevant fault data can be quickly captured, providing a basis for subsequent fault analysis and processing.

[0048] Document data preprocessing: Document data needs to be converted into a format that can be processed by LLM in the subsequent steps. Use the Pandoc tool to convert the Word document to Markdown format using the command pandoc device_manual.docx -f docx -t markdown -o device_manual.md. Use the MinerU tool to parse and convert the PDF file to Markdown format using the command magic-pdf -pdevice_manual.pdf -o . / -m auto.

[0049] Equipment operation data preprocessing: Equipment operation data includes temperature, pressure, flow, fault code, operation time, downtime, etc. These data are collected in real time through sensors and monitoring systems. The collected data needs to be preprocessed by cleaning, de-duplication and de-noising, normalization, etc. to ensure data quality and availability.

[0050] Clean and transform supporting documents and historical fault data to generate structured data. The specific operations are as follows: Support document structured processing: Remove text formatting such as fonts, colors, tables, etc. from supported documents to remove irrelevant information.

[0051] Delete duplicate paragraphs and irrelevant comments to extract the plain text content.

[0052] Leverage the natural language processing library to extract key information such as equipment descriptions, maintenance procedures, and troubleshooting methods from plain text.

[0053] Convert the extracted content into Markdown format for subsequent processing and display.

[0054] Structural processing of historical fault data: Delete incomplete, irrelevant, and abnormal historical fault data to ensure data accuracy and availability.

[0055] The missing values ​​are filled with the average values ​​of the same period to reduce the impact of missing data on the analysis results.

[0056] The average smoothing algorithm is used to reduce the deviation of random values ​​and improve the stability of data.

[0057] Using the natural language processing library, key entities and relationships are extracted from the cleaned data, and the extracted content is saved in Markdown format.

[0058] Building a knowledge graph After receiving the structured data, a knowledge graph is established, and the entity and relationship data are output as search text. The knowledge graph is a structured semantic knowledge base that stores the relationships between entities in the form of a graph. By representing different entities and their relationships in a structured manner, a networked knowledge structure is formed to facilitate subsequent query and analysis. In the present invention, the establishment of a knowledge graph helps to integrate scattered fault data and supporting document information to form a systematic knowledge system. The process of building a knowledge graph mainly involves entity recognition and relationship extraction, so the open source project GraphRAG based on a large model is used to establish the knowledge graph.

[0059] Convert search text into high-dimensional vector The search text is converted into a high-dimensional vector using sentence embedding and / or paragraph embedding methods. Here, bge-large-zh-v1.5 is selected as the vector embedding model. Sentence embedding and paragraph embedding are commonly used techniques in natural language processing. They can convert text into numerical vectors for computer processing and comparison. Compared with other vector embedding methods, sentence embedding can capture the semantic information of the sentence, and paragraph embedding represents the entire paragraph as a vector, which is more suitable for information retrieval in subsequent steps. In this way, text information can be effectively converted into a high-dimensional vector, providing a basis for subsequent similarity calculation and search.

[0060] Create a search library An empty database is established as a search library, the converted search text is stored in the search library, and Milvus and other tools are used for storage and retrieval. In the present invention, the search library has a retrieval function and a corresponding search formula is built in. The establishment of the search library provides a rapid retrieval capability for subsequent fault diagnosis.

[0061] Receive events triggered by inquiries from operation and maintenance personnel When the operation and maintenance personnel make a troubleshooting request, the system will obtain the real-time operation data of the industrial control equipment and output the query prompt words based on the operation and maintenance personnel's questions. This step is the trigger point for fault diagnosis, which combines real-time data with the operation and maintenance personnel's inquiries to provide contextual information for subsequent fault analysis.

[0062] Convert query hints into high-dimensional vectors The author topic model is used to convert the query prompt words into high-dimensional vectors. The author topic model is a natural language processing technology that can infer the author's topic preference based on the text content, which can improve the accuracy and usability of the query prompt words. In the present invention, the query of the operation and maintenance personnel can be converted into a high-dimensional vector by this method, so as to facilitate comparison with the vectors in the search library.

[0063] Invoke the search library through enhanced search algorithms This step uses the enhanced search algorithm to call the search library to achieve fault diagnosis. The specific operations are as follows: Enter the converted query prompt term into the search library.

[0064] Through vectorization and Milvus's cosine similarity calculation, the vectors in the search library are sorted from low to high in terms of similarity. Cosine similarity is a method of measuring vector similarity, which evaluates the similarity of two vectors by calculating the angle between them.

[0065] After the search database executes the query task through the search formula, multiple most similar vectors and their corresponding document IDs are returned.

[0066] Get the document ID and the knowledge fragment corresponding to the most similar vector, and output them to LLM.

[0067] Generate diagnostic results After receiving the knowledge fragments, LLM generates a natural language description of the fault handling method and steps, or outputs a conclusion that there is no fault. This step is the final output of fault diagnosis. It converts complex technical information into easy-to-understand natural language, providing intuitive fault handling guidance for operation and maintenance personnel.

[0068] Electronic devices An electronic device that can be implemented in combination with a method for on-site operation and maintenance troubleshooting of industrial computer faults includes a central processing unit, a storage medium, and a detachable communication module. The communication module is used to connect to the industrial computer. This electronic device can call an API to read historical and real-time operation data stored in the industrial computer through the communication module and the IO interface of the industrial computer itself.

[0069] The storage medium stores a computer program, and the central processing unit will run the computer program when it is running. The program will cause the central processing unit to execute the method for on-site operation and maintenance to troubleshoot and handle industrial computer failures.

[0070] Specific working process Please refer to Figure 2 , 1. Question input and real-time data acquisition: The operation and maintenance personnel enter troubleshooting queries through the user interface, which may involve specific symptoms, abnormal performance or fault codes of the equipment. While entering the questions, the system automatically obtains the real-time operating data of the current industrial control equipment from sensors and monitoring systems, such as temperature, pressure, flow and other parameters.

[0071] 2. Query prompt word generation: The system combines real-time operation data with the questions input by the operation and maintenance personnel, and generates a query prompt word through natural language processing technology. This query prompt word is a high-level summary of the operation and maintenance personnel's questions and is used in the subsequent search and fault diagnosis process.

[0072] 3. Query hint word vectorization: The generated query prompt words will be converted into high-dimensional vectors through the author topic model. This step is to convert text information into a numerical form that can be processed by computers to facilitate comparison with the data in the search library.

[0073] 4. Search library query: The query prompt word vector is input into the search library. The search library has a built-in enhanced search algorithm, which compares the vectors in the search library with the query prompt word vectors through cosine similarity calculation and sorts them by similarity.

[0074] 5. Fault diagnosis: The system performs the search task and retrieves the vectors most similar to the query prompt and their corresponding document IDs. These document IDs are associated with the troubleshooting methods and historical cases in the knowledge graph.

[0075] 6. Knowledge fragment extraction: According to the retrieved document ID, the system extracts the corresponding knowledge fragments from the search library. These knowledge fragments contain possible fault causes, processing methods and steps.

[0076] 7. Generation of diagnostic results: The extracted knowledge fragments are input into the LLM, which combines this information to generate fault handling methods and steps described in natural language. If the retrieved information is not sufficient to determine the fault, the LLM will output a conclusion that there is no fault.

[0077] 8. Result feedback: Finally, the system feeds back the fault handling suggestions or fault-free conclusions generated by LLM to the operation and maintenance personnel through the user interface. The operation and maintenance personnel can perform actual operations based on these suggestions to solve or troubleshoot the fault.

[0078] 9. Execution and Monitoring: The operation and maintenance personnel perform the corresponding troubleshooting steps according to the guidance provided by the system. During the execution process, the system continues to monitor the operating status of the equipment to ensure that the troubleshooting measures are effectively implemented and update the equipment's operating data in real time.

[0079] 10. Loop iteration: If the fault is not resolved, the operation and maintenance personnel can enter the problem again, and the system will repeat the above process and continue fault diagnosis until the problem is resolved.

[0080] Through this continuous dynamic process, the system can provide real-time and accurate fault diagnosis support for operation and maintenance personnel, helping them to quickly respond to and handle industrial computer failures.

[0081] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for on-site operation and maintenance troubleshooting of industrial computer failures, characterized in that: The following steps are involved: Receive externally imported supporting documents; Collect historical fault data; Structural processing of supporting documents and historical fault data, and output of structured data; Receive structured data, build a knowledge graph, and output entity and relationship data as search text; Receive search text and convert the search text into a high-dimensional vector; Establish a search library and store the converted search text in the search library; After receiving the event triggered by the operation and maintenance personnel's inquiry, the real-time operation data of the industrial control equipment is obtained, the real-time operation data and the operation and maintenance personnel's questions are combined, and the query prompt word of the question is output; The query prompt words are input into the LLM, and the diagnosis results generated by the LLM after calling the search library through the enhanced search algorithm are received.

2. A method for on-site operation and maintenance troubleshooting and processing of industrial computer failures as claimed in claim 1, characterized in that: LLM calls the search library through the enhancement algorithm including the following steps: Convert the query prompt word into a high-dimensional vector, and input the converted query prompt word into the search library; By calculating the cosine similarity, the vectors in the search library are sorted from low to high in terms of similarity; After the search database executes the query task through the search formula, multiple most similar vectors and their corresponding document IDs are returned; Get the document ID and the knowledge fragment corresponding to the most similar vector, and output them to LLM.

3. A method for on-site operation and maintenance troubleshooting of industrial computer failures as claimed in claim 2, characterized in that: The method of converting the search text into a high-dimensional vector is one or more of sentence embedding and paragraph embedding; The method of converting query hint words into high-dimensional vectors is the author topic model.

4. A method for on-site operation and maintenance troubleshooting and processing of industrial computer failures according to claim 1, characterized in that: The diagnosis results include the fault handling methods and steps output in natural language, and the fault-free conclusion output in natural language.

5. A method for on-site operation and maintenance troubleshooting and processing of industrial computer failures as claimed in claim 1, characterized in that: Structuring the supporting document data includes: Delete text formatting in supported documents, remove duplicate paragraphs and irrelevant comments, and get plain text; Through the natural language processing library, key information such as equipment description, maintenance steps, and troubleshooting methods are extracted from plain text; Through document conversion, the extracted content is converted into Markdown format; Get structured data.

6. A method for on-site operation and maintenance troubleshooting and processing of industrial computer failures as claimed in claim 5, characterized in that: Structural processing of historical fault data includes: Delete incomplete, irrelevant, and abnormal historical fault data; Fill the missing values ​​with the mean value of the same period; Reduce the deviation of random values ​​through average smoothing algorithm; Get the cleaned data; Through the natural language processing library, key entities and relationships are extracted from the cleaned data, and the extracted content is saved in Markdown format to obtain structured data.

7. A method for on-site operation and maintenance troubleshooting of industrial computer failures as claimed in claim 1, characterized in that: Collecting historical fault data includes: real-time monitoring of the operating status of the industrial computer through communication and automatic recording of abnormal events; obtaining historical fault data.

8. A method for on-site operation and maintenance troubleshooting of industrial computer failures as claimed in claim 2, characterized in that: When calculating cosine similarity, vectors have different weighting factors.

9. An electronic device, based on the method for on-site operation and maintenance troubleshooting and processing of industrial computer failures according to any one of claims 1 to 8, characterized in that: It includes a central processing unit, a storage medium, and a detachable communication module, wherein the communication module is used to connect to an industrial computer.

10. An electronic device as claimed in claim 9, characterized in that: The storage medium stores a computer program, which, when executed by the central processing unit, enables the central processing unit to execute the method for on-site operation and maintenance troubleshooting and processing of industrial computer failures as described in any one of claims 1 to 8.

Images