Production line fault information processing method and system based on industrial agent

Abstract

The invention provides a production line fault information processing method and system based on an industrial agent. The method comprises the following steps that various monitoring data of a production line are collected through various monitoring devices; aligning various monitoring data according to timestamps or event identifiers; extracting features of the aligned monitoring data; mapping the multi-source feature data to the same semantic layer to form a descriptive intermediate text; converting the descriptive intermediate text into a readable text; vectorizing the readable text; converting the troubleshooting manual into a vector database; inputting the fault query vector into a vector database to obtain a fault query result; determining target monitoring data corresponding to the fault query result, and obtaining the target monitoring data again; and inputting a fault query result, the re-acquired target monitoring data and the multi-source feature data into the trained vertical domain large model to obtain a diagnosis result. According to the scheme, the production line fault information processing method can be more intelligent, accurate and rapid.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information processing, and in particular relates to a production line fault information processing method and system based on an industrial agent. BACKGROUND

[0002] Some industrial production lines have a large number of devices and complex processes, and the monitoring of the running state of the production line and the response to faults become key links to maintain the continuity of production. In order to monitor the running state of the production line, monitoring devices such as sensors and video monitoring devices are usually arranged on the production line, and by analyzing the monitoring data of the monitoring devices, it can be determined whether the production line is abnormal.

[0003] However, the current production line fault information processing method has the following problems:

[0004] 1. Lack of unified information fusion and intelligent judgment mechanism:

[0005] The sources of fault information are diverse, which may come from sensors, business system logs, video monitoring or PLC data, and there is a lack of information fusion and intelligent judgment mechanism between different monitoring devices.

[0006] 2. Fault processing depends on the experience of operators:

[0007] The types of on-site faults are various and the causes are complex, and the operators need to combine fault information, fault maintenance manuals and personal experience to make comprehensive judgments to determine the fault processing scheme, which leads to low efficiency and easy mistakes in fault processing. SUMMARY

[0008] The present application provides a production line fault information processing method and system based on an industrial agent to solve the above problems existing in the current production line fault information processing method.

[0009] To solve the above technical problems, the present application provides a production line fault information processing method based on an industrial agent, comprising the following steps:

[0010] S1, collecting various types of monitoring data of the production line through various types of monitoring devices;

[0011] S2, aligning various types of monitoring data according to timestamps or event identifiers, and establishing event context;

[0012] S3, extracting features of the aligned various types of monitoring data to obtain multi-source feature data;

[0013] S4, mapping the multi-source feature data to the same semantic layer to form a descriptive intermediate text;

[0014] S5, converting the descriptive intermediate text into readable text using a rule template or a small language model;

[0015] S6. Vectorize the readable text to obtain the fault query vector;

[0016] S7. Convert the troubleshooting manual into a vector database;

[0017] S8. Input the fault query vector into the vector database to obtain the fault query result;

[0018] S9. Determine the target monitoring data corresponding to the fault query result, and reacquire the target monitoring data;

[0019] S10. Input the fault query results, the reacquired target monitoring data, and the multi-source feature data into the trained vertical domain large model to obtain the diagnostic results.

[0020] Preferably, after step S10, the following steps are also included:

[0021] S11. The diagnostic results are compressed and simplified to obtain simplified diagnostic results;

[0022] S12. Output the simplified diagnostic results through the output device.

[0023] Preferably, step S11 includes the following steps: extracting the fault number and the location information of the faulty device from the diagnostic results.

[0024] Preferably, step S11 further includes the following step: extracting fault level information from the diagnostic results.

[0025] Preferably, step S11 further includes the following step: extracting the cause of the fault from the diagnostic results.

[0026] Preferably, the output device is a voice broadcasting device.

[0027] Preferably, after step S10, the following steps are also included:

[0028] S13. Receive the evaluation information of the diagnostic results; the vertical domain large model uses the diagnostic results and corresponding evaluation information to learn and update, generating a new fault analysis knowledge base.

[0029] Preferably, the production line includes a dryer.

[0030] Preferably, the monitoring data includes material height, material temperature, dryer speed, and dryer vibration signal.

[0031] This invention also provides a production line fault information processing system based on industrial intelligent agents, comprising the following modules:

[0032] The data acquisition module is used to collect various monitoring data from the production line through various monitoring devices;

[0033] The data preprocessing module is used to align various monitoring data by timestamp or event identifier and establish event context;

[0034] The feature extraction module is used to extract features from various types of aligned monitoring data to obtain multi-source feature data;

[0035] The feature fusion module is used to map the multi-source feature data to the same semantic layer to form descriptive intermediate text;

[0036] The text generation module uses rule templates or a small language model to convert the descriptive intermediate text into readable text;

[0037] The vector conversion module is used to vectorize the readable text to obtain a fault query vector.

[0038] The vector conversion module is also used to convert the fault diagnosis manual into a vector database;

[0039] The matching module is used to input the fault query vector into the vector database to obtain the fault query result;

[0040] The matching module is also used to determine the target monitoring data corresponding to the fault query result and to reacquire the target monitoring data;

[0041] The output module is used to input the fault query results, the reacquired target monitoring data, and the multi-source feature data into the trained vertical domain large model to obtain the diagnostic results.

[0042] This invention provides a method and system for processing production line fault information based on industrial intelligent agents. The industrial intelligent agent can automatically execute each step, thereby replacing human decision-making and mobilizing external equipment to perform specific tasks, such as directly sending stop commands to the production line control terminal. Traditional large-scale language models can only think and output results according to fixed patterns and cannot mobilize external equipment to perform specific tasks. This solution first fuses multi-source feature data to obtain query vectors, then converts the fault repair manual into a vector database. The query vectors are used to automatically retrieve fault query results from the vector database. Finally, the vertical domain large-scale model outputs diagnostic results based on the fault query results, reacquired target monitoring data, and multi-source feature data. This makes the production line fault information processing method more intelligent, accurate, and fast, and no longer relies on the personal experience of operators. Attached Figure Description

[0043] Figure 1This is a flowchart of a production line fault information processing method based on an industrial intelligent agent, provided by an embodiment of the present invention. Detailed Implementation

[0044] To make the objectives, advantages, and features of the present invention clearer, the following detailed description of a production line fault information processing method and system based on industrial intelligence, proposed by the present invention, is provided in conjunction with the accompanying drawings. It should be noted that the drawings are all in a very simplified form and use non-precise proportions, and are only used to facilitate and clearly illustrate the objectives of the embodiments of the present invention.

[0045] In the description of this invention, the terms "first," "second," and other qualifiers are added for convenience of description and reference, and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with qualifiers such as "first" and "second" may explicitly or implicitly include one or more of that feature.

[0046] like Figure 1 As shown, this embodiment provides a method for processing production line fault information based on industrial intelligent agents, including the following steps:

[0047] S1. Collect various monitoring data from the production line using various monitoring devices. These devices include sensors and video surveillance equipment, which can collect real-time status data of the production line.

[0048] S2. Align various monitoring data according to timestamps or event identifiers to establish event context.

[0049] S3. Extract features from the aligned monitoring data to obtain multi-source feature data. Before feature extraction, the data can be processed with filtering, noise reduction, and missing data completion. For structured data (such as temperature and current), statistical features and trend change points can be extracted. For text data, keyword extraction and named entity recognition can be performed. For video streams, keyframe detection, target recognition, and abnormal action detection can be performed.

[0050] S4. Map the multi-source feature data to the same semantic layer to form descriptive intermediate text. The descriptive intermediate text can be represented as: Dryer B section inlet - temperature rises by 5°C - conveyor belt speed is normal - material level drops to 0 - time 10:13.

[0051] S5. Using rule templates or a small language model, the descriptive intermediate text is converted into readable text. The readable text can be expressed as: There are signs of material interruption at the inlet of section B of the dryer. The equipment is operating normally but the material level is zero, and the temperature is rising abnormally, suspected to be due to upstream material supply blockage.

[0052] S6. Vectorize the readable text to obtain the fault query vector. The readable text can be processed by sentence segmentation, redundancy removal, synonym substitution, and keyword tagging to obtain the fault query vector.

[0053] S7. Convert the troubleshooting manual into a vector database. The troubleshooting manual can be segmented into logical paragraphs (each paragraph representing a fault or handling item). Each paragraph retains metadata (including equipment type, process step, fault number, keywords, and repair measures). A text embedding model can be used to convert each paragraph into a high-dimensional vector representation; non-text content (tables, images) is generated into text through OCR (Optical Character Recognition) or structural recognition and also vectorized; the vectors and metadata are stored together in the vector database.

[0054] S8. Input the fault query vector into the vector database to obtain the fault query results. Similarity retrieval can be performed in the vector database based on the fault query vector to obtain the most relevant paragraphs; then, a re-ranking model is used to fine-tune the results based on semantic similarity, temporal context, and device consistency. The result ranked first, i.e., with the highest confidence, can be used as the fault query result.

[0055] S9. Determine the target monitoring data corresponding to the fault query result, and reacquire the target monitoring data. For example, if the fault query result is "abnormal material level," the corresponding target monitoring data is the material height. If the fault query result is "material temperature too high," the corresponding target monitoring data is the material temperature.

[0056] S10. Input the fault query result, the reacquired target monitoring data, and the multi-source feature data into the trained vertical domain large model to obtain the diagnostic result. After training, the vertical domain large model can execute the following logic: determine whether the fault query result matches the current operating condition (i.e., the reacquired target monitoring data); analyze whether the causal relationship of each piece of evidence (e.g., material interruption causing temperature increase rather than temperature decrease) is correct. Based on the input data, the vertical domain large model can output the diagnostic result, which may include the fault number, cause, confidence level, and recommended measures. The diagnostic result can be represented as follows:

[0057] Fault number: E-314.

[0058] Cause: Blockage in the upstream feed conveyor belt.

[0059] Confidence level: 92%.

[0060] Recommended measures: Suspend the feeding equipment and inspect the material conveyor.

[0061] This embodiment provides a production line fault information processing method based on industrial intelligent agents. The industrial intelligent agent can automatically execute each step, thereby replacing human decision-making and mobilizing external equipment to perform specific tasks, such as directly sending stop commands to the production line control terminal. Traditional large-scale language models, on the other hand, can only think and output results according to fixed patterns and cannot mobilize external equipment to perform specific tasks. This solution first fuses multi-source feature data to obtain query vectors, then converts the fault repair manual into a vector database. The query vectors are used to automatically retrieve fault query results from the vector database. Finally, the vertical domain large-scale model outputs diagnostic results based on the fault query results, reacquired target monitoring data, and multi-source feature data. This makes the production line fault information processing method more intelligent, accurate, and fast, and no longer relies on the personal experience of operators.

[0062] The present invention provides a production line fault information processing method based on industrial intelligent agents, which can be applied to automated production lines in fields such as rail transportation, tobacco, automobiles, electronics, and chemicals.

[0063] Preferably, after step S10, the following steps are also included:

[0064] S11. The diagnostic results are compressed and simplified to obtain simplified diagnostic results. Semantic analysis is performed on the diagnostic results to extract key phrases (such as "fault number E-314", "B segment entry", "severity level"), unnecessary modifiers and redundant causal descriptions are deleted, and only core information is retained.

[0065] S12. Output the simplified diagnostic results through the output device.

[0066] This embodiment provides a production line fault information processing method based on industrial intelligent agents. By compressing and simplifying diagnostic results, the output efficiency of the output devices can be improved. After compression and simplification, the mapping relationship between the "diagnostic result" and the "simplified diagnostic result" is retained, ensuring information traceability and reverse lookup. If a detailed review is required later, the "diagnostic result" containing the complete context can be restored from the "simplified diagnostic result".

[0067] Preferably, step S11 includes the following steps: extracting the fault number and the location information of the faulty equipment from the diagnostic results. The simplified diagnostic results can be divided into multiple modes, which can be automatically switched or manually switched according to the severity of the fault and the type of broadcast terminal. If the simplified diagnostic results only include the fault number and the location information of the faulty equipment, such as "Dryer B section fault E-314", this mode can be called mode A or minimalist mode.

[0068] Preferably, step S11 further includes the following step: extracting fault level information from the diagnostic results. For example, the simplified diagnostic result could be "Dryer B section fault E-314 Emergency level", which can be called Mode B or Briefing Mode.

[0069] Preferably, step S11 further includes the following step: extracting the cause of the fault from the diagnostic result. For example, the simplified diagnostic result may be "Dryer B section fault E-314 upstream feed blockage", which can be called mode C or extended mode. This mode is suitable for training or debriefing.

[0070] Preferably, the output device is a voice broadcasting device. Real-time text-to-speech (TTS) technology can be used to generate voice signals from compressed text; the voice supports multiple languages ​​and voice timbres, ensuring clear intelligibility in high-noise industrial environments; the voice output can be pushed to the corresponding responsible person via broadcast, headset, or mobile terminal.

[0071] Step S10 is followed by the following steps:

[0072] S13. Receive the evaluation information of the diagnostic results; the vertical domain large model uses the diagnostic results and corresponding evaluation information to learn and update, generating a new fault analysis knowledge base. Operators can evaluate the diagnostic results, and the evaluation information includes the correctness of the diagnostic results, ease of understanding, and corrected diagnostic results. The vertical domain large model uses the diagnostic results and corresponding evaluation information to learn and update, generating a new fault analysis knowledge base to improve the accuracy of subsequent fault cause determination.

[0073] Preferably, the production line includes a dryer. This embodiment provides a production line fault information processing method based on industrial intelligence, which can perform fault analysis on equipment such as dryers.

[0074] Preferably, the monitoring data includes signals such as material height, material temperature, dryer speed, and dryer vibration. By collecting multi-source characteristic data of the dryer, the cause of dryer malfunction can be accurately determined.

[0075] Based on the same technical concept as the above-mentioned production line fault information processing method based on industrial intelligent agents, this embodiment provides a production line fault information processing system based on industrial intelligent agents, including the following modules:

[0076] The data acquisition module is used to collect various monitoring data from the production line through various monitoring devices;

[0077] The data preprocessing module is used to align various monitoring data by timestamp or event identifier and establish event context;

[0078] The feature extraction module is used to extract features from various types of aligned monitoring data to obtain multi-source feature data;

[0079] The feature fusion module is used to map the multi-source feature data to the same semantic layer to form descriptive intermediate text;

[0080] The text generation module uses rule templates or a small language model to convert the descriptive intermediate text into readable text;

[0081] The vector conversion module is used to vectorize the readable text to obtain a fault query vector.

[0082] The vector conversion module is also used to convert the fault diagnosis manual into a vector database;

[0083] The matching module is used to input the fault query vector into the vector database to obtain the fault query result;

[0084] The matching module is also used to determine the target monitoring data corresponding to the fault query result and to reacquire the target monitoring data;

[0085] The output module is used to input the fault query results, the reacquired target monitoring data, and the multi-source feature data into the trained vertical domain large model to obtain the diagnostic results.

[0086] In summary, this invention provides a production line fault information processing method and system based on industrial intelligent agents. The industrial intelligent agent can automatically execute each step, thereby replacing human decision-making and mobilizing external equipment to perform specific tasks, such as directly sending stop commands to the production line control terminal. Traditional large-scale language models, on the other hand, can only think and output results according to fixed patterns and cannot mobilize external equipment to perform specific tasks. This solution first fuses multi-source feature data to obtain query vectors, then converts the fault repair manual into a vector database. The query vectors are used to automatically retrieve fault query results from the vector database. Finally, the vertical domain large-scale model outputs diagnostic results based on the fault query results, reacquired target monitoring data, and multi-source feature data. This makes the production line fault information processing method more intelligent, accurate, and rapid, and no longer reliant on the personal experience of operators.

[0087] The above description is only a description of preferred embodiments of the present invention and is not intended to limit the scope of the present invention in any way. Any changes or modifications made by those skilled in the art based on the above disclosure shall fall within the protection scope of the present invention.

Claims

1. A method for processing production line fault information based on industrial intelligent agents, characterized in that, Includes the following steps: S1. Collect various monitoring data from the production line using various monitoring devices; S2. Align various monitoring data according to timestamps or event identifiers to establish event context; S3. Extract the features of various types of aligned monitoring data to obtain multi-source feature data; S4. Map the multi-source feature data to the same semantic layer to form descriptive intermediate text; S5. Using rule templates or small language models, the descriptive intermediate text is converted into readable text; S6. Vectorize the readable text to obtain the fault query vector; S7. Convert the troubleshooting manual into a vector database; S8. Input the fault query vector into the vector database to obtain the fault query result; S9. Determine the target monitoring data corresponding to the fault query result, and reacquire the target monitoring data; S10. Input the fault query results, the reacquired target monitoring data, and the multi-source feature data into the trained vertical domain large model to obtain the diagnostic results.

2. The production line fault information processing method based on industrial intelligent agents as described in claim 1, characterized in that, Step S10 is followed by the following steps: S11. The diagnostic results are compressed and simplified to obtain simplified diagnostic results; S12. Output the simplified diagnostic results through the output device.

3. The production line fault information processing method based on industrial intelligent agents as described in claim 2, characterized in that, Step S11 includes the following steps: extracting the fault number and the location information of the faulty device from the diagnostic results.

4. The production line fault information processing method based on industrial intelligent agents as described in claim 3, characterized in that, Step S11 also includes the following step: extracting fault level information from the diagnostic results.

5. The production line fault information processing method based on industrial intelligent agents as described in claim 3, characterized in that, Step S11 also includes the following step: extracting the cause of the fault from the diagnostic results.

6. The production line fault information processing method based on industrial intelligent agents as described in claim 2, characterized in that, The output device is a voice broadcasting device.

7. The production line fault information processing method based on industrial intelligent agents as described in claim 1, characterized in that, Step S10 is followed by the following steps: S13. Receive the evaluation information of the diagnostic results; the vertical domain large model uses the diagnostic results and corresponding evaluation information to learn and update, generating a new fault analysis knowledge base.

8. The production line fault information processing method based on industrial intelligent agents as described in claim 1, characterized in that, The production line includes a dryer.

9. The production line fault information processing method based on industrial intelligent agents as described in claim 8, characterized in that, The monitoring data includes material height, material temperature, dryer speed, and dryer vibration signal.

10. A production line fault information processing system based on industrial intelligent agents, characterized in that, Includes the following modules: The data acquisition module is used to collect various monitoring data from the production line through various monitoring devices; The data preprocessing module is used to align various monitoring data by timestamp or event identifier and establish event context; The feature extraction module is used to extract features from various types of aligned monitoring data to obtain multi-source feature data; The feature fusion module is used to map the multi-source feature data to the same semantic layer to form descriptive intermediate text; The text generation module uses rule templates or a small language model to convert the descriptive intermediate text into readable text; The vector conversion module is used to vectorize the readable text to obtain a fault query vector. The vector conversion module is also used to convert the fault diagnosis manual into a vector database; The matching module is used to input the fault query vector into the vector database to obtain the fault query result; The matching module is also used to determine the target monitoring data corresponding to the fault query result and to reacquire the target monitoring data; The output module is used to input the fault query results, the reacquired target monitoring data, and the multi-source feature data into the trained vertical domain large model to obtain the diagnostic results.

Images