A mechanical rotating component fault diagnosis method based on size model cooperation

Abstract

The application provides a mechanical rotating component fault diagnosis method based on size model cooperation, and belongs to the technical field of equipment fault diagnosis. The method comprises the following steps: obtaining the running state signal of a current mechanical rotating component; inputting the running state signal into a pre-trained large-scale language model (LLM) to obtain a large model fault prediction result; collecting the original vibration and acoustic emission signals of the current mechanical rotating component and inputting the original vibration and acoustic emission signals into a convolutional neural network to output original features; encoding the large model fault prediction result into a continuous conditional vector, generating scaling and offset parameters, modulating the original features based on the scaling and offset parameters, processing the modulated original features by using a full connection network, and obtaining a small model fault prediction result; when the large model fault prediction result is inconsistent with the small model fault prediction result, an error judgment mechanism is used for judgment, and the best fault prediction result is output. The method realizes efficient cooperation of multiple models, improves equipment maintenance efficiency, and ensures efficient operation of production equipment.

Description

Technical Field

[0001] This invention belongs to the field of equipment fault diagnosis technology, specifically relating to a fault diagnosis method for mechanical rotating parts based on the collaboration of large and small models. Background Technology

[0002] With the continuous development of intelligent and automated industrial equipment, fault diagnosis plays an increasingly important role in equipment maintenance and fault prediction. Traditional fault diagnosis methods mainly rely on model-based techniques or data-driven machine learning methods. However, these methods face significant challenges when dealing with highly heterogeneous data, especially the coordination problem between large-scale language models (LLMs) and traditional neural network models. Large-scale language models, due to their powerful versatility and natural language processing capabilities, can typically handle large amounts of industrial data and provide users with accurate preliminary fault diagnoses. However, because LLMs require significant computational resources and cannot directly adapt to the details of specific application scenarios, smaller models such as recurrent neural networks or convolutional neural networks are often used as supplementary methods, specifically for handling fault diagnosis tasks in certain domains.

[0003] However, when existing large and small models work together, the lower reliability of the small model is due to the difference in model structure, which affects the coordination efficiency of the large and small models. Furthermore, when the judgment results of the two are inconsistent, it is usually difficult to determine whether the error is due to the large model or the small model, which affects the operation and maintenance efficiency of the equipment. Summary of the Invention

[0004] To address the shortcomings of existing large-scale model collaboration methods in industrial equipment fault diagnosis, this invention provides a fault diagnosis method for mechanical rotating components based on large-scale model collaboration.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A fault diagnosis method for mechanical rotating parts based on size model collaboration includes the following steps: The current operating status signal of the rotating mechanical component is obtained, and the operating status signal is input into a pre-trained large-scale language model (LLM) to obtain the fault prediction result of the large model. The original vibration and acoustic emission signals of the rotating mechanical component are collected, and the original vibration and acoustic emission signals are input into a convolutional neural network to output the original features. The large model fault prediction result is encoded into a continuous conditional vector, and scaling and offset parameters are generated based on the conditional vector. The original features are modulated based on the scaling and offset parameters, and the modulated original features are processed using a fully connected network to obtain the small model fault prediction result. When the fault prediction results of the large model are inconsistent with those of the small model, an error judgment mechanism is used to make a judgment and output the best fault prediction result.

[0006] Preferably, structured data relating industrial equipment to fault characteristics is extracted based on industrial equipment data; the large-scale language model LLM is trained based on the structured data to obtain a pre-trained large-scale language model LLM.

[0007] Preferably, the step of using an error judgment mechanism to determine and output the best fault prediction result specifically includes the following steps: Obtain the confidence scores of large-scale language models and convolutional neural networks, set high confidence thresholds and low confidence thresholds, and if the confidence score of one side is greater than the high confidence threshold and the difference between the confidence scores of the other side is greater than a set value, then it is preliminarily determined that the low confidence side is wrong. This paper analyzes the inference chain of a large-scale language model (LLM) and extracts the key feature descriptions that the LLM depends on. It uses an attribution algorithm to visualize the actual signal feature regions that the small model focuses on in its decision-making. If the key feature descriptions that the LLM depends on are highly consistent with the actual signal feature regions, but the results conflict, it supports the preliminary judgment that the low-confidence side is wrong. If the feature focus is significantly different, the model result that is more consistent with the physical mechanism is adopted. The paper outputs the best fault prediction result.

[0008] Preferably, if the convolutional neural network fails, it also includes using incremental learning and adaptive optimization strategies to dynamically adjust the parameters of the convolutional neural network.

[0009] Preferably, the structured data that correlates entity information with fault features is extracted based on industrial equipment data using the GPT-4 large language model.

[0010] Preferably, the method further includes performing noise reduction and normalization preprocessing on the original vibration and acoustic emission signals before processing them by the convolutional neural network, converting them into a two-dimensional time-frequency graph through short-time Fourier transform, and inputting the two-dimensional time-frequency graph into the convolutional neural network.

[0011] This invention also provides a fault diagnosis system for mechanical rotating parts based on size model collaboration, specifically including: The first model prediction module is used to acquire the current operating status signal of the rotating mechanical component, input the operating status signal into the pre-trained large-scale language model LLM, and obtain the large model fault prediction result.

[0012] The second model prediction module is used to collect the original vibration and acoustic emission signals of the current rotating mechanical parts, input the original vibration and acoustic emission signals into a convolutional neural network, and output the original features; encode the large model fault prediction result into a continuous condition vector, generate scaling and offset parameters based on the condition vector, modulate the original features based on the scaling and offset parameters, and process the modulated original features using a fully connected network to obtain the small model fault prediction result.

[0013] An optimized output module is used to determine the best fault prediction result when the fault prediction results of the large model and the small model are inconsistent, by using an error judgment mechanism.

[0014] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps described in the method for fault diagnosis of mechanical rotating parts based on size model coordination.

[0015] The present invention also provides a computer-readable storage medium storing a computer program that, when loaded by a processor, can execute the steps described in the method for fault diagnosis of mechanical rotating parts based on size model coordination.

[0016] The fault diagnosis method for mechanical rotating parts based on size model collaboration provided by this invention has the following beneficial effects: This invention utilizes a large-scale language model for preliminary fault diagnosis based on the equipment's operating status. A small model is used to process the equipment's original vibration and acoustic emission signals. Based on the preliminary fault diagnosis, the features obtained from the small model are modulated. The large model guides the small model's diagnosis, enhancing the reliability of the small model's processing. The proposed error source judgment mechanism effectively identifies the sources of diagnostic errors in both the large and small models. When the small model malfunctions, incremental learning and adaptive optimization strategies are employed to adjust the small model promptly, thereby ensuring the accuracy and robustness of the overall diagnostic system. Attached Figure Description

[0017] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a fault diagnosis method for mechanical rotating parts based on size model collaboration according to the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0020] Example This invention provides a fault diagnosis method for mechanical rotating parts based on size model collaboration, which includes the following steps: Step 1: Convert industrial data (such as maintenance manuals, historical maintenance records, etc.) into structured data and store it in the Neo4j graph database.

[0021] Large models such as gpt4 are used to extract entities from industrial data, such as bearing model: 6205-2RS, fault type: outer ring spalling, characteristic frequency: 0.4×BPFO, possible cause: poor lubrication, repair measures: replace the bearing and clean the oil circuit; fault features extracted from vibration / acoustic emission signals (such as "the amplitude of the characteristic frequency of the bearing outer ring fault increases with harmonics" and "the RMS value of the acoustic emission signal continues to rise when the tool is worn") are used as independent nodes.

[0022] Step 2: Use a large-scale language model (LLM) for preliminary fault diagnosis. Based on the equipment's operating status or fault description, the LLM generates preliminary fault prediction results.

[0023] (1) Based on the operating status or fault description of the equipment, the user submits a fault problem or diagnosis request to the system.

[0024] (2) LLM utilizes its powerful natural language processing capabilities to combine knowledge extracted from historical maintenance records, equipment operating status and other relevant documents to generate preliminary fault prediction results.

[0025] (3) Based on the input device status information, LLM generates a preliminary diagnosis for a specific fault and proposes possible fault types.

[0026] (4) Based on the preliminary fault diagnosis, the important features corresponding to the fault are assigned higher weights.

[0027] (5) The preliminary diagnostic results of LLM can help users quickly understand the potential problems of the equipment and provide a preliminary basis for subsequent detailed analysis.

[0028] Step 3: Construct a fault diagnosis model based on a small neural network (CNN) to obtain the fault prediction results of the small model. Perform detailed analysis on specific types of faults to optimize the preliminary diagnostic results of the large model.

[0029] (1) The original vibration and acoustic emission signals were preprocessed by noise reduction and normalization, and then converted into a two-dimensional time-frequency graph by short-time Fourier transform (STFT) as the input of the network.

[0030] (2) The preliminary diagnostic results (such as fault type text) output by LLM are encoded into a continuous condition vector through an embedding layer.

[0031] (3) Introduce the above conditional vectors into the feature extraction process of the small model. Specifically, a conditional modulation layer is introduced into the key layer after feature extraction in the convolutional neural network (CNN) (such as the last convolutional layer). This layer generates a set of scaling (γ) and offset (β) parameters based on the conditional vectors through a lightweight fully connected network, and performs an affine transformation on the current features (i.e., modulated features = γ * original features + β), so that the features extracted by the network focus more on the patterns related to the preliminary diagnosis of LLM.

[0032] (4) The AdamW optimizer is used, and the layered learning rates are set for the feature extraction layer, conditional modulation layer and classification layer in the network to achieve stable and efficient training.

[0033] (5) Integrate the diagnostic results and confidence scores of the small model with the preliminary results of the LLM. If the confidence score of the small model is higher than the set threshold (e.g., 0.9), the results of the small model shall prevail; otherwise, retain the LLM results or trigger manual review.

[0034] Step 4: When there is an inconsistency between the diagnostic results of the large model and the small model, the error source judgment mechanism is activated. By comparing the confidence and diagnostic paths of the two, it is determined whether the error is due to the large model or the small model, and corresponding measures are taken to make adjustments.

[0035] (1) The system automatically detects whether the output categories of LLM and SM conflict. If they conflict, the confidence levels of both sides are compared. Two confidence thresholds are set: high and low (e.g., θ_high=0.95, θ_low=0.6). If the confidence level of one side is much higher than that of the other side and exceeds θ_high, it is preliminarily determined that the side with lower confidence level is wrong.

[0036] (2) To verify the preliminary judgment, a diagnostic path analysis was conducted. The inference chain of the LLM was analyzed, and the key feature descriptions it depends on were extracted (such as "focus on the bearing outer ring failure frequency BPFO"). At the same time, attribution algorithms (such as Grad-CAM) were used to visualize the actual signal feature regions that the SM focuses on during decision-making. If the features on which the LLM inference depends are highly consistent with the feature regions that the SM actually focuses on, but the results conflict, the preliminary judgment is supported; if the feature focus points are significantly different, the model results that are more consistent with the physical mechanism are more likely to be accepted.

[0037] (3) Execute final arbitration based on the above analysis: Execute arbitration based on the above analysis: If it is determined that the LLM is wrong, the system adopts the diagnostic result of the SM; if it is determined that the SM is wrong, the LLM result is adopted and the incremental learning process of the small model is automatically triggered; if the confidence of both parties is low or the diagnostic path analysis cannot reach a conclusion, the system submits the case to human expert for adjudication.

[0038] Step 5: If the error originates from the small model, incremental learning and adaptive optimization strategies are used to update the small model, dynamically adjusting its parameters to improve its diagnostic accuracy. The diagnostic results of both the large and small models are then comprehensively evaluated. The accuracy of the diagnostic results is verified through evaluation of different indicators.

[0039] (1) Based on the feedback of arbitration in the error source judgment mechanism, identify the sample set of small model diagnosis errors and analyze its error mode (such as confused fault categories).

[0040] (2) By combining the above-mentioned erroneous samples with a small amount of recent new fault data to form an incremental learning dataset, the model can be prevented from catastrophic forgetting.

[0041] (3) Based on the incremental learning dataset, an adaptive optimization strategy is adopted to dynamically adjust the parameters of the small model. The optimization algorithm includes adjusting the learning rate and updating the model weights, so that it can make a more accurate judgment when it encounters a similar problem next time.

[0042] A hierarchical learning rate strategy is employed to fine-tune a small model. A lower learning rate (e.g., 0.0001) is set for the original feature extraction layer to preserve existing knowledge, while a higher learning rate (e.g., 0.001) is set for the conditional modulation layer and classification layer closer to the output layer. This allows the model to quickly adapt to new error patterns, enabling more accurate judgments when encountering similar problems in the future. This is a simple yet effective adaptive optimization strategy.

[0043] (4) Continuously monitor the performance of the small model after the update and periodically evaluate its diagnostic accuracy. If the performance deteriorates or a new error pattern is encountered, repeat the above incremental learning process to ensure that the model is continuously optimized in long-term operation.

[0044] (5) Continuously monitor the performance of the small model, regularly evaluate its diagnostic accuracy, and make further adjustments as necessary to ensure that the diagnostic accuracy of the small model continues to improve in long-term use.

[0045] Step Six: Validate the model using equipment datasets from different environmental conditions to evaluate its generalization ability. Ensure the diagnostic system performs stably under various industrial equipment and environments.

[0046] The present invention has the following advantages, including: (1) By combining the collaborative work of large and small models, a more accurate fault diagnosis method is provided. When there is an error in the diagnosis results of the large and small models, the source of error can be automatically identified through the error source judgment mechanism, and the small model can be dynamically updated and optimized. Specifically, when the diagnosis results of the large and small models are inconsistent, it can be determined whether the error originates from the large or small model, thereby effectively improving the accuracy and robustness of fault diagnosis. Through this mechanism, the present invention can effectively reduce the misdiagnosis rate, improve equipment maintenance efficiency, reduce equipment downtime and maintenance costs caused by fault judgment errors, and ensure the efficient operation of production equipment.

[0047] (2) By adopting incremental learning and adaptive optimization strategies, the small model can quickly adapt to new data and fault modes during the fault diagnosis process. When faced with constantly changing fault types, the stability and accuracy of the diagnostic system can be maintained by continuously optimizing the small model, especially when equipment and environment change. By combining dynamic node activation mechanism and incremental learning, the small model can adjust its learning process through real-time data, ensuring that each diagnostic result is optimal and improving the long-term adaptability and generalization ability of the model.

[0048] (3) It has broad application prospects, especially in the fields of industrial internet and intelligent manufacturing. By combining the advantages of large and small models and introducing knowledge graph support, it can more comprehensively analyze the equipment's operating data, fault history, and maintenance records, thereby providing more accurate diagnostic results. Compared with traditional single-model diagnostic methods, this invention can adapt to more complex and variable industrial scenarios, solve the problem of collaboration between multiple models, and improve the efficiency of equipment management and fault prediction.

[0049] (4) It can effectively improve the accuracy of fault diagnosis and the robustness of the system, helping enterprises to grasp the operating status of equipment in real time. Through the collaborative work of large and small models, errors in fault diagnosis and prediction are significantly reduced, providing accurate guidance for equipment maintenance, extending equipment service life, reducing maintenance costs, and improving the overall stability and production efficiency of the production line. It solves the problems in traditional fault diagnosis methods. Traditional fault diagnosis systems usually rely on a single model for prediction, but often cannot adapt to complex production environments and changes in equipment status. By introducing the collaborative work of large and small models and designing an error source judgment mechanism, efficient collaboration of multiple models is achieved. Through incremental learning and dynamic optimization, it maintains efficient and stable fault diagnosis capabilities in constantly changing industrial environments, significantly improving the accuracy and timeliness of equipment management and maintenance.

[0050] (5) It has strong innovation and practicality, especially in the application of smart devices and industrial Internet. It can reduce equipment downtime through accurate fault diagnosis, improve the operating efficiency of production lines, and ensure that the most appropriate measures are taken during equipment maintenance to reduce production costs and improve the reliability and safety of production equipment.

[0051] This invention also provides a fault diagnosis system for mechanical rotating parts based on size model collaboration, comprising: The first model prediction module is used to acquire the current operating status signal of the rotating mechanical component, input the operating status signal into the pre-trained large-scale language model LLM, and obtain the large model fault prediction result.

[0052] The second model prediction module is used to collect the original vibration and acoustic emission signals of the current rotating mechanical parts, input the original vibration and acoustic emission signals into the convolutional neural network, and output the original features; encode the large model fault prediction results into continuous condition vectors, generate scaling and offset parameters based on the condition vectors, modulate the original features based on the scaling and offset parameters, and process the modulated original features using a fully connected network to obtain the small model fault prediction results.

[0053] The optimized output module is used to determine the best fault prediction result when the fault prediction results of the large model and the small model are inconsistent, by using an error judgment mechanism.

[0054] The modules in the aforementioned fault diagnosis system for mechanical rotating parts based on large and small model collaboration can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0055] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps in an embodiment of a method for fault diagnosis of mechanical rotating parts based on size model coordination. Specific implementation methods can be found in the method embodiments, and will not be repeated here.

[0056] Furthermore, the present invention also provides a non-transitory computer-readable storage medium containing instructions on which a computer program is stored. For example, a memory containing instructions that can be executed by a processor of a computer device to perform the aforementioned method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc. When the computer program is executed by the processor, it can implement the steps in an embodiment of a method for diagnosing faults in mechanical rotating parts based on size-model coordination. Specific implementation methods can be found in the method embodiments, which will not be repeated here.

[0057] Those skilled in the art will understand that embodiments of the present invention can provide methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0058] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0059] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0060] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0061] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail in this specification and embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered within the protection scope of the present invention patent. No reference numerals in the claims should be construed as limiting the scope of the claims. Any simple variations or equivalent substitutions of technical solutions that can be readily obtained by those skilled in the art within the scope of the technology disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A fault diagnosis method for mechanical rotating parts based on size model collaboration, characterized in that, Includes the following steps: The current operating status signal of the rotating mechanical component is obtained, and the operating status signal is input into a pre-trained large-scale language model (LLM) to obtain the fault prediction result of the large model. The original vibration and acoustic emission signals of the currently rotating mechanical parts are collected, and the original vibration and acoustic emission signals are input into a convolutional neural network to output the original features; The large model fault prediction result is encoded into a continuous conditional vector. Scaling and offset parameters are generated based on the conditional vector. The original features are modulated based on the scaling and offset parameters. The modulated original features are processed using a fully connected network to obtain the small model fault prediction result. When the fault prediction results of the large model are inconsistent with those of the small model, an error judgment mechanism is used to make a judgment and output the best fault prediction result. The error judgment mechanism specifically includes: obtaining the confidence levels of the large-scale language model and the convolutional neural network; setting high confidence thresholds and low confidence thresholds; if the confidence level of one side is greater than the high confidence threshold and the difference between the confidence levels of the other side is greater than a set value, then the low confidence side is initially judged to be wrong; parsing the inference chain of the large-scale language model LLM and extracting the key feature descriptions that the LLM depends on; using an attribution algorithm to visualize the actual signal feature regions that the small model's decision focuses on; if the key feature descriptions that the LLM depends on are highly consistent with the actual signal feature regions, but the results conflict, then the initial judgment that the low confidence side is wrong is supported; if the differences in feature focus points are significant, then the model result that is more consistent with the physical mechanism is adopted; and outputting the best fault prediction result.

2. The method for fault diagnosis of mechanical rotating parts based on size model collaboration according to claim 1, characterized in that, Structured data that correlates industrial equipment with fault characteristics is extracted based on industrial equipment data; The large-scale language model LLM is trained based on the structured data to obtain a pre-trained large-scale language model LLM.

3. The method for fault diagnosis of mechanical rotating parts based on size model collaboration according to claim 1, characterized in that, It also includes using incremental learning and adaptive optimization strategies to dynamically adjust the parameters of the convolutional neural network if the convolutional neural network makes a mistake.

4. The method for fault diagnosis of mechanical rotating parts based on size model collaboration according to claim 1, characterized in that, We utilize the GPT-4 large language model to extract structured data that correlates entity information with fault characteristics based on industrial equipment data.

5. The method for fault diagnosis of mechanical rotating parts based on size model collaboration according to claim 1, characterized in that, It also includes performing noise reduction and normalization preprocessing on the original vibration and acoustic emission signals before processing them by the convolutional neural network, converting them into a two-dimensional time-frequency graph through short-time Fourier transform, and inputting the two-dimensional time-frequency graph into the convolutional neural network.

6. A fault diagnosis system for mechanical rotating parts based on large-scale model collaboration, characterized in that, include: The first model prediction module is used to acquire the current operating status signal of the rotating mechanical component, input the operating status signal into the pre-trained large-scale language model LLM, and obtain the large model fault prediction result. The second model prediction module is used to collect the original vibration and acoustic emission signals of the current mechanical rotating parts, input the original vibration and acoustic emission signals into the convolutional neural network, and output the original features; The large model fault prediction result is encoded into a continuous conditional vector. Scaling and offset parameters are generated based on the conditional vector. The original features are modulated based on the scaling and offset parameters. The modulated original features are processed using a fully connected network to obtain the small model fault prediction result. The optimized output module is used to make a judgment using an error judgment mechanism when the fault prediction results of the large model and the fault prediction results of the small model are inconsistent, and output the best fault prediction result. The error judgment mechanism specifically includes: obtaining the confidence levels of the large-scale language model and the convolutional neural network; setting high confidence thresholds and low confidence thresholds; if the confidence level of one side is greater than the high confidence threshold and the difference between the confidence levels of the other side is greater than a set value, then the low confidence side is initially judged to be wrong; parsing the inference chain of the large-scale language model LLM and extracting the key feature descriptions that the LLM depends on; using an attribution algorithm to visualize the actual signal feature regions that the small model's decision focuses on; if the key feature descriptions that the LLM depends on are highly consistent with the actual signal feature regions, but the results conflict, then the initial judgment that the low confidence side is wrong is supported; if the differences in feature focus points are significant, then the model result that is more consistent with the physical mechanism is adopted; and outputting the best fault prediction result.

7. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is loaded by the processor, it is able to perform the steps of the method according to any one of claims 1 to 5.

Images