Fine adjustment method and system for infrared fault diagnosis multi-mode large model of power equipment and medium
By constructing a knowledge base for infrared fault diagnosis of power equipment and fine-tuning the training of a multimodal large model, simulated fault samples are generated, solving the problems of sample scarcity and high annotation costs in infrared diagnosis of power equipment, and achieving efficient and standardized fault diagnosis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-16
- Publication Date
- 2026-03-27
AI Technical Summary
Existing infrared fault diagnosis methods for power equipment face the problems of scarce fault samples and high annotation costs, resulting in insufficient model training and insufficient diagnostic accuracy.
By constructing a knowledge base for infrared fault diagnosis of power equipment, generating simulated fault samples, and using a multimodal large model for standardized processing and fine-tuning training, a fine-tuned multimodal large model is generated to achieve automated and standardized fault diagnosis.
It effectively solves the problem of scarce fault samples, reduces annotation costs, and improves the accuracy and standardization of diagnosis, making it suitable for specialized power infrared diagnostics.
Smart Images
Figure CN121746845A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of smart grid technology, specifically to a method, system, and medium for fine-tuning a large-scale multimodal infrared diagnostic model of power equipment. Background Technology
[0002] With the rapid development of smart grids and the increasing complexity of power equipment, the monitoring of power equipment operating status and fault diagnosis have become particularly important. Infrared thermal imaging technology, due to its advantages such as non-contact operation, visualization, and fault location, has been widely used in power equipment condition monitoring and fault diagnosis. By analyzing the temperature distribution characteristics of infrared images of power equipment, abnormal phenomena such as overheating can be effectively detected, thereby preventing equipment failures and ensuring the safe and stable operation of the power grid.
[0003] Existing technologies for equipment fault detection include target detection and segmentation techniques, which require a certain number of samples to train the model; and generative models such as Generative Adversarial Networks (GANs) to create fault samples, but the realism and physical plausibility of the generated results are difficult to guarantee. Meanwhile, although generalized large language models (LLMs) or visual language models (VLMs) exhibit strong cognitive capabilities, their direct application in the highly specialized field of power infrared diagnostics can lead to "illusions" due to a lack of specific domain knowledge, resulting in diagnostic conclusions that do not conform to industry standards or are even erroneous, thus limiting their practicality.
[0004] Therefore, current deep learning-based infrared fault diagnosis methods for power equipment face two major challenges:
[0005] 1. Scarcity of Fault Samples: Power equipment typically operates under normal conditions, with a low probability of failure, resulting in an extremely limited number of fault samples that can be collected. Deep learning models require a large amount of labeled data for training; insufficient samples can lead to overfitting, poor generalization ability, and difficulty in effective practical application.
[0006] 2. High annotation costs and standardization issues: The diagnosis of infrared images requires specialized knowledge and typically relies on experienced experts to manually interpret and annotate them according to industry standards. This process is time-consuming, labor-intensive, and costly, and is easily affected by subjective factors, making it difficult to guarantee the consistency and accuracy of the annotations. Summary of the Invention
[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, system and medium for fine-tuning a multimodal large model for infrared fault diagnosis of power equipment, thereby improving the accuracy and reliability of fault detection.
[0008] To achieve the above objectives, in a first aspect, the present invention provides a method for fine-tuning a multimodal large model for infrared fault diagnosis of power equipment, the method comprising:
[0009] S1. Collect power parameters of power equipment and build a knowledge base for infrared fault diagnosis of power equipment;
[0010] S2. Collect and preprocess the operating data of power equipment under normal conditions to generate a normal sample dataset;
[0011] S3. Based on the normal sample dataset, select a multimodal large model and edit the large model according to the image to generate a simulated fault sample dataset.
[0012] S4. Based on the power equipment infrared fault diagnosis knowledge base, the normal sample dataset, and the simulated fault sample dataset, the standardized sample dataset is obtained by standardizing the data using the multimodal large model.
[0013] S5. Based on the standardized sample dataset, fine-tune the multimodal large model to generate a fine-tuned multimodal large model;
[0014] S6. Input the real-time infrared data of the power equipment into the fine-tuned multimodal large model for processing to generate infrared diagnostic results of the power equipment.
[0015] In a further implementation, in step S1, the power parameters include temperature parameters, fault type, fault diagnosis steps, analysis logic, and recommended handling measures under normal and fault conditions of the power equipment.
[0016] In a further implementation, step S2 specifically includes: collecting infrared image data of power equipment under normal conditions, annotating the infrared images, and annotating the equipment name and the type and location of possible faults on the infrared images.
[0017] In a further implementation, rectangular boxes are used to mark the locations where malfunctions may occur on the infrared image.
[0018] In a further implementation, step S3 specifically includes: generating infrared image data of the power equipment under normal conditions based on the acquired infrared image data of the power equipment and the types of faults that may occur on the infrared image, simulating a fault infrared image, and adding and / or modifying temperature parameters on the simulated fault infrared image.
[0019] In a further implementation, step S4 specifically includes:
[0020] S41. Based on the fault type corresponding to the normal sample dataset and the simulated fault sample dataset, retrieve the professional knowledge corresponding to the power equipment infrared fault diagnosis knowledge base to obtain the professional knowledge corresponding to the fault type; the professional knowledge includes fault diagnosis steps, analysis logic and recommended handling measures.
[0021] S42. Based on the professional knowledge corresponding to the fault types in the normal sample dataset and the simulated fault sample dataset, construct structured prompt words;
[0022] S43. Input the structured prompt words into the multimodal large model to generate a standardized sample dataset. The standardized sample dataset includes simulated fault infrared images, fault types, temperature parameters, and corresponding fault diagnosis steps, analysis logic, and recommended processing measures.
[0023] In a further implementation, step S5 employs LoRA (Low-Rank Adaptation) technology.
[0024] In a further implementation, step S5 specifically includes:
[0025] S51. Load the multimodal large model and initialize the model parameters;
[0026] S52. Train the standardized sample dataset in batches, and perform forward propagation and loss calculation.
[0027] S53. Update the model parameters using the optimizer based on the loss value;
[0028] S54. Repeat steps S51 to S53 to complete iterative training, save model parameters, and generate a fine-tuned multimodal large model.
[0029] In a further implementation, the electrical equipment is one or more of bushings, surge arresters, insulators, and circuit breakers.
[0030] Secondly, the present invention provides a fine-tuning system for a multimodal large-scale infrared fault diagnosis model of power equipment, the system comprising:
[0031] The knowledge base construction module is used to collect power parameters of power equipment and build an infrared power equipment fault diagnosis knowledge base.
[0032] The data preprocessing module is used to collect infrared images of power equipment under normal conditions and preprocess them to generate a normal sample dataset.
[0033] The data simulation processing module selects a multimodal large model based on the normal sample dataset and edits the large model according to the image to generate a simulated fault sample dataset.
[0034] The data standardization processing module, based on the infrared power equipment fault diagnosis knowledge base, the normal sample dataset, and the simulated fault sample dataset, performs standardization processing through the multimodal large model to obtain a standardized sample dataset.
[0035] The model building module fine-tunes the multimodal large model based on the standardized sample dataset to generate a fine-tuned multimodal large model.
[0036] The fault diagnosis module inputs real-time infrared data of the power equipment into the fine-tuned multimodal large model for processing, and generates infrared diagnostic results for the power equipment.
[0037] Thirdly, the present invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-described method.
[0038] In summary, the present invention has the following advantages compared with the prior art:
[0039] (1) Efficiently solve the problem of sample scarcity: Use multimodal large models to simulate faults on a large number of normal samples to generate high-quality and diverse simulated fault samples, breaking through the bottleneck of difficult collection of real fault samples.
[0040] (2) Significantly reduce data processing costs: Through the “simple annotation + knowledge-enhanced large model automatic annotation” mode, only simple title annotation is required, realizing the automation and intelligence of the annotation process and significantly improving efficiency.
[0041] (3) Improve the standardization and accuracy of diagnosis: By constructing a knowledge base for infrared fault diagnosis of power equipment and using the training process of fine-tuning the multimodal large model, it is ensured that all generated sample labels meet industry standards, the diagnostic logic is clear and the basis is sufficient, and the problems of excessive subjectivity and inconsistent standards in traditional methods are effectively avoided.
[0042] (4) The method is highly versatile: The method of the present invention can be extended to other industrial fault diagnosis scenarios that rely on professional standards and visual inspection, and has high promotion value. Attached Figure Description
[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, 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.
[0044] Figure 1 This invention provides a method for fine-tuning a large multimodal infrared fault diagnosis model for power equipment.
[0045] Figure 2A schematic diagram of the modules for fine-tuning the multimodal large model of infrared fault diagnosis for power equipment provided in an embodiment of the present invention; Detailed Implementation
[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0047] This invention proposes a fine-tuning method, system, and medium for a multimodal large-scale model of infrared fault diagnosis for power equipment. By combining image editing large-scale models and knowledge-enhanced multimodal large-scale models, the invention effectively solves the problems of scarce fault samples and annotation, and realizes automated and standardized diagnosis of infrared power equipment faults.
[0048] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0049] In one exemplary embodiment, the present invention provides a method for fine-tuning a multimodal large model for infrared fault diagnosis of power equipment. In this embodiment, as shown... Figure 1 As shown, the method includes:
[0050] Step S1: Collect power parameters of power equipment and build a knowledge base for infrared fault diagnosis of power equipment.
[0051] The power parameters include temperature parameters, fault types, fault diagnosis steps, analysis logic, and recommended handling measures under normal and fault conditions of the power equipment.
[0052] The knowledge base includes, but is not limited to:
[0053] (1) Temperature limits, temperature rise criteria, relative temperature difference criteria, etc. of various power equipment (such as bushings, surge arresters, insulators, circuit breakers, etc.) under normal and fault conditions.
[0054] (2) Description of infrared image features of different fault types (such as poor contact, internal moisture, abnormal medium loss, oil shortage, etc.).
[0055] (3) Fault diagnosis steps, analysis logic and recommended handling measures.
[0056] The above knowledge base content can be derived from industry specifications or standards such as DL / T 664 Application Specification for Infrared Diagnosis of Live Equipment and Typical Examples of Infrared Power Equipment Faults.
[0057] Step S2: Collect and preprocess the operating data of power equipment under normal conditions to generate a normal sample dataset;
[0058] Collect a large number of infrared images of electrical equipment under normal conditions. Perform simple annotations on these normal samples, including two parts:
[0059] (1) Equipment name, such as "casing head", etc.
[0060] (2) Typical fault types that may occur (e.g., "poor contact and overheating of the casing head") and their locations on the image, with the locations marked in the form of rectangular boxes.
[0061] Since the workload of simple annotation is far less than that of fine annotation, it is intended to provide guidance for subsequent fault simulation and knowledge retrieval.
[0062] Step S3: Based on the normal sample dataset, select a multimodal large model, and edit the large model according to the image to generate a simulated fault sample dataset.
[0063] Select a multimodal large model such as Qwen, and use the normal sample with the title marked in step S2 as the base image. Generate the corresponding preliminary simulation sample according to the fault type indicated by the label.
[0064] Specifically, the input is a normal image of a "sleeve head" and the prompt "Simulates localized overheating of the sleeve head due to poor contact; other positions remain unchanged." Position A is the rectangular frame marked S2. Based on its understanding of the laws of the physical world, the image editing model can generate an abnormally heated area conforming to heat conduction characteristics at the corresponding position in the normal image, thus obtaining a visually realistic fault image. Key temperature data information is then added or corrected to the generated simulated image. This temperature data includes, but is not limited to, hotspot temperature values, reference body temperature values, and ambient temperature values.
[0065] Step S4: Based on the power equipment infrared fault diagnosis knowledge base, the normal sample dataset, and the simulated fault sample dataset, the standardized sample dataset is obtained by standardizing the data using the multimodal large model.
[0066] Specifically, the following steps are included:
[0067] Step S41: Based on the fault types corresponding to the normal sample dataset and the simulated fault sample dataset, retrieve the professional knowledge corresponding to the power equipment infrared fault diagnosis knowledge base to obtain the relevant professional knowledge. This professional knowledge includes fault diagnosis steps, analysis logic, and recommended handling measures.
[0068] Specifically, based on the description of the fault type to be matched in the normal sample dataset and the simulated fault sample dataset (such as "poor contact and overheating of the sleeve head"), a search is performed in the knowledge base constructed in step S1 to obtain the professional knowledge corresponding to the fault type. The professional knowledge includes fault diagnosis steps, analysis logic, and recommended treatment measures.
[0069] Step S42: Construct structured prompt words based on the professional knowledge corresponding to the fault types in the normal sample dataset and the simulated fault sample dataset.
[0070] Specifically, the retrieved professional knowledge, the normal sample dataset, and the simulated fault sample dataset (e.g., detailed text descriptions, such as image classification, temperature matrix data, etc.) are combined to construct a structured prompt word according to the preset annotation task requirements.
[0071] For example, the label indicates a power equipment fault type: poor contact and overheating of the bushing head. The retrieved professional knowledge defines bushing equipment defects as follows: Emergency defect: hot spot temperature > 80℃ or δ ≥ 95% and hot spot temperature > 55℃; Severe defect: 55℃ ≤ hot spot temperature ≤ 80℃ or δ ≥ 80% but hot spot temperature does not reach the emergency defect temperature value; General defect: δ ≥ 35% but hot spot temperature does not reach the severe defect temperature value. Here, δ = (ΔT1 / ΔT2) × 100% = [(T1 - T2) / (T1 - T0)] × 100%, where T1 is the fault point temperature, T2 is the normal point temperature, and T0 is the ambient temperature. The infrared image is then analyzed based on these defect definitions.
[0072] Step S43: Input the structured prompt words into the multimodal large model to generate a standardized sample dataset. The standardized sample dataset includes simulated fault infrared images, fault types, temperature parameters, and corresponding fault diagnosis steps, analysis logic, and recommended processing measures.
[0073] Specifically, the constructed prompts are input into the large model, which is then tasked with acting as a power equipment diagnostic expert, analyzing images according to the power equipment infrared fault diagnosis knowledge base. The annotations output by the multimodal large model should include: the location of the faulty equipment, the fault type, the highest temperature, the ambient temperature, the reference temperature, the relative temperature difference, the defect level (e.g., general defect, severe defect, critical defect), as well as the diagnostic basis and recommendations.
[0074] Step S5: Fine-tune the multimodal large model based on the standardized sample dataset to generate a fine-tuned multimodal large model;
[0075] Using the standardized sample dataset (including the normal sample dataset and the simulated fault sample dataset) generated in step S4, the parameters of the multimodal large model are efficiently fine-tuned to generate a fine-tuned multimodal large model.
[0076] By employing LoRA (Low-Rank Adaptation) technology, without altering the main parameters of the original model, a low-rank adapter module is introduced, enabling the model to quickly adapt to the professional field of infrared power equipment fault diagnosis.
[0077] Specifically, the following steps are included:
[0078] S51: Load the multimodal large model and initialize the model parameters.
[0079] During initialization, all weight parameters of the large multimodal model (such as the Qwen model) are loaded, denoted as . This value is then set to a frozen state, meaning it remains unchanged during subsequent training. LoRA adapters are added in parallel alongside the linear layers (query, key, value, output projection layers, etc.) in the self-attention mechanism of the base model. The LoRA adapter is a pair of low-rank matrices. and And initialize it, where k, d, r represent the matrix dimensions, and the low-rank matrix {A, B} is the trainable state.
[0080] Step S52: Train the standardized sample dataset in batches, and perform forward propagation and loss calculation;
[0081] The standardized sample dataset is divided into several training batches. First, the forward propagation of the base model is performed. In each linear layer injected with the LoRA adapter, the actual computation is as follows:
[0082] Let the input matrix of this layer be h, then the output of this layer is y = Wh + BAh. The final output of the model and the target labels from the training data are fed into the loss function to calculate the loss value.
[0083] Step S53: Update the model parameters using the optimizer based on the loss value.
[0084] Calculate the gradient of the loss function with respect to all trainable parameters (i.e., the parameters of the low-rank matrix {A,B} of all LoRA adapters). Update the matrix parameters of the LoRA adapters using an optimizer (e.g., AdamW) based on the calculated gradient.
[0085] Step S54: Repeat steps S51 to S53 to complete iterative training, save model parameters, and generate a fine-tuned multimodal large model.
[0086] Repeat steps S51 to S53 until the model's performance on the validation set no longer improves or the predetermined number of training epochs is reached. After training, save only the trained LoRA adapter weights, i.e., the matrix parameters of all low-rank matrices, to generate a fine-tuned multimodal large model.
[0087] After fine-tuning, this fine-tuned multimodal large model can understand the complex mapping relationship between infrared image features and text diagnostic specifications end-to-end, enabling accurate and automatic identification and diagnosis of power equipment faults.
[0088] S6. Input the real-time infrared data of the power equipment into the fine-tuned multimodal large model for processing to generate infrared diagnostic results of the power equipment.
[0089] Specifically, the system acquires infrared images of the power equipment in operation, inputs them into a fine-tuned multimodal large model for analysis, identifies and judges faults in the power equipment, obtains analysis logic, diagnostic results, and recommended handling measures, and provides feedback to power equipment inspection personnel for reference.
[0090] In one exemplary embodiment, such as Figure 2 As shown, this invention provides a fine-tuning system for a multimodal large-scale infrared fault diagnosis model of power equipment, the system comprising:
[0091] The knowledge base construction module 201 is used to collect power parameters of power equipment and build an infrared power equipment fault diagnosis knowledge base.
[0092] Data preprocessing module 202 is used to collect infrared images of power equipment under normal conditions and perform preprocessing to generate a normal sample dataset;
[0093] The data simulation processing module 203 selects a multimodal large model based on the normal sample dataset and edits the large model according to the image to generate a simulated fault sample dataset.
[0094] The data standardization processing module 204, based on the infrared power equipment fault diagnosis knowledge base, the normal sample dataset, and the simulated fault sample dataset, performs standardization processing through the multimodal large model to obtain a standardized sample dataset.
[0095] The model building module 205 fine-tunes the multimodal large language model based on the standardized sample dataset to generate a fine-tuned multimodal large model.
[0096] The fault diagnosis module 206 inputs the real-time operating infrared data of the power equipment into the fine-tuned multimodal large model for processing, and generates infrared diagnostic results for the power equipment.
[0097] In one exemplary embodiment, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.
[0098] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., SSD), etc.
[0099] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed, it can include the processes of the embodiments of the above methods.
[0100] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for fine-tuning a multimodal large-scale model for infrared fault diagnosis of power equipment, characterized in that, The method includes: S1. Collect power parameters of power equipment and build a knowledge base for infrared fault diagnosis of power equipment; S2. Collect and preprocess the operating data of power equipment under normal conditions to generate a normal sample dataset; S3. Based on the normal sample dataset, select a multimodal large model, and edit the large model according to the image to generate a simulated fault sample dataset; S4. Based on the power equipment infrared fault diagnosis knowledge base, the normal sample dataset, and the simulated fault sample dataset, the standardized sample dataset is obtained by standardizing the data using the multimodal large model. S5. Based on the standardized sample dataset, fine-tune the multimodal large model to generate a fine-tuned multimodal large model; S6. Input the real-time infrared data of the power equipment into the fine-tuned multimodal large model for processing to generate infrared diagnostic results of the power equipment.
2. The fine-tuning method for a multimodal large model of infrared fault diagnosis of power equipment as described in claim 1, characterized in that: In step S1, the power parameters include temperature parameters under normal and fault conditions, fault type, fault diagnosis steps, analysis logic, and recommended handling measures.
3. The fine-tuning method for a multimodal large model of infrared fault diagnosis of power equipment as described in claim 1, characterized in that: Step S2 specifically includes: collecting infrared image data of power equipment under normal conditions, annotating the infrared images, and annotating the equipment name and the type and location of possible faults on the infrared images.
4. The fine-tuning method for a multimodal large model of infrared fault diagnosis of power equipment as described in claim 3, characterized in that: Use rectangles to mark the locations of potential faults on the infrared image.
5. The fine-tuning method for a multimodal large model of infrared fault diagnosis of power equipment as described in claim 3, characterized in that: Step S3 specifically includes: generating infrared image data of the power equipment under normal conditions based on the obtained infrared image data of the power equipment and the types of faults that may occur on the infrared image, simulating a fault infrared image, and adding and / or modifying temperature parameters on the simulated fault infrared image.
6. The fine-tuning method for a multimodal large model of infrared fault diagnosis of power equipment as described in claim 5, characterized in that: Step S4 specifically includes: S41. Based on the fault type corresponding to the normal sample dataset and the simulated fault sample dataset, retrieve the professional knowledge corresponding to the power equipment infrared fault diagnosis knowledge base to obtain the professional knowledge corresponding to the fault type; the professional knowledge includes fault diagnosis steps, analysis logic and recommended handling measures. S42. Based on the professional knowledge corresponding to the fault types in the normal sample dataset and the simulated fault sample dataset, construct structured prompt words; S43. Input the structured prompt words into the multimodal large model to generate a standardized sample dataset. The standardized sample dataset includes simulated fault infrared images, fault types, temperature parameters, and corresponding fault diagnosis steps, analysis logic, and recommended processing measures.
7. The fine-tuning method for a multimodal large model of infrared fault diagnosis of power equipment as described in claim 1, characterized in that: Step S5 specifically includes: S51. Load the multimodal large model and initialize the model parameters; S52. Train the standardized sample dataset in batches, and perform forward propagation and loss calculation. S53. Update the model parameters using the optimizer based on the loss value; S54. Repeat steps S51 to S53 to complete iterative training, save model parameters, and generate a fine-tuned multimodal large model.
8. The fine-tuning method for a multimodal large model for infrared fault diagnosis of power equipment as described in any one of claims 1-7, characterized in that: The electrical equipment is one or more of bushings, surge arresters, insulators, and circuit breakers.
9. A fine-tuning system for a multimodal large-scale infrared fault diagnosis model of power equipment, characterized in that, The system includes: The knowledge base construction module is used to collect power parameters of power equipment and build an infrared power equipment fault diagnosis knowledge base. The data preprocessing module is used to collect infrared images of power equipment under normal conditions and preprocess them to generate a normal sample dataset. The data simulation processing module selects a multimodal large model based on the normal sample dataset, and edits the large model according to the image to generate a simulated fault sample dataset. The data standardization processing module, based on the infrared power equipment fault diagnosis knowledge base, the normal sample dataset, and the simulated fault sample dataset, performs standardization processing through the multimodal large model to obtain a standardized sample dataset. The model building module fine-tunes the multimodal large model based on the standardized sample dataset to generate a fine-tuned multimodal large model. The fault diagnosis module inputs real-time infrared data of the power equipment into the fine-tuned multimodal large model for processing, and generates infrared diagnostic results for the power equipment.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the method as described in any one of claims 1 to 8.