Stator core defect diagnosis method, system and equipment and storage medium

Abstract

The invention discloses a stator core defect diagnosis method, system and device and a storage medium, and the method comprises the following steps: constructing a cross-modal neural network model which comprises a thermal modal branch and an electromagnetic modal branch, and fusing the output results of the two types of modals through a cross attention mechanism; obtaining virtual simulation data and actual measurement data of the stator core, and carrying out transfer learning on the cross-modal neural network model to adapt to parameters of the cross-modal neural network model; infrared thermal imaging data and ELCID electromagnetic signal data of the stator core are obtained, and the two kinds of data are preprocessed; and respectively inputting the two types of preprocessed data into the thermal mode branch and the electromagnetic mode branch in the adaptive cross-mode neural network model, and fusing through a cross attention mechanism to obtain a diagnosed defect result. Comprehensive diagnosis of stator core defects is realized, and the diagnosis precision is improved.

Description

Technical Field

[0001] The present invention belongs to the field of power equipment detection, and relates to a stator core defect diagnosis method, system, device and storage medium. Background Art

[0002] Traditional stator core defect detection methods mainly rely on infrared thermal imaging (monitoring temperature rise) and the ELCID method (detecting flux distortion).

[0003] The essence of the infrared thermal imaging method is to apply a power frequency or different frequency excitation current on the stator core to simulate the electromagnetic field environment of the actual operating condition, use an infrared thermal imager (thermal sensitivity ≤ 0.05 °C) to capture the temperature distribution on the core surface, generate a thermal image, and indirectly diagnose internal defects based on the abnormal temperature distribution on the stator core surface. Its principle is based on the following physical processes: 1) Local energy loss caused by defects: When there are defects such as insulation breakage and lamination short circuit in the stator core, the alternating magnetic field will induce abnormal eddy currents in the defect area, resulting in a significant increase in Joule heat.

[0004] 2) Heat conduction and surface temperature rise: The local heat is conducted to the core surface through the silicon steel sheets, forming a temperature gradient. The surface temperature of the defect area is usually higher than that of the normal area (the temperature rise ΔT can reach several °C to dozens of °C).

[0005] The ELCID method generates a power frequency alternating magnetic flux in the stator core through an excitation coil, and uses a detection coil (such as a Chattock coil) to detect the magnetic potential difference on the core surface. When there are lamination short circuits or insulation defects in the core, the local eddy current loss increases, resulting in abnormal magnetic potential differences, thereby locating the defects.

[0006] However, the above two methods still have the following problems: Strong dependence on manual operation: It requires operators to combine thermal images and electromagnetic signal experience to judge the defect type, and the quantization accuracy is low.

[0007] High miss rate of small defects: A single method (thermal or electromagnetic) has insufficient sensitivity for diagnosing early insulation breakage and local short circuits. Summary of the Invention

[0008] The purpose of the present invention is to overcome the above-mentioned disadvantages of the prior art, and provide a stator core defect diagnosis method, system, device and storage medium, which combines the advantages of the infrared thermal imaging method and the ELCID method, realizes the comprehensive diagnosis of stator core defects, and improves the diagnosis accuracy.

[0009] To achieve the above purpose, the present invention adopts the following technical solutions: A stator core defect diagnosis method includes the following processes: Construct a cross-modal neural network model. The cross-modal neural network model includes a thermal modality branch and an electromagnetic modality branch, and uses a cross-attention mechanism to fuse the output results of the two modalities; Obtain the virtual simulation data and measured data of the stator core, and perform transfer learning on the cross-modal neural network model to adapt the parameters of the cross-modal neural network model; Obtain the infrared thermal imaging data and ELCID electromagnetic signal data of the stator core, and preprocess the two types of data; Input the two preprocessed data into the thermal modality branch and the electromagnetic modality branch in the adapted cross-modal neural network model respectively. After fusion by the cross-attention mechanism, the diagnosed defect results are obtained.

[0010] Preferably, the thermal modality branch uses a channel-expanded ResNet-18 network, with the input being a thermal image and polar coordinate parameters, and extracts thermal image features; the electromagnetic modality branch includes two one-dimensional convolutional layers. The first layer is used to extract the electromagnetic features between adjacent slots, and the second layer is used to extract periodic abnormal patterns; through the cross-attention mechanism, the electromagnetic features are used as query vectors to dynamically weight the spatial features of the thermal image.

[0011] Preferably, during the fusion process of the output results of the two modalities: perform loss function processing with physical constraints on the fusion results.

[0012] Preferably, the transfer learning process is: train the cross-modal neural network model based on the virtual simulation data, adjust the simulation parameters to match the actual physical parameters of the stator core, and collect the measured data to optimize the output layer parameters of the cross-modal neural network model.

[0013] Preferably, the preprocessing process of the infrared thermal imaging data is: perform denoising, histogram equalization, cropping, and temperature value normalization on the infrared thermal image, and append polar coordinate position encoding.

[0014] Preferably, the preprocessing process of the ELCID electromagnetic signal data is: map the ELCID electromagnetic signal data to a polar coordinate space sequence according to the slot number, and generate an electromagnetic feature matrix containing amplitude, phase, and complex voltage.

[0015] Preferably, during the diagnosis process of the stator core, divide the stator core into multiple fan-shaped regions to independently obtain data; convert the local coordinates to global polar coordinates and merge the results of adjacent regions; verify the authenticity of the defect through the spatial matching of the ELCID electromagnetic amplitude abnormal region and the thermal map temperature rise region; retain the defect results with a confidence level higher than the preset threshold.

[0016] A stator core defect diagnosis system, comprising: A model construction module, which is used to construct a cross-modal neural network model. The cross-modal neural network model includes a thermal modality branch and an electromagnetic modality branch, and uses a cross-attention mechanism to fuse the output results of the two modalities; A transfer learning module, which is used to obtain virtual simulation data and measured data of the stator core, and perform transfer learning on the cross-modal neural network model to adapt the parameters of the cross-modal neural network model; A data acquisition module, which is used to obtain infrared thermal imaging data and ELCID electromagnetic signal data of the stator core, and preprocess the two types of data; A defect diagnosis module, which is used to respectively input the two preprocessed types of data into the thermal modality branch and the electromagnetic modality branch in the adapted cross-modal neural network model. After being fused by the cross-attention mechanism, the diagnosed defect results are obtained.

[0017] A computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the stator core defect diagnosis method are implemented.

[0018] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the steps of the stator core defect diagnosis method are implemented.

[0019] Compared with the prior art, the present invention has the following beneficial effects: The present invention combines infrared thermal imaging and ELCID electromagnetic signals through deep learning and physical mechanisms, breaks through the bottleneck of single-modal detection, realizes automatic defect location, classification, and severity quantification, realizes comprehensive diagnosis of stator core defects, and improves the diagnosis accuracy. And it supports area scanning and position coding, adapting to the detection requirements of large equipment. The pre-training data can strictly cover common defect types and boundary conditions in engineering, ensuring the generalization of the model. It has a fast transfer ability and realizes fast cross-unit adaptation with a small amount of data through virtual calibration and on-site adaptation. Description of the Drawings

[0020] Figure 1 It is a flowchart of the stator core defect diagnosis method of the present invention. Detailed Embodiments

[0021] The embodiments described below by referring to the drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation of the present invention.

[0022] The following disclosure provides many different embodiments or examples for implementing different aspects of the present invention. To simplify the disclosure of the present invention, the components and settings of specific examples are described below. Of course, they are merely examples and are not intended to limit the present invention. In addition, the present invention may repeat reference numerals and / or reference letters in different examples. This repetition is for the purpose of simplification and clarity and does not itself indicate the relationship between the various embodiments and / or settings discussed.

[0023] Embodiment 1: This embodiment provides a method for diagnosing stator core defects, including the following processes: Construct a cross-modal neural network model. The cross-modal neural network model includes a thermal modality branch and an electromagnetic modality branch, and uses a cross-attention mechanism to fuse the output results of the two modalities.

[0024] Obtain the virtual simulation data and measured data of the stator core, and perform transfer learning on the cross-modal neural network model to adapt the parameters of the cross-modal neural network model.

[0025] Obtain the infrared thermal imaging data and ELCID electromagnetic signal data of the stator core, and preprocess the two types of data.

[0026] Input the preprocessed two types of data into the thermal modality branch and the electromagnetic modality branch of the adapted cross-modal neural network model respectively. After fusion by the cross-attention mechanism, the diagnosed defect results are obtained.

[0027] Embodiment 2: As Figure 1 shown, this embodiment provides an intelligent diagnosis method for stator core defects based on infrared-electromagnetic data fusion, including the following processes: 1. Multi-modal data acquisition and preprocessing: The goal of this step is to convert infrared thermal imaging and ELCID signals into a format suitable for input to the neural network while retaining physical characteristics.

[0028] 1.1 Infrared thermal imaging data: Image acquisition: The temperature field distribution map (grayscale image, pixel value corresponding to temperature) collected on-site or generated by simulation, covering all or part of the stator core area (the resolution of the on-site collected picture should not be lower than 640×480 pixels, and the thermal sensitivity ≤ 0.05°C); Local processing: If the input is a local thermal map, its polar coordinate position (r,θ) in the iron core needs to be marked and input into the network through position encoding (normalized to [0,1]); Preprocessing: Gaussian filtering for noise reduction (eliminating Gaussian noise in the image) → histogram equalization (enhancing the contrast of the image by changing the gray-level distribution) → cropping to 256×256 pixels → normalizing the temperature values in the image.

[0029] 1.2 ELCID electromagnetic data: Excitation coil: Apply a sinusoidal current with a power frequency (50Hz / 60Hz) (amplitude 1 to 5A) and maintain a steady state. Detection coil: Continuously test along the inner circumference of the iron core to obtain the voltage amplitude (V) and phase (φ) of each slot. The typical data format is as follows (taking a stator with a certain number of slots N as an example):

[0030] Data processing: Data alignment: Map the slot numbers to the polar coordinates (θ) of the iron core circumference to construct a spatial distribution sequence. Example: Slot number 1 corresponds to θ = 0°, slot number 2 corresponds to θ = 360 / N°, and so on.

[0031] Feature construction: Amplitude feature: The voltage amplitude of each slot .

[0032] Phase feature: The phase difference of each slot . The reference phase can be the average value or the phase of the first slot.

[0033] Composite feature: and , which respectively represent the imaginary part and real part of the complex voltage.

[0034] Standardization: Perform Z-score standardization on each feature.

[0035] Construct an electromagnetic feature matrix: A two-dimensional matrix (number of slots × number of features). Taking a stator core with 72 slots as an example, the shape of the feature matrix is 72×4.

[0036] 2. Cross-modal neural network architecture: 2.1 Input layer: The input layer is mainly divided into two branches, namely the thermal modality branch and the electromagnetic modality branch. After processing the two branches according to the method in step 1, they are used as the input layer.

[0037] 2.2 Feature extraction module: Thermal modality branch (ResNet-18): Input layer: Change the 3-channel input of the original ResNet-18 to 3 channels (heat map + position encoding), and the first layer convolution parameters are set as:

[0038] Residual block configuration: There are 4 groups of residual blocks ([2, 2, 2, 2] layers), and the number of output channels for each group is 64, 128, 256, and 512 respectively.

[0039] Output: The output of the last convolutional layer is an 8×8×512 feature tensor.

[0040] Electromagnetic mode branch (1D-CNN, taking the stator core with 72 slots as an example): Convolutional layer design:

[0041] Conv1d(in_channels = 4, out_channels = 16, kernel_size = 3, padding = 1).

[0042] Input shape: (batch_size, 4, 72) → Output shape: (batch_size, 16, 72).

[0043] Spatial feature extraction: The first convolutional layer captures the correlation between adjacent slots (such as the amplitude mutation between slots 1 - 3).

[0044] The second convolutional layer expands the receptive field and identifies periodic patterns (such as anomalies occurring every 6 slots).

[0045] Global average pooling: Output a 32-dimensional feature vector.

[0046] 2.3 Cross-modal fusion module: Dynamically aligns the key regions of thermal images and electromagnetic signals.

[0047] Cross-attention mechanism: Uses electromagnetic features as queries (Query), thermal features as key-values (Key-Value), calculates attention weights, and performs weighted fusion.

[0048] Thermal feature reshaping: Flattens the 8×8×512 thermal features into a 64×512 matrix (64 spatial positions, with 512-dimensional features at each position).

[0049] Electromagnetic feature extension: Maps the 32-dimensional electromagnetic features to 512 dimensions through a fully connected layer as the query vector .

[0050] Attention weight calculation:

[0051] Feature weighted fusion:

[0052] In the formula, the matrix Q is the query vector, which is the eigenvector from the electromagnetic mode branch and represents the electromagnetic feature mode currently of concern; The matrix K is the eigenmatrix (spatial thermal feature) from the thermal mode branch and represents the potential area of concern in the thermal image.

[0053] V is the value matrix representing the specific feature content that needs to be weighted and fused.

[0054] b is the dimension scaling factor, and its value is the dimension of the query vector and the key vector. For example, here d = 512, which is used to prevent the dot product value from being too large.

[0055] Loss function design: The loss function needs to optimize multi-modal feature fusion, multi-task output (regression + classification), and introduce physical constraints.

[0056]

[0057] Among them 、 、 is the weight coefficient, which can be initially set to 、 、 . is the regression loss, and the goal is to optimize the defect coordinates ( ), defect size ( ), and resistance drop rate ( ). The calculation formula is

[0058] In the formula 、 、 、 is the predicted value, 、 、 、 is the actual value (from simulation or measured data).

[0059] is the classification loss, and the goal is to optimize the discrimination accuracy of the defect type. The calculation formula is

[0060] In the formula is the actual type (one-hot encoding), . It is the Softmax probability for network output.

[0061] It is the physical constraint loss. The objective is to force the prediction result to conform to physical laws (such as energy conservation, material properties, etc.). Taking the Joule heat power constraint as an example, the calculation formula is

[0062] In the formula is the defective loss power predicted by the network, is the temperature rise distribution in the thermal image, is the material thermal conductivity coefficient (which can be calibrated through simulation).

[0063] 2.4 Output layer: The output layer can perform regression and classification tasks. Through the regression task, it can predict the defective center coordinates (x, y), equivalent diameter, and insulation resistance drop rate; it can also classify different defective types (insulation breakage, lamination short circuit, local overheating, material aging) and generate confidence levels.

[0064] 3. Transfer learning and virtual calibration: 3.1 Pretraining stage: Simulation data generation: Establish a parametric iron core model in COMSOL and randomly generate 10,000 groups of defective data (covering all types and parameter ranges); General model training: Train the initial network based on the simulation data to learn the general mapping relationship between thermal - electromagnetic features and defects.

[0065] Defective types and parameter ranges (for example only):

[0066] Simulation model diversity control: Geometric parameters: Iron core diameter (3m - 10m), lamination thickness (0.3mm - 0.5mm), number of slots (48 - 96).

[0067] Material parameters: Permeability of silicon steel sheet (μ_r = 2000 - 5000), conductivity (σ = 2e6 - 6e6 S / m).

[0068] 3.2 Virtual calibration stage: Parameter adjustment (example): The iron core diameter of the target unit is 8m, and the material is M250 - 35A silicon steel sheet (μ_r = 2500, σ = 4.17e6 S / m).

[0069] Modify the geometric dimensions and material library in COMSOL to generate 100 sets of data. The defect parameters can be set according to the historical fault statistical distribution of this unit (for example, the number of short-circuited lamination layers is preferably 3-5 layers).

[0070] Noise model (optional): Infrared thermal image: Apply noise (Poisson noise (simulating photon noise) + Gaussian noise (σ = 0.1 °C)) that matches the actual infrared camera to the thermal image.

[0071] ELCID signal: Superimpose the actually recorded background noise (signal-to-noise ratio SNR = 30 dB).

[0072] Freeze all layers of ResNet-18 and the first two convolutional layers of 1D-CNN. Select the optimizer AdamW, learning rate = 5e-5, weight decay = 1e-4, batch size = 16. Use the 100 sets of data generated in this stage to train the model for 50 epochs to initially adapt to the electromagnetic response characteristics of the target unit.

[0073] 3.3 On-site adaptation stage: Then test the target unit on-site to obtain 10-20 sets of measured data. Unfreeze the last two layers of ResNet-18 and use the measured data to further train the model (optimizer: Adam, learning rate = 1e-5, batch size = 8, train for 10 epochs) to adapt to the heat conduction characteristics of the target unit.

[0074] 4. Online diagnosis and result fusion: 4.1 Sub-region detection process: Core partitioning: Divide the inner circumference of the stator core into 12 regions in 30° sectors.

[0075] Data acquisition: Sequentially take infrared thermal images of each region and perform local ELCID scans synchronously.

[0076] Independent inference: Input the data of each region into the network to obtain local defect prediction results.

[0077] Coordinate transformation: Convert the local coordinates (based on the sector center) to global polar coordinates (r, θ).

[0078] 4.2 Global result fusion rules: Conflict resolution: If adjacent regions predict a defect at the same location, select the one with the highest confidence.

[0079] ELCID verification: The regions with abnormal global ELCID amplitude and phase need to match the thermal image prediction results spatially.

[0080] Confidence threshold: Only output defects with Softmax probability ≥ 85%.

[0081] Example 3: In this embodiment, a stator core defect diagnosis system is provided. This stator core defect diagnosis system can be used to implement the above-mentioned stator core defect diagnosis method. Specifically, this stator core defect diagnosis system includes a model construction module, a transfer learning module, a data acquisition module, and a defect diagnosis module.

[0082] Among them, the load acquisition module is used to obtain the residential electricity load in real time.

[0083] The model construction module is used to construct a cross-modal neural network model. The cross-modal neural network model includes a thermal modality branch and an electromagnetic modality branch, and uses a cross-attention mechanism to fuse the output results of the two modalities.

[0084] The transfer learning module is used to obtain the virtual simulation data and measured data of the stator core, and perform transfer learning on the cross-modal neural network model to adapt the cross-modal neural network model parameters.

[0085] The data acquisition module is used to obtain the infrared thermal imaging data and ELCID electromagnetic signal data of the stator core, and preprocess the two types of data.

[0086] The defect diagnosis module is used to input the two types of preprocessed data into the thermal modality branch and the electromagnetic modality branch of the adapted cross-modal neural network model respectively. After being fused by the cross-attention mechanism, the diagnosed defect results are obtained.

[0087] Example 4: In this embodiment, a terminal device is provided. The terminal device includes a processor and a memory. The memory is used to store a computer program, and the computer program includes program instructions. The processor is used to execute the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function. The processor described in the embodiment of the present invention can be used for the operation of the stator core defect diagnosis method, including: constructing a cross-modal neural network model, where the cross-modal neural network model includes a thermal modality branch and an electromagnetic modality branch, and using a cross-attention mechanism to fuse the output results of the two modalities; obtaining virtual simulation data and measured data of the stator core, and performing transfer learning on the cross-modal neural network model to adapt the parameters of the cross-modal neural network model; obtaining infrared thermal imaging data and ELCID electromagnetic signal data of the stator core, and preprocessing the two types of data; respectively inputting the preprocessed two types of data into the thermal modality branch and the electromagnetic modality branch of the adapted cross-modal neural network model, and after fusion by the cross-attention mechanism, obtaining the diagnosed defect results.

[0088] Embodiment 5: In this embodiment, a computer-readable storage medium (Memory) is provided. The computer-readable storage medium is a memory device in the terminal device and is used to store programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and, of course, the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, and this storage space stores the operating system of the terminal. And, one or more instructions suitable for being loaded and executed by the processor are also stored in this storage space. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed Random Access Memory (RAM), or a non-volatile memory, such as at least one disk memory.

[0089] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the stator core defect diagnosis method in the above embodiments; the one or more instructions in the computer-readable storage medium are loaded and executed by the processor to perform the following steps: constructing a cross-modal neural network model, the cross-modal neural network model includes a thermal modality branch and an electromagnetic modality branch, and using a cross-attention mechanism to fuse the output results of the two modalities; obtaining virtual simulation data and measured data of the stator core, and performing transfer learning on the cross-modal neural network model to adapt the cross-modal neural network model parameters; obtaining infrared thermal imaging data and ELCID electromagnetic signal data of the stator core, and preprocessing the two types of data; respectively inputting the preprocessed two types of data into the thermal modality branch and the electromagnetic modality branch of the adapted cross-modal neural network model, and after fusion by the cross-attention mechanism, obtaining the diagnosed defect results.

[0090] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, optical storage, etc.) containing computer-usable program code.

[0091] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0092] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0093] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are executed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions for implementing the functions specified in one process or a plurality of processes and / or boxes Figure 1 one process or a plurality of processes and / or boxes Figure 1 steps for the functions specified in one box or a plurality of boxes.

[0094] In the above embodiments of the present application, the descriptions of the respective embodiments have their own emphases. For parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0095] The above are only the preferred embodiments of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

[0096] It should be understood that the above description is for the purpose of illustration and not limitation. Many embodiments and many applications other than the examples provided will be apparent to those skilled in the art upon reading the above description.

Claims

1. A method for diagnosing stator core defects, characterized in that: The process includes: Construct a cross-modal neural network model, which includes a thermal modal branch and an electromagnetic modal branch, and use a cross-attention mechanism to fuse the output results of the two modalities; Obtain virtual simulation data and measured data of the stator core, and perform transfer learning on the cross-modal neural network model to adapt the cross-modal neural network model parameters; Obtain infrared thermal imaging data and ELCID electromagnetic signal data of the stator core, and pre-process the two types of data; The two types of preprocessed data are respectively input into the thermal modal branch and the electromagnetic modal branch of the adapted cross-modal neural network model. After fusion through the cross-attention mechanism, the diagnosed defect results are obtained.

2. The stator core defect diagnosis method according to claim 1, characterized in that: The thermal modality branch uses a channel-expanded ResNet-18 network, with thermal images and polar coordinate parameters as input to extract thermal image features. The electromagnetic modality branch includes two one-dimensional convolutional layers. The first layer is used to extract electromagnetic features between adjacent slots, and the second layer is used to extract periodic abnormal patterns. Through the cross-attention mechanism, the electromagnetic features are used as query vectors to dynamically weight the thermal image spatial features.

3. The stator core defect diagnosis method according to claim 1, characterized in that: During the fusion process of the output results of the two types of modalities: the fusion results are processed by the loss function of physical constraints.

4. The stator core defect diagnosis method according to claim 1, characterized in that: The transfer learning process is as follows: training the cross-modal neural network model based on virtual simulation data, adjusting the simulation parameters to match the actual physical parameters of the stator core, and collecting measured data to optimize the output layer parameters of the cross-modal neural network model.

5. The stator core defect diagnosis method according to claim 1, characterized in that: The preprocessing process of infrared thermal imaging data is: denoising, histogram equalization, cropping and temperature value standardization of infrared thermal images, and adding polar coordinate position coding.

6. The stator core defect diagnosis method according to claim 1, characterized in that: The preprocessing process of ELCID electromagnetic signal data is as follows: mapping the ELCID electromagnetic signal data into a polar coordinate space sequence according to the slot number, and generating an electromagnetic characteristic matrix containing amplitude, phase and complex voltage.

7. The stator core defect diagnosis method according to claim 1, characterized in that: During the diagnosis of the stator core, the stator core is divided into multiple sector-shaped areas to obtain data independently; the local coordinates are converted into global polar coordinates and the results of adjacent areas are merged; the authenticity of the defect is verified by spatial matching between the ELCID electromagnetic amplitude abnormal area and the thermal map temperature rise area; and the defect results with confidence levels higher than the preset threshold are retained.

8. A stator core defect diagnosis system, characterized in that: include: The model building module is used to build a cross-modal neural network model. The cross-modal neural network model includes a thermal modal branch and an electromagnetic modal branch, and uses a cross-attention mechanism to fuse the output results of the two modalities. The transfer learning module is used to obtain virtual simulation data and measured data of the stator core, and to perform transfer learning on the cross-modal neural network model to adapt the parameters of the cross-modal neural network model; A data acquisition module is used to acquire infrared thermal imaging data and ELCID electromagnetic signal data of the stator core and pre-process the two types of data; The defect diagnosis module is used to input the two preprocessed data into the thermal modal branch and the electromagnetic modal branch in the adapted cross-modal neural network model respectively, and obtain the diagnosed defect results after fusion through the cross-attention mechanism.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the stator core defect diagnosis method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the stator core defect diagnosis method according to any one of claims 1 to 7 are implemented.

Images