Motor thermal imaging diagnosis method and system based on pulse neural network
By constructing an adaptive residual coding module and pulse residual network, the motor thermal imaging diagnosis method is solved in the existing technology of information loss and calculation complexity, and high-precision motor fault diagnosis is achieved, which improves the biointerpretation of the model and the energy consumption efficiency.
Patent Information
- Application Number
- CN202510690859.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-27
- Publication Date
- 2025-09-02
AI Technical Summary
The existing motor thermal imaging diagnostic method based on deep learning adopts convolutional neural networks, which lack interpretability and complex calculations. The real-time encoding method leads to information loss and reduces diagnostic accuracy.
Adaptive residual coding module and pulse residual network are adopted to adaptively adjust the weight coefficient of branch coding features, and combine the main and branch coding features to build a motor thermal imaging diagnostic model based on pulsed neural network to reduce information loss and improve diagnostic accuracy.
It improves the biointerpretation of the model, reduces the computational complexity and energy consumption, and improves the accuracy and information retention capabilities of motor fault diagnosis.
Smart Images
Figure CN120580487A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of intelligent diagnosis of motor faults, and in particular to a motor thermal imaging diagnosis method, system and computer-readable storage medium based on a pulse neural network. Background Art
[0002] Motors are crucial components of automated industrial manufacturing systems and transportation equipment transmission systems. Failures in motors can severely impact the economic benefits of industrial production, increase maintenance costs for transportation equipment like high-speed trains, and even threaten the lives of workers and passengers. Therefore, accurately diagnosing motor fault types is crucial.
[0003] Existing fault diagnosis methods can be broadly categorized as those based on mathematical modeling and data-driven approaches. With the development of artificial intelligence and big data technologies, data-driven approaches have garnered widespread attention. Data-driven approaches primarily include signal processing and machine learning. Signal processing and traditional machine learning methods typically require prior knowledge and theoretical understanding of statistical learning. Deep learning, on the other hand, can adaptively extract important features and autonomously learn, providing a powerful tool for fault diagnosis.
[0004] In recent years, a number of deep learning network models have emerged, such as convolutional neural networks, long short-term memory networks, generative adversarial networks, graph neural networks, and the Transformer. These models have been widely used in fields such as image recognition and intelligent fault diagnosis, achieving excellent results. However, these deep learning models are based on traditional artificial neural networks (ANNs), which are black-box models. While they offer excellent performance, they lack interpretability.
[0005] Driven by the need for biological interpretability, spiking neural networks (SNNs) have been proposed to simulate biological neural networks. SNNs not only possess strong biological interpretability, but their discrete binary nature also allows for energy savings when deployed on hardware. Early in the application of SNNs in fault diagnosis, researchers first employed a simple, single-layer SNN to diagnose rolling bearing faults. Experimental results confirmed the feasibility of SNNs for mechanical equipment fault diagnosis and their unique advantages in biological interpretability. Current research on SNNs encompasses multiple aspects, including encoding methods, training algorithms, neural models, and network architectures. Encoding methods primarily focus on efficiently encoding continuous input signals into discrete pulse signals that the network can recognize. Training algorithms primarily focus on developing training algorithms tailored to the characteristics of SNNs to improve network performance. Neuron models primarily focus on improving existing models. Finally, network architectures primarily focus on developing deep networks suitable for SNNs, drawing on existing network structures.
[0006] Existing deep learning-based motor thermal imaging diagnostic methods all employ convolutional neural networks (ANNs), which learn the features of thermal images to diagnose motor faults. While these methods offer excellent feature extraction and representation capabilities, they lack interpretability. Furthermore, convolutional neural networks transmit continuous values, resulting in computational complexity and high power consumption. This makes them unsuitable for deployment on constrained hardware and for processing complex data-intensive tasks.
[0007] On this basis, SNN with biological interpretability, transmission of discrete pulse values and lower energy consumption can be used for motor thermal imaging diagnosis.
[0008] Existing SNN encoding methods primarily utilize rate coding, delay coding, and real-time coding. Rate coding converts input intensity into a firing rate or spike count; delay coding converts input intensity into spike duration; and real-time coding directly encodes input into spikes through the mapping of spiking neurons. In recent years, most SNNs have adopted real-time coding. This approach requires no complex preprocessing and can directly convert external stimuli or input into spikes, making it both simple to use and more biologically compatible.
[0009] However, due to the sparse nature of discrete pulses themselves, when processing motor thermal imaging images, the real-time encoding of SNN will lose a lot of information in the process of encoding the input into pulses. This information may be extremely important for diagnostic decisions, thus reducing the accuracy of motor thermal imaging diagnosis. Summary of the Invention
[0010] To this end, the technical problem to be solved by the present invention is to overcome the existing technology of using SNN for motor thermal imaging diagnosis, in which the real-time encoding method adopted will lead to information loss and reduce the accuracy of the diagnosis results.
[0011] To solve the above technical problems, the present invention provides a motor thermal imaging diagnostic method based on a pulse neural network, comprising:
[0012] Construct a motor thermal imaging diagnostic model, including an adaptive residual coding module and a pulse residual network; the adaptive residual coding module includes a main circuit and a branch circuit;
[0013] The motor thermal imaging image is input into the adaptive residual coding module, and the main path coding feature and branch coding feature are obtained through the main path and branch paths respectively. The information entropy of each channel in the main path coding feature is calculated and normalized. The normalized information entropy is subtracted from 1 to obtain the corresponding weight coefficient of each channel of the branch coding feature. The weight coefficient of each channel of the branch coding feature is used to perform a weighted summation of each channel of the branch coding feature and the main path coding feature to obtain the target pulse coding feature.
[0014] The target pulse code features are input into the pulse residual network to obtain the motor fault category prediction label.
[0015] Preferably, the main path and the branch path of the adaptive residual coding module both include a convolutional layer, a batch normalization layer and a pulse neuron connected in sequence, and the branch path is connected before the convolutional layer and after the pulse neuron of the main path.
[0016] Preferably, in the adaptive residual coding module, the spiking neurons of the main path and the spiking neurons of the branch path are spiking neurons with different membrane potential thresholds, and the membrane potential threshold of the spiking neurons of the branch path is smaller than the membrane potential threshold of the spiking neurons of the main path.
[0017] Preferably, the weight coefficients of each channel of the branch coding feature are used to perform weighted summation on each channel of the branch coding feature and the main channel coding feature to obtain the target pulse coding feature, and the formula is:
[0018] O c,h,w =M c,h,w +θ c ×B c,h,w
[0019] Among them, O c,h,w represents the value of the target pulse coding feature at channel c and position (h, w), M c,h,w represents the value of the main path encoding feature at channel c and position (h, w), B c,h,w represents the value of the branch encoding feature at channel c and position (h, w), where h and w represent the height index and width index respectively, θ cRepresents the weight coefficient of channel c.
[0020] Preferably, Spiking-ResNet, SEW-ResNet or MS-ResNet is used as the pulse residual network.
[0021] Preferably, the pulse neurons in the adaptive residual coding module and the pulse residual network adopt any one of the integral firing neurons, leaky integral firing neurons and their variant models or parallel pulse neurons.
[0022] Preferably, the motor thermal imaging diagnosis model is trained using a mean square error loss function between the motor fault category prediction label and the true label.
[0023] Preferably, the optimization algorithm for training the motor thermal imaging diagnostic model includes a square root transfer algorithm, a stochastic gradient descent method or an adaptive moment estimation algorithm.
[0024] The present invention also provides a motor thermal imaging diagnostic system based on a pulse neural network, comprising:
[0025] A model building module is used to build a motor thermal imaging diagnostic model, including an adaptive residual coding module and a pulse residual network; the adaptive residual coding module includes a main path and a branch path;
[0026] The encoding module is used to input the motor thermal imaging image into the adaptive residual encoding module, and obtain the main path encoding feature and branch encoding feature through the main path and branch paths respectively; calculate the information entropy of each channel in the main path encoding feature and normalize it, and subtract the normalized information entropy from 1 to obtain the corresponding weight coefficient of each channel of the branch encoding feature; use the weight coefficient of each channel of the branch encoding feature to perform weighted summation on each channel of the branch encoding feature and the main path encoding feature to obtain the target pulse encoding feature;
[0027] The prediction module is used to input the target pulse coding features into the pulse residual network to obtain the motor fault category prediction label.
[0028] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned motor thermal imaging diagnostic method based on pulse neural network are implemented.
[0029] The above technical solution of the present invention has the following beneficial effects compared with the prior art:
[0030] The present invention describes a motor thermal imaging diagnostic method based on a pulsed neural network. This method uses a pulsed neural network to build a diagnostic model, improving the model's biological interpretability while reducing computational complexity and energy consumption. Furthermore, to address the information loss caused by real-time pulsed neural network encoding, the present invention constructs an adaptive residual coding module that adaptively adjusts the weight coefficients of each channel of branch coding features based on the information entropy of the main coding path. The branch coding results are used to supplement the main coding information, achieving adaptive information retention. This reduces information loss during the encoding process, enhances the pulse coding results' ability to characterize motor fault states, and improves the accuracy of motor fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:
[0032] Figure 1 is a structural diagram of the adaptive residual coding module of the present invention;
[0033] Figure 2 1 is a schematic diagram of the training of the motor thermal imaging diagnostic model of the present invention;
[0034] Figure 3 Schematic diagram of the test of the motor thermal imaging diagnostic model of the present invention;
[0035] Figure 4 This is a t-SNE dimensionality reduction visualization comparison diagram of the classification results of the test samples by the adaptive residual coding pulse neural network model and the real-time coding pulse neural network model of the present invention, where Figure 4 (a) is a t-SNE dimension reduction visualization diagram of the classification results of the test sample by the adaptive residual coding pulse neural network model of the present invention. Figure 4 (b) is the t-SNE dimension reduction visualization diagram of the classification results of the test sample by the real-time encoding spike neural network model;
[0036] Figure 5 This is a comparison chart of the confusion matrices of the prediction results of the test samples by the adaptive residual coding pulse neural network model and the real-time coding pulse neural network model of the present invention, where Figure 5 (a) is the confusion matrix of the prediction results of the adaptive residual coding pulse neural network model for the test sample of the present invention, Figure 5 (b) Confusion matrix of the prediction results of the real-time coding spike neural network model for the test sample;
[0037] Figure 6 is a visualization diagram of the encoding results of adaptive residual coding and real-time coding in this embodiment, where Figure 6 (a) is the input motor thermal imaging image, Figure 6 (b) in the figure is the coding result obtained by adaptive residual coding. Figure 6 (c) in the figure is the encoding result obtained by real-time encoding. DETAILED DESCRIPTION
[0038] The present invention will be further described below with reference to the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it. However, the embodiments are not intended to limit the present invention.
[0039] A first embodiment of the present invention provides a motor thermal imaging diagnostic method based on a pulse neural network, comprising:
[0040] Construct a motor thermal imaging diagnostic model, including an adaptive residual coding module and a pulse residual network; the adaptive residual coding module includes a main circuit and a branch circuit;
[0041] The motor thermal imaging image is input into the adaptive residual coding module, and the main path coding feature and branch coding feature are obtained through the main path and branch paths respectively. The information entropy of each channel in the main path coding feature is calculated and normalized. The normalized information entropy is subtracted from 1 to obtain the corresponding weight coefficient of each channel of the branch coding feature. The weight coefficient of each channel of the branch coding feature is used to perform a weighted summation of each channel of the branch coding feature and the main path coding feature to obtain the target pulse coding feature.
[0042] The target pulse code features are input into the pulse residual network to obtain the motor fault category prediction label.
[0043] The basic idea of the adaptive residual coding module is to add an adaptive weighted residual coding branch during the pulse coding process to reduce the information loss caused by the coding process. Figure 1 As shown, the main path and branch paths of the adaptive residual coding module both include sequentially connected convolutional layers, batch normalization layers, and pulse neurons, and the branch paths are connected before the convolutional layers and after the pulse neurons of the main path to adaptively reduce information loss.
[0044] Preferably, in the adaptive residual coding module, the spiking neurons of the main path and the spiking neurons of the branch path are spiking neurons with different membrane potential thresholds, and the membrane potential threshold of the spiking neurons of the branch path is smaller than the membrane potential threshold of the spiking neurons of the main path, so as to retain information that may be lost in the main path.
[0045] Specifically, the weight coefficients of each channel of the branch coding feature are used to perform weighted summation on each channel of the branch coding feature and the main channel coding feature to obtain the target pulse coding feature. The formula is:
[0046] O c,h,w =M c,h,w +θ c ×Bc,h,w
[0047] Among them, O c,h,w represents the value of the target pulse coding feature at channel c and position (h, w), M c,h,w represents the value of the main path encoding feature at channel c and position (h, w), B c,h,w represents the value of the branch encoding feature at channel c and position (h, w), where h and w represent the height index and width index respectively, θ c Represents the weight coefficient of channel c.
[0048] The adaptive residual coding module is used to encode the input motor thermal imaging image and adaptively calculate the weight coefficient of the branch coding feature based on the information entropy of the main path coding, so that the target pulse coding feature can combine the coding results under the two membrane potential thresholds of the main path and the branch path, thereby reducing information loss. It is beneficial to further extract the feature of the coding content after inputting the target pulse coding feature into the pulse residual network, thereby improving the accuracy of motor thermal imaging diagnosis.
[0049] Preferably, the pulse residual network can adopt a pulse version of the residual network such as Spiking-ResNet, SEW-ResNet or MS-ResNet.
[0050] Preferably, the pulse neurons used in the adaptive residual coding module and the pulse residual network are any one of the integral firing (IF) neurons, leaky integral firing (LIF) neurons and their variant models (such as PLIF, KLIF, GLIF, etc.) or parallel pulse neurons.
[0051] Reference Figure 2 As shown, the process of training the motor thermal imaging diagnosis model includes:
[0052] S1: Data preprocessing.
[0053] The collected motor thermal imaging images are subjected to data enhancement. After data enhancement, the number of images of each fault category in the dataset is unified and divided into training dataset and test dataset.
[0054] The main methods of data enhancement include rotation, mirroring, adding Gaussian noise, adjusting brightness, and Gaussian blurring. The training dataset is used to train the fault diagnosis model, while the test dataset is not used in model training and is only used to test the accuracy of the model results.
[0055] The dataset used in this example contains eleven fault types: healthy state (H), cooling fan fault (F), rotor stuck fault (R), and stator winding short-circuit faults. A short-circuit rate of 10% is divided into single-phase short circuit (A10), two-phase short circuit (AB10), and three-phase short circuit (ABC10); a short-circuit rate of 30% is divided into single-phase short circuit (A30), two-phase short circuit (AB30), and three-phase short circuit (ABC30); and a short-circuit rate of 50% is divided into single-phase short circuit (A50) and two-phase short circuit (AB50). Fault category labels are represented by 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, and 10, respectively. It should be noted that the number of images in each category in the original dataset is small and unbalanced. Therefore, image data augmentation is performed on the original dataset, and the number of image samples in each category is unified to 450. The description of the thermal imaging image dataset after data augmentation is shown in Table 1. In this embodiment, the training data set and the test data set are divided into a ratio of 4:1, that is, each category has 360 image samples for training and 90 image samples for testing.
[0056] Table 1. Motor thermal imaging dataset after data enhancement
[0057]
[0058]
[0059] S2: Construct a motor thermal imaging diagnostic model.
[0060] The pulse neurons used in this embodiment are all leaky integrating firing neurons.
[0061] In the adaptive residual coding module, the pulse neurons of the main path are set as leaky integration firing neurons, and their threshold is the default 1.0; the pulse neurons of the branch path are set as leaky integration firing neurons, and their threshold is 0.5.
[0062] The pulse residual network of this embodiment uses Spiking-ResNet18 as the basic network structure, and the time step is set to 4.
[0063] S3: Input the training dataset into the constructed motor thermal imaging diagnostic model and perform model training according to the given loss function and optimization algorithm.
[0064] In this embodiment, the batch size of the training data set is set to 16.
[0065] Preferably, for the input training data set, the model is trained using a mean square error loss function between the motor fault category prediction label and the true label. The purpose of training is to enable the model to classify the input samples into the correct fault category label.
[0066] Preferably, the optimization algorithm for training the motor thermal imaging diagnostic model includes but is not limited to one of the root mean square propagation algorithm (RMSprop), the stochastic gradient descent method (SGD), and the adaptive moment estimation algorithm (Adam).
[0067] This embodiment uses an adaptive moment estimation algorithm with a learning rate of 0.001. After 64 iterations, the loss function tends to be balanced, completing the model training.
[0068] Reference Figure 3 As shown in Figure 1, the test dataset is fed into the trained motor thermal imaging diagnostic model to perform motor fault diagnosis. Specifically, the test accuracy result is calculated based on the class prediction labels and actual labels obtained by inputting the test samples into the model.
[0069] In this embodiment, the batch size of the test data set is set to 16.
[0070] In order to further illustrate the advantages of the proposed encoding method, this embodiment is also compared with a real-time encoding model with the same structure.
[0071] Figure 4 This is a t-SNE dimensionality reduction visualization comparison diagram of the classification results of the test samples by the adaptive residual coding pulse neural network model and the real-time coding pulse neural network model of the present invention, where Figure 4 (a) is a t-SNE dimension reduction visualization diagram of the classification results of the test sample by the adaptive residual coding pulse neural network model of the present invention. Figure 4 (b) is a t-SNE dimension reduction visualization of the classification results of the test samples by the real-time coding pulse neural network model. The network structures of the adaptive residual coding pulse neural network model and the real-time coding pulse neural network model are the same except for the encoding method. It can be seen that the method of the present invention can effectively cluster samples of the same category and create a clearer boundary between the features of samples of different categories. Compared with the method of the present invention, the real-time coding method has obvious classification errors, which proves that the method of the present invention can retain more effective information important for diagnostic decision-making and improve the intra-class clustering and inter-class separability of fault categories.
[0072] The confusion matrix of the prediction results of the test samples by the adaptive residual coding pulse neural network model and the real-time coding pulse neural network model of the present invention is as follows: Figure 5 As shown, Figure 5 (a) is the confusion matrix of the prediction results of the adaptive residual coding pulse neural network model for the test sample of the present invention, Figure 5(b) Confusion matrix of the prediction results of the real-time coding spiking neural network model for the test samples. As can be seen, the proposed method has a high diagnostic accuracy of 99.49%, with only five misclassifications among 990 test samples of eleven fault types. In comparison, the real-time coding method has a diagnostic accuracy of only 97.47%, with 25 misclassifications. This demonstrates the superior diagnostic performance and generalization ability of the proposed method for motor thermal imaging.
[0073] In addition, this embodiment also visualizes the output of the two encoding methods, and the results are as follows: Figure 6 As shown, Figure 6 (a) is the input motor thermal imaging image, Figure 6 (b) in the figure is the coding result obtained by adaptive residual coding. Figure 6 Figure (c) shows the encoding result obtained using real-time encoding. As can be seen, real-time encoding loses a significant amount of information during the encoding process, resulting in large expanses of black or white. However, the adaptive residual coding proposed in this invention retains more effective information, which significantly improves the diagnostic model's final performance.
[0074] In summary, the motor thermal imaging diagnostic method based on a pulse neural network described in the present invention uses a pulse neural network to build a diagnostic model, which improves the biological interpretability of the model while reducing computational complexity and energy consumption. Furthermore, to address the problem of information loss caused by the real-time encoding method of the pulse neural network, the present invention constructs an adaptive residual coding module that adaptively adjusts the weight coefficients of each channel of the branch coding feature based on the information entropy of the main coding path, and uses the branch coding results to supplement the main coding information, thereby achieving adaptive information retention. This can reduce information loss caused by the encoding process, enhance the pulse coding results' ability to characterize the motor fault state, and improve the accuracy of motor fault diagnosis.
[0075] Based on the above-mentioned motor thermal imaging diagnosis method based on a pulse neural network, this embodiment further provides a motor thermal imaging diagnosis system based on a pulse neural network, including:
[0076] A model building module is used to build a motor thermal imaging diagnostic model, including an adaptive residual coding module and a pulse residual network; the adaptive residual coding module includes a main path and a branch path;
[0077] The encoding module is used to input the motor thermal imaging image into the adaptive residual encoding module, and obtain the main path encoding feature and branch encoding feature through the main path and branch paths respectively; calculate the information entropy of each channel in the main path encoding feature and normalize it, and subtract the normalized information entropy from 1 to obtain the corresponding weight coefficient of each channel of the branch encoding feature; use the weight coefficient of each channel of the branch encoding feature to perform weighted summation on each channel of the branch encoding feature and the main path encoding feature to obtain the target pulse encoding feature;
[0078] The prediction module is used to input the target pulse coding features into the pulse residual network to obtain the motor fault category prediction label.
[0079] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned motor thermal imaging diagnostic method based on pulse neural network are implemented.
[0080] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.
[0081] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0082] These computer program instructions may 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 produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.
[0083] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0084] Obviously, the above embodiments are merely examples for clarity of explanation and are not intended to limit the implementation methods. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all implementation methods here. Obvious variations or modifications arising therefrom remain within the scope of protection of the present invention.
Claims
1. A motor thermal imaging diagnostic method based on pulse neural network, characterized in that: include: Construct a motor thermal imaging diagnostic model, including an adaptive residual coding module and a pulse residual network; The adaptive residual coding module includes a main path and a branch path; The motor thermal imaging image is input into the adaptive residual coding module, and the main path coding features and branch coding features are obtained respectively through the main path and branch path; Calculate the information entropy of each channel in the main channel coding feature and normalize it. Subtract the normalized information entropy from 1 to obtain the corresponding weight coefficient of each channel of the branch coding feature. Use the weight coefficient of each channel of the branch coding feature to perform weighted summation of each channel of the branch coding feature and the main channel coding feature to obtain the target pulse coding feature. The target pulse code features are input into the pulse residual network to obtain the motor fault category prediction label.
2. The motor thermal imaging diagnostic method based on pulse neural network according to claim 1 is characterized in that: Both the main path and the branch path of the adaptive residual coding module include a convolutional layer, a batch normalization layer, and a pulse neuron connected in sequence, and the branch path is connected before the convolutional layer and after the pulse neuron of the main path.
3. The motor thermal imaging diagnostic method based on pulse neural network according to claim 2, characterized in that: In the adaptive residual coding module, the spiking neurons of the main path and the spiking neurons of the branch path are spiking neurons with different membrane potential thresholds, and the membrane potential threshold of the spiking neurons of the branch path is smaller than the membrane potential threshold of the spiking neurons of the main path.
4. The motor thermal imaging diagnostic method based on pulse neural network according to claim 1, characterized in that: The weight coefficients of each channel of the branch coding feature are used to perform weighted summation on each channel of the branch coding feature and the main channel coding feature to obtain the target pulse coding feature. The formula is: The c,h,w =M c,h,w +θ c ×B c,h,w Among them, O c,h,w represents the value of the target pulse coding feature at channel c and position (h, w), M c,h,w represents the value of the main path encoding feature at channel c and position (h, w), B c,h,w represents the value of the branch encoding feature at channel c and position (h, w), where h and w represent the height index and width index respectively, θ c Represents the weight coefficient of channel c.
5. The motor thermal imaging diagnostic method based on pulse neural network according to claim 1, characterized in that: Spiking-ResNet, SEW-ResNet or MS-ResNet is used as the pulse residual network.
6. The motor thermal imaging diagnostic method based on pulse neural network according to claim 1, characterized in that: The pulse neurons in the adaptive residual coding module and the pulse residual network adopt any one of the integral firing neurons, the leaky integral firing neurons and their variant models or the parallel pulse neurons.
7. The motor thermal imaging diagnostic method based on pulse neural network according to claim 1, characterized in that: The motor thermal imaging diagnosis model is trained using a mean square error loss function between the motor fault category prediction label and the true label.
8. The motor thermal imaging diagnostic method based on pulse neural network according to claim 1 is characterized in that: The optimization algorithm for training the motor thermal imaging diagnostic model includes a square root transfer algorithm, a stochastic gradient descent method or an adaptive moment estimation algorithm.
9. A motor thermal imaging diagnostic system based on pulse neural network, characterized in that: include: A model building module, used to build a motor thermal imaging diagnostic model, including an adaptive residual coding module and a pulse residual network; The adaptive residual coding module includes a main path and a branch path; The encoding module is used to input the motor thermal imaging image into the adaptive residual encoding module, and obtain the main path encoding feature and the branch encoding feature through the main path and the branch path respectively; Calculate the information entropy of each channel in the main channel coding feature and normalize it. Subtract the normalized information entropy from 1 to obtain the corresponding weight coefficient of each channel of the branch coding feature. Use the weight coefficient of each channel of the branch coding feature to perform weighted summation of each channel of the branch coding feature and the main channel coding feature to obtain the target pulse coding feature. The prediction module is used to input the target pulse coding features into the pulse residual network to obtain the motor fault category prediction label.
10. 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 a motor thermal imaging diagnostic method based on a pulse neural network as claimed in any one of claims 1 to 8.