Electrical control circuit fault adaptive diagnosis system and method based on deep learning

Abstract

The invention discloses an electrical control circuit fault adaptive diagnosis system and method based on deep learning. The system comprises a data acquisition module, a preprocessing module, a feature extraction module, a diagnosis model module, an adaptive optimization module and a visual interaction module. Based on an improved Transform architecture, a time convolution operator is embedded to enhance the local feature perception capability of a self-attention mechanism, high-dimensional time sequence features are extracted, a dual-channel deep residual network is constructed, electrical features and thermal-mechanical features are analyzed respectively, a fault classification result is output after fusion, and model parameters are updated through online incremental learning. And on the basis of a transfer learning controller, new fault sample distribution characteristic transfer pre-training parameters are adapted, and a genetic algorithm is utilized to dynamically optimize the network depth and the convolution kernel size so as to improve the generalization ability. The method has the advantages that high-precision and multi-dimensional information fusion is realized through an improved Transform architecture and a self-adaptive optimization mechanism in combination with time convolution and transfer learning, and the method can quickly adapt to new fault types.

Description

Technical Field

[0001] The present invention relates to deep learning, and particularly to an adaptive fault diagnosis system and method for electrical control circuits based on deep learning. Background Art

[0002] Electrical control circuits are an indispensable part of modern industrial automation systems and are widely used in various fields such as mechanical equipment, production lines, and transportation vehicles. With the development of electrical control technology, the complexity of circuits has gradually increased, leading to an increase in the likelihood of faults. Faults in electrical control circuits usually result in equipment downtime or performance degradation, posing risks to production efficiency and safety. Therefore, it is particularly important to diagnose circuit faults promptly and accurately and take corresponding repair measures.

[0003] Currently, most of the adaptive fault diagnosis systems for electrical control circuits on the market are based on rule engines or shallow machine learning, relying on manual feature extraction and threshold setting. Although such methods can identify conventional faults, they have obvious shortcomings: firstly, they have insufficient ability to analyze complex non-linear signals, resulting in a relatively high false detection rate; secondly, the model has poor generalization ability, and when the circuit structure or load changes, the rule base needs to be redesigned, with high maintenance costs; thirdly, traditional algorithms have insufficient global correlation modeling of time series features and are difficult to capture the dynamic coupling relationship of multi-sensor data, easily missing early hidden faults. Summary of the Invention

[0004] In order to improve the existing adaptive fault diagnosis system and method for electrical control circuits, an adaptive fault diagnosis system and method for electrical control circuits based on deep learning are provided, which integrates an improved version of the Transformer architecture and a temporal convolutional network, captures long-range temporal dependencies through the multi-head self-attention mechanism, greatly reduces the dependence on manual experience and full-scale data, and provides a more cost-effective solution for complex industrial scenarios.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] An adaptive fault diagnosis system for electrical control circuits based on deep learning, comprising:

[0007] A data acquisition module, including a high-precision current / voltage sensor array, an infrared thermal imaging unit, and a vibration sensor group, for real-time acquisition of the operating time series signals, temperature field distribution data, and mechanical vibration spectra of the electrical control circuit;

[0008] A preprocessing module, including a signal denoising unit and a data alignment unit, where the signal denoising unit uses a wavelet threshold denoising algorithm to perform non-linear filtering on the time series signals and applies an adaptive contrast enhancement algorithm to process infrared images;

[0009] The feature extraction module includes an improved Transformer architecture, which embeds a temporal convolution operator in the self-attention mechanism to form a temporal encoder with local feature perception ability;

[0010] The diagnostic model module includes a dual-channel deep residual network and an online incremental learning unit. The dual-channel network processes electrical features and thermal-mechanical features respectively to obtain the fault diagnosis result of the electrical control circuit;

[0011] The adaptive optimization module includes a transfer learning controller and a model structure dynamic adjustment unit. The transfer learning controller selects pre-trained model parameters according to the distribution characteristics of new fault samples, and the dynamic adjustment unit optimizes the network depth and convolution kernel size through a genetic algorithm;

[0012] The visualization and interaction module includes a fault heat map generator and a diagnostic decision tree interpreter, which can display the fault location result in real time and provide an interpretable analysis of the diagnostic basis.

[0013] Preferably, the data acquisition module specifically includes:

[0014] A high-precision current / voltage sensor array, which uses multi-channel synchronous sampling technology to capture the dynamic time-series signals of the electrical control circuit in real time, completes signal filtering, analog-to-digital conversion and data calibration, and synchronously encapsulates the raw signals with timestamps and operating condition parameters through a high-speed communication interface, and transmits them to the preprocessing module for further processing;

[0015] An infrared thermal imaging unit, which synchronously captures the temperature field distribution data by scanning the surface of the electrical equipment in real time, associates and integrates the temporal temperature matrix with the equipment operating condition parameters, and constructs a dynamic thermal feature data set for streaming transmission to the preprocessing module;

[0016] A vibration sensor group, which collects vibration analog signals in real time and converts them into digital signals, and generates time-domain waveform data after filtering, amplification and analog-to-digital conversion and transmits it to the preprocessing module.

[0017] Preferably, the preprocessing module specifically includes:

[0018] A signal denoising unit, which uses the wavelet threshold denoising algorithm to perform non-linear filtering on the time-series signal, specifically including:

[0019] The original signal based on the data acquisition module is decomposed into sub-bands of different frequencies through multi-scale wavelet transform, and the approximate coefficients and detail coefficients are extracted for wavelet decomposition. The formula is:

[0020]

[0021] Among them, x(n) is the original signal, φ(t) is the wavelet basis function, is the scaling function, cA j (k), cD j (k) are the approximation coefficient and the detail coefficient of the j-th layer respectively, and k is the position index;

[0022] Based on the extracted approximation coefficients and detail coefficients, distinguish the signal from the noise, and perform non-linear filtering to suppress the noise;

[0023] Reconstruct the noise-reduced signal based on the processed wavelet coefficients;

[0024] The data alignment unit, the specific process of processing the infrared image by the adaptive contrast enhancement algorithm includes:

[0025] Calculate the local mean and standard deviation of the infrared image by block;

[0026] Dynamically design the gain factor based on the standard deviation, perform non-linear contrast stretching on each sub-block, and fuse the overlapping regions to avoid block effects;

[0027] Obtain the standard dynamic range based on global normalization, suppress noise and over-enhancement while enhancing the characteristics of the fault area.

[0028] Preferably, the feature extraction module specifically includes:

[0029] The improved Transformer architecture specifically includes:

[0030] Based on the data processed by the preprocessing module, perform one-dimensional causal convolution, and the number of convolution output channels is d to obtain local features;

[0031] Add the local features output by the temporal convolution to the original data embedding element-wise to form a fused feature;

[0032] Generate query (Q), key (K), and value (V) matrices from the fused data through a linear transformation, and divide the query (Q), key (K), and value (V) into h heads, and the dimension of each head is

[0033] Concatenate the attention outputs of each head and project them back to the original input dimension through a linear layer;

[0034] Add the output of the multi-head attention to the original input to retain the initial information, alleviate the vanishing gradient, and stabilize the training process, adjust the output distribution, and accelerate convergence;

[0035] Introduce non-linear transformation through a feed-forward network to refine features position by position;

[0036] The output result of the feedforward network is added to the normalized attention result, and the output distribution is stabilized through layer normalization, serving as the final output of the encoder layer;

[0037] Multiple encoder layers are repeatedly stacked to extract high-order temporal features layer by layer. The parameters of each layer are independent, and local features are gradually fused.

[0038] Preferably, the diagnostic model module specifically includes:

[0039] A dual-channel deep residual network model. After obtaining electrical feature data and thermal-mechanical feature data, the dual-channel deep residual network model avoids gradient disappearance in the deep network based on residual connections. For each channel, the following residual block structure can be used:

[0040] y = F(x,{W i}) + x

[0041] where F(x,{W i}) is a convolution operation, and W i is the convolution kernel;

[0042] After processing the electrical features and thermal-mechanical features separately in the two channels, the outputs of the two channels are fused, and the initial fault diagnosis result is obtained through a fully connected layer;

[0043] An online incremental learning unit. The online incremental learning unit uses methods such as online gradient descent to gradually update the model. For each new data batch, the loss function is updated based on the cross-entropy loss, in the form of:

[0044]

[0045] where y i is the true label, is the predicted output of the model, and N is the total number of data;

[0046] Based on the output of the dual-channel network and the adaptability of the incremental learning unit, the diagnostic result is classified through the probability of the model output to obtain the final electrical control circuit fault diagnosis result and identify the fault type.

[0047] Preferably, the adaptive optimization module specifically includes:

[0048] A transfer learning controller. The transfer learning controller calculates the distribution difference between the new sample and the source domain sample by extracting the feature vector of the new fault sample and selects a parameter transfer strategy;

[0049] A model structure dynamic adjustment unit. The model structure dynamic adjustment unit optimizes the structure of the neural network through a genetic algorithm, including the network depth and the size of the convolution kernel;

[0050] Transfer learning selects a suitable model and fine-tunes it, dynamically adjusts the units to optimize the network structure, and further fine-tunes the model according to the new samples and structure until the diagnostic effect reaches the expectation.

[0051] Preferably, the transfer learning controller specifically includes:

[0052] Construct a pre-trained model library containing multiple device models, and each model establishes a hypersphere decision boundary in the parameter space;

[0053] Calculate the Wasserstein distance between the distribution of new fault data and each pre-trained model;

[0054] Select the three models with the closest distances for model parameter interpolation, and the interpolation weights are determined according to the reciprocals of the relative distances;

[0055] During the transfer process, freeze the underlying feature extraction layer and only fine-tune the upper classifier.

[0056] Preferably, the model structure dynamic adjustment unit specifically includes:

[0057] Automatically increase or decrease the number of residual blocks based on the change rate of the validation set accuracy for network depth evolution, and the increase or decrease decision formula is:

[0058]

[0059] where ΔL is the change in the number of residual blocks, η is the adjustment coefficient, A val(t) -A val(t-1) is the difference in the validation set accuracy between the current stage and the previous stage, reflecting the change in model performance, and σ A is the standard deviation of the validation set accuracy, used to normalize the adjustment amount;

[0060] Optimize the convolutional kernel size through genetic algorithm evolution selection in the candidate set;

[0061] Dynamically adjust the number of channels in each layer based on the information entropy value of the feature map.

[0062] Preferably, the visualization interaction module specifically includes:

[0063] The heat map generator uses gradient class activation mapping technology to superimpose the fault feature areas concerned by the neural network on the circuit topology map in the form of color mapping, and real-time annotates the abnormal current / voltage signal distribution to accurately locate the fault components;

[0064] The diagnostic decision tree interpreter parses the fault classification probability output by the model into an understandable logical reasoning chain through rule extraction and path backtracking, dynamically displays the decision-making process of "signal distortion → feature extraction → hierarchical determination", and associates with the historical fault case library.

[0065] Furthermore, the electrical control circuit fault adaptive diagnosis method based on deep learning includes:

[0066] Simultaneously collect the timing signals, temperature field distribution, and mechanical vibration spectrum data of the circuit through a high-precision sensor array, infrared thermal imaging, and vibration sensor group;

[0067] Use the wavelet threshold denoising algorithm to denoise the timing signals, eliminate the multi-sensor timing deviation through data alignment, and optimize the infrared image quality using adaptive contrast enhancement;

[0068] Based on the improved Transformer architecture, embed the temporal convolutional operator to enhance the local feature perception ability of the self-attention mechanism and extract high-dimensional temporal features;

[0069] Construct a dual-channel deep residual network to separately analyze electrical features and thermal-mechanical features, fuse them to output the fault classification result, and update the model parameters through online incremental learning;

[0070] Based on the transfer learning controller, adapt the pre-trained parameters to the distribution characteristics of new fault samples, and use the genetic algorithm to dynamically optimize the network depth and convolution kernel size to improve the generalization ability;

[0071] Generate a fault thermal map to locate the abnormal area, and combine with the decision tree interpreter to provide a visual analysis and interpretable report of the diagnostic basis.

[0072] Compared with the prior art, the advantages of the present invention are as follows:

[0073] The improved Transformer architecture can effectively perceive the local features in the timing signals by embedding the temporal convolutional operator in the self-attention mechanism, improving the accuracy and robustness of fault diagnosis. Especially for complex electrical control circuits, it can accurately capture the timing patterns and evolution processes of faults. The model dynamic adjustment mechanism combines the transfer learning controller and the genetic algorithm, which can automatically optimize the network structure according to different fault samples and changing environmental conditions, such as adjusting the network depth and convolution kernel size, so as to achieve rapid adaptation to new fault types. This adaptive ability enables the system to continuously learn and improve the accuracy of fault diagnosis. Especially in the face of unknown or complex faults, the system can effectively borrow existing knowledge through transfer learning, avoiding the inefficiency and misdiagnosis risks of traditional methods. The system separates electrical features and thermal-mechanical features through a dual-channel deep residual network, further improving the accuracy of multi-dimensional information fusion and providing a more comprehensive and accurate analysis for fault location. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 It is a schematic diagram of the system proposed by the present invention;

[0075] Figure 2 Schematic diagram of the method proposed by the present invention;

[0076] Figure 3 Schematic diagram of the Transformer architecture proposed by the present invention;

[0077] Figure 4 Schematic diagram of the transfer learning controller proposed by the present invention;

[0078] Figure 5 Schematic diagram of the model structure dynamic adjustment unit proposed by the present invention;

[0079] Figure 6 Architecture diagram of the electronic device in this solution;

[0080] Figure 7 Schematic diagram of the structure of the computer-readable storage medium in this solution. Detailed implementation manners

[0081] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments in the following description are only examples, and those skilled in the art can think of other obvious variations.

[0082] Refer to Figure 1 As shown, the deep learning-based electrical control circuit fault adaptive diagnosis system includes:

[0083] Data acquisition module, including a high-precision current / voltage sensor array, an infrared thermal imaging unit, and a vibration sensor group, for real-time acquisition of the operation timing signals, temperature field distribution data, and mechanical vibration spectra of the electrical control circuit;

[0084] Preprocessing module, including a signal denoising unit and a data alignment unit. The signal denoising unit uses the wavelet threshold denoising algorithm to perform non-linear filtering on the timing signals and applies the adaptive contrast enhancement algorithm to process the infrared images;

[0085] Feature extraction module, including an improved Transformer architecture, embedding a temporal convolutional operator in the self-attention mechanism to form a timing encoder with local feature perception ability;

[0086] Diagnostic model module, including a dual-channel deep residual network and an online incremental learning unit. The dual-channel network processes electrical features and thermal-mechanical features respectively to obtain the fault diagnosis results of the electrical control circuit;

[0087] Adaptive optimization module, including a transfer learning controller and a model structure dynamic adjustment unit. The transfer learning controller selects pre-trained model parameters according to the distribution characteristics of new fault samples, and the dynamic adjustment unit optimizes the network depth and convolutional kernel size through a genetic algorithm;

[0088] The visual interaction module, including a fault heat map generator and a diagnostic decision tree interpreter, can display the fault location results in real time and provide an interpretable analysis of the diagnostic basis.

[0089] See Figure 1 As shown, the data acquisition module specifically includes:

[0090] A high-precision current / voltage sensor array that captures the dynamic timing signals of the electrical control circuit in real time through multi-channel synchronous sampling technology, completes signal filtering, analog-to-digital conversion, and data calibration, and synchronously packages the original signals with time stamps and operating condition parameters through a high-speed communication interface and transmits them to the preprocessing module for further processing;

[0091] An infrared thermal imaging unit that synchronously captures the temperature field distribution data by scanning the surface of the electrical equipment in real time, correlates and integrates the timed temperature matrix with the equipment operating condition parameters, and constructs a dynamic thermal feature data set for streaming transmission to the preprocessing module;

[0092] A vibration sensor group that collects vibration analog signals in real time and converts them into digital signals, generates time-domain waveform data after filtering, amplification, and analog-to-digital conversion, and transmits it to the preprocessing module.

[0093] It can be understood that sensor data is usually high-dimensional and needs to be correlated and integrated with the operating condition data of electrical equipment. The large amount of data and complex correlations may lead to delays in data processing. Data dimensionality reduction and feature extraction technologies are adopted to optimize the data processing algorithm. In addition, a real-time data stream processing framework, such as Apache Kafka, can be introduced to ensure the real-time nature and processing efficiency of the data.

[0094] See Figure 1 As shown, the preprocessing module specifically includes:

[0095] A signal denoising unit that uses the wavelet threshold denoising algorithm to perform nonlinear filtering on the timing signal, specifically including:

[0096] The original signal based on the data acquisition module is decomposed into sub-bands of different frequencies through multi-scale wavelet transform, and the approximate coefficients and detail coefficients are extracted for wavelet decomposition. The formula is:

[0097]

[0098] Among them, x(n) is the original signal, φ(t) is the wavelet basis function, is the scaling function, cA j (k), cD j(k) are the approximation coefficients and detail coefficients of the j-th layer respectively, and k is the position index;

[0099] Distinguish signals from noise based on the extracted approximation coefficients and detail coefficients, and perform non-linear filtering to suppress noise;

[0100] Reconstruct the denoised signal based on the processed wavelet coefficients;

[0101] The data alignment unit, the specific process of processing the infrared image through the adaptive contrast enhancement algorithm includes:

[0102] Calculate the local mean and standard deviation of the infrared image by block;

[0103] Dynamically design the gain factor based on the standard deviation, perform non-linear contrast stretching on each sub-block, and fuse the overlapping regions to avoid block effects;

[0104] Obtain the standard dynamic range based on global normalization, while enhancing the characteristics of the fault area, suppressing noise and over-enhancement.

[0105] Specifically, in practical applications, the wavelet coefficients are usually processed by soft threshold or hard threshold methods. Soft threshold can provide smoother signal reconstruction, while hard threshold can better retain important detail information. Therefore, choosing the appropriate threshold method and strategy is crucial for improving the denoising effect and retaining signal characteristics. In addition, multiple decomposition levels can be combined to process noises in different frequency bands, improving the overall performance of denoising, especially more effective in the processing of complex signals.

[0106] Refer to Figure 3 As shown, the feature extraction module specifically includes:

[0107] The improved Transformer architecture specifically includes:

[0108] Based on the data processed by the preprocessing module, perform one-dimensional causal convolution, the number of convolution output channels is d, and obtain local features;

[0109] Add the local features output by the temporal convolution to the original data embedding element-wise to form fused features;

[0110] Generate query (Q), key (K), and value (V) matrices from the fused data through linear transformation, and divide the query (Q), key (K), and value (V) into h heads, and the dimension of each head is

[0111] Concatenate the attention outputs of each head and project them back to the original input dimension through a linear layer;

[0112] Add the output of the multi-head attention to the original input to retain the initial information, alleviate the vanishing gradient, and stabilize the training process, adjust the output distribution, and accelerate convergence;

[0113] Introduce non-linear transformation through a feed-forward network to refine features position by position;

[0114] Based on the output result of the feed-forward network and the normalized attention result, add them together, and stabilize the output distribution through layer normalization as the final output of the encoder layer;

[0115] Repeat and stack multiple encoder layers to extract high-order temporal features layer by layer. The parameters of each layer are independent, and local features are gradually fused.

[0116] Specifically, in the multi-head attention mechanism, in addition to splitting the query (Q), key (K), and value (V) into multiple heads and performing parallel calculations, it is worth noting that multi-head attention enables the model to simultaneously focus on different parts of the input data in different subspaces, thereby capturing more abundant temporal features. Each head learns different weights during calculation, so the diversity and complexity of information can be captured within different attention subspaces. By concatenating the outputs of these heads and restoring the dimensions through a linear transformation, the expressive power of the model can be effectively enhanced.

[0117] Refer to Figure 1 As shown, the diagnostic model module specifically includes:

[0118] A dual-channel deep residual network model. After obtaining electrical feature data and thermal-mechanical feature data, the dual-channel deep residual network model avoids the vanishing gradient in the deep network based on residual connections. For each channel, the following residual block structure can be used:

[0119] y = F(x,{W i}) + x

[0120] where F(x,{W i}) is a convolution operation, and W i is a convolution kernel;

[0121] After processing the electrical features and thermal-mechanical features separately in the two channels, fuse the outputs of the two channels and obtain the initial fault diagnosis result through a fully connected layer;

[0122] An online incremental learning unit. The online incremental learning unit uses methods such as online gradient descent to gradually update the model. For each new data batch, the loss function is updated based on the cross-entropy loss, in the form of:

[0123]

[0124] where, yi is the real label, is the predicted output of the model, and N is the total number of data;

[0125] Based on the output of the dual-channel network and the adaptability of the incremental learning unit, the diagnostic results are classified by the probability of the model output to obtain the final fault diagnosis result of the electrical control circuit and identify the fault type.

[0126] It can be understood that in the training of deep networks, especially in deep residual networks, the phenomenon of gradient vanishing or gradient explosion may occur, resulting in the ineffective update of network parameters. Appropriate initialization methods (such as Xavier initialization or He initialization) can be used to ensure the reasonable distribution of weights in the initial stage or when the gradient exceeds the preset threshold, clip the gradient to prevent gradient explosion.

[0127] Refer to Figure 1 shown, the adaptive optimization module specifically includes:

[0128] A transfer learning controller that calculates the distribution difference between the new sample and the source domain sample by extracting the feature vector of the new fault sample and selects a parameter transfer strategy;

[0129] A model structure dynamic adjustment unit that optimizes the structure of the neural network, including the network depth and the size of the convolutional kernel, through a genetic algorithm;

[0130] The transfer learning selects a suitable model and fine-tunes it, the dynamic adjustment unit optimizes the network structure, and further fine-tunes the model according to the new samples and structure until the diagnostic effect reaches the expectation.

[0131] Refer to Figure 4 shown, the transfer learning controller specifically includes:

[0132] Construct a pre-trained model library containing multiple device models, and each model establishes a hypersphere decision boundary in the parameter space;

[0133] Calculate the Wasserstein distance between the new fault data distribution and each pre-trained model;

[0134] Select the three models with the closest distances for model parameter interpolation, and the interpolation weights are determined according to the reciprocal of the relative distances;

[0135] Freeze the underlying feature extraction layer during the transfer process and only fine-tune the upper classifier.

[0136] Refer to Figure 5 shown, the model structure dynamic adjustment unit specifically includes:

[0137] Automatically increase or decrease the number of residual blocks for network depth evolution based on the change rate of the validation set accuracy. The increase or decrease decision formula is:

[0138]

[0139] where ΔL is the change in the number of residual blocks, η is the adjustment coefficient, A val(t) -A val(t-1) is the difference in the validation set accuracy between the current stage and the previous stage, reflecting the change in model performance, and σ A is the standard deviation of the validation set accuracy, used to normalize the adjustment amount;

[0140] Optimize the convolutional kernel size through evolutionary selection in the candidate set using the genetic algorithm;

[0141] Dynamically adjust the number of channels in each layer based on the information entropy value of the feature map.

[0142] Refer to Figure 1 As shown, the visual interaction module specifically includes:

[0143] The heat map generator uses the gradient class activation mapping technology to superimpose the fault feature areas concerned by the neural network on the circuit topology diagram in a color mapping manner, and real-time annotates the abnormal current / voltage signal distribution to accurately locate the fault components;

[0144] The diagnostic decision tree interpreter parses the fault classification probability output by the model into an understandable logical reasoning chain through rule extraction and path backtracking, dynamically displays the decision-making process of "signal distortion → feature extraction → hierarchical determination", and associates with the historical fault case library.

[0145] Refer to Figure 2 As shown, the deep learning-based adaptive fault diagnosis method for electrical control circuits includes:

[0146] Step 1: Synchronously collect the timing signals, temperature field distribution, and mechanical vibration spectrum data of the circuit through a high-precision sensor array, infrared thermal imaging, and vibration sensor group;

[0147] Step 2: Denoise the timing signals using the wavelet threshold denoising algorithm, eliminate the multi-sensor timing deviation through data alignment, and optimize the infrared image quality using adaptive contrast enhancement;

[0148] Step 3: Based on the improved Transformer architecture, embed the time convolution operator to enhance the local feature perception ability of the self-attention mechanism and extract high-dimensional timing features;

[0149] Step 4: Construct a two-channel deep residual network to separately analyze the electrical features and thermal-mechanical features, fuse and output the fault classification results, and update the model parameters through online incremental learning;

[0150] Step Five: Based on the transfer learning controller, adapt the pre-trained parameters to the characteristics of the new fault sample distribution, and use the genetic algorithm to dynamically optimize the network depth and convolution kernel size to improve the generalization ability;

[0151] Step Six: Generate a fault heat map to locate the abnormal area, and combine it with the decision tree interpreter to provide a visualization analysis and an interpretable report of the diagnostic basis.

[0152] Furthermore, the method according to the embodiment of the present application can also be implemented with the aid of Figure 6 the architecture of the electronic device shown. As Figure 6 shown, the electronic device 500 may include a bus 501, one or more CPUs 502, a read-only memory (ROM) 503, a random access memory (RAM) 504, a communication port 505 connected to the network, an input / output component 506, a hard disk 507, etc. The storage device in the electronic device 500, such as the ROM 503 or the hard disk 507, may store the electrical control circuit fault adaptive diagnosis system and method based on deep learning provided by the present application. The electronic device 500 may also include a terminal interface 508. Of course, Figure 6 the architecture shown is only exemplary. When implementing different devices, one or more components shown in the Figure 6 electronic device may be omitted according to actual needs.

[0153] Figure 7 is a schematic diagram of the structure of a computer-readable storage medium provided by an embodiment of the present application. As Figure 7 shown, it is a computer-readable storage medium 600 according to an embodiment of the present application. Computer-readable instructions are stored on the computer-readable storage medium 600. When the computer-readable instructions are run by a processor, the electrical control circuit fault adaptive diagnosis system and method according to the embodiment of the present application described with reference to the above drawings can be executed. The storage medium 600 includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory, etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0154] It should be noted that: the above sequence of the embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. And the above specific embodiments of this specification have been described. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0155] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments.

[0156] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. An adaptive fault diagnosis system for electrical control circuits based on deep learning, characterized in that: include: Data acquisition module, including a high-precision current / voltage sensor array, an infrared thermal imaging unit, and a vibration sensor group, for real-time acquisition of operating timing signals, temperature field distribution data, and mechanical vibration spectra of the electrical control circuit; A preprocessing module, including a signal noise reduction unit and a data alignment unit, wherein the signal noise reduction unit uses a wavelet threshold noise reduction algorithm to perform nonlinear filtering on the time series signal and applies an adaptive contrast enhancement algorithm to process the infrared image; The feature extraction module includes an improved Transformer architecture that embeds a temporal convolution operator in the self-attention mechanism to form a temporal encoder with local feature perception capabilities; A diagnostic model module, including a dual-channel deep residual network and an online incremental learning unit, wherein the dual-channel network processes electrical characteristics and thermal-mechanical characteristics respectively to obtain electrical control circuit fault diagnosis results; An adaptive optimization module, including a transfer learning controller and a model structure dynamic adjustment unit, wherein the transfer learning controller selects pre-trained model parameters according to the distribution characteristics of new fault samples, and the dynamic adjustment unit optimizes the network depth and convolution kernel size through a genetic algorithm; The visualization interaction module, including the fault heat map generator and the diagnosis decision tree interpreter, can display the fault location results in real time and provide interpretable analysis of the diagnosis basis.

2. The deep learning-based electrical control circuit fault adaptive diagnosis system according to claim 1, characterized in that: The data acquisition module specifically includes: A high-precision current / voltage sensor array, which uses multi-channel synchronous sampling technology to capture the dynamic timing signals of the electrical control circuit in real time, completes signal filtering, analog-to-digital conversion and data calibration, and synchronously encapsulates the original signal with time stamp and operating condition parameters through a high-speed communication interface, and transmits them to the preprocessing module for further processing; Infrared thermal imaging unit, which scans the surface of electrical equipment in real time, synchronously captures temperature field distribution data, associates and integrates the time-series temperature matrix with equipment operating parameters, and constructs a dynamic thermal feature data set for streaming to the preprocessing module; The vibration sensor group collects vibration analog signals in real time and converts them into digital signals. After filtering, amplification and analog-to-digital conversion, time domain waveform data is generated and transmitted to the preprocessing module.

3. The electrical control circuit fault adaptive diagnosis system based on deep learning according to claim 1, characterized in that: The preprocessing module specifically includes: The signal denoising unit, wherein the nonlinear filtering of the time series signal using the wavelet threshold denoising algorithm specifically includes: The original signal based on the data acquisition module is decomposed into sub-bands of different frequencies through multi-scale wavelet transform, and the approximate coefficients and detail coefficients are extracted for wavelet decomposition. The formula is: Among them, x(n) is the original signal, φ(t) is the wavelet basis function, is the scale function, cA j (k), cD j (k) are the approximation coefficient and detail coefficient of the jth layer, respectively, and k is the position index; Based on the extracted approximate coefficients and detail coefficients, the signal and noise are distinguished, and nonlinear filtering is performed to suppress the noise; reconstructing the denoised signal based on the processed wavelet coefficients; The data alignment unit, wherein the processing of the infrared image by the adaptive contrast enhancement algorithm specifically includes: Calculate the local mean and standard deviation of the infrared image by dividing the image into blocks; The gain factor is dynamically designed based on the standard deviation, nonlinear contrast stretching is performed on each sub-block, and overlapping areas are fused to avoid block effects; The standard dynamic range is obtained based on global normalization, which suppresses noise and over-enhancement while enhancing the features of the fault area.

4. The electrical control circuit fault adaptive diagnosis system based on deep learning according to claim 1, characterized in that: The feature extraction module specifically includes: The improved Transformer architecture specifically includes: Based on the data processed by the preprocessing module, one-dimensional causal convolution is performed, and the number of convolution output channels is d to obtain local features; The local features output by the temporal convolution are embedded in the original data and added element by element to form a fused feature. Generate query (Q), key (K), value (V) matrices from the fused data through linear transformation, and split query (Q), key (K), value (V) into h heads, each with a dimension of Concatenate the attention outputs of each head and project them back to the original input dimension through a linear layer; Add the output of multi-head attention to the original input to retain the initial information, alleviate the gradient disappearance, stabilize the training process, adjust the output distribution, and accelerate convergence; Nonlinear transformation is introduced through the feedforward network to refine features position by position; The output of the feedforward network is added to the normalized attention result, and the output distribution is stabilized by layer normalization as the final output of the encoder layer; Multiple encoder layers are repeatedly stacked to extract high-order temporal features layer by layer. The parameters of each layer are independent and local features are gradually integrated.

5. The electrical control circuit fault adaptive diagnosis system based on deep learning according to claim 1, characterized in that: The diagnostic model module specifically includes: A dual-channel deep residual network model, after acquiring electrical characteristic data and thermal-mechanical characteristic data, avoids gradient vanishing in the deep network based on residual connections. For each channel, the following residual block structure can be used: y=F(x,{W i })+x Among them, F(x,{W i }) is the convolution operation, W i is the convolution kernel; After the two channels process the electrical features and the thermal-mechanical features respectively, the outputs of the two channels are fused and the preliminary fault diagnosis results are obtained through the fully connected layer; An online incremental learning unit, which gradually updates the model using an online gradient descent method, updates the loss function based on the cross entropy loss for each new data batch in the form of: Among them, y i is the true label, is the predicted output of the model, and N is the total number of data; Based on the output of the dual-channel network and the adaptability of the incremental learning unit, the diagnosis results are classified according to the probability of the model output to obtain the final electrical control circuit fault diagnosis results and identify the fault type.

6. The deep learning-based electrical control circuit fault adaptive diagnosis system according to claim 1, characterized in that: The adaptive optimization module specifically includes: A transfer learning controller, wherein the transfer learning controller calculates the distribution difference between the new sample and the source domain sample by extracting the feature vector of the new fault sample, and selects a parameter migration strategy; A model structure dynamic adjustment unit, wherein the model structure dynamic adjustment unit optimizes the structure of the neural network, including the network depth and the size of the convolution kernel, through a genetic algorithm; Transfer learning selects a suitable model and fine-tunes it, dynamically adjusts units to optimize the network structure, and further fine-tunes the model according to new samples and structures until the diagnostic effect reaches the expected level.

7. The deep learning-based electrical control circuit fault adaptive diagnosis system according to claim 6, characterized in that: The transfer learning controller specifically includes: Build a library of pre-trained models for various device models, each of which establishes a hyperspherical decision boundary in parameter space; Calculate the Wasserstein distance between the new fault data distribution and each pre-trained model; The three models with the closest distance are selected for model parameter interpolation, and the interpolation weight is determined according to the inverse of the relative distance; The underlying feature extraction layers are frozen during the transfer process, and only the upper classifier is fine-tuned.

8. The deep learning-based electrical control circuit fault adaptive diagnosis system according to claim 6, characterized in that: The model structure dynamic adjustment unit specifically includes: The number of residual blocks is automatically increased or decreased based on the accuracy change rate of the validation set to perform network deep evolution. The decision formula for increase or decrease is: Among them, ΔL is the change in the number of residual blocks, η is the adjustment coefficient, and A val(t) -A val(t-1) is the difference between the validation set accuracy of the current stage and the previous stage, reflecting the change in model performance, σ A is the standard deviation of the accuracy of the validation set, which is used to normalize the adjustment amount; Optimize the convolution kernel size by evolutionary selection in the candidate set through genetic algorithm; The number of channels in each layer is dynamically adjusted based on the information entropy value of the feature map.

9. The electrical control circuit fault adaptive diagnosis system based on deep learning according to claim 1, characterized in that: The visual interaction module specifically includes: The heat map generator uses gradient activation mapping technology to superimpose the fault feature areas that the neural network focuses on onto the circuit topology map in a color mapping manner, annotate the abnormal current / voltage signal distribution in real time, and accurately locate the faulty components; The diagnostic decision tree interpreter parses the fault classification probability output by the model into an understandable logical reasoning chain through rule extraction and path backtracking, dynamically displays the decision-making process of "signal distortion → feature extraction → hierarchical judgment", and associates it with the historical fault case library.

10. An adaptive fault diagnosis method for electrical control circuits based on deep learning, characterized in that: The deep learning-based adaptive fault diagnosis system for electrical control circuits according to any one of claims 1 to 9 comprises: Through high-precision sensor arrays, infrared thermal imaging and vibration sensor groups, the circuit timing signal, temperature field distribution and mechanical vibration spectrum data are synchronously collected; The wavelet threshold noise reduction algorithm is used to denoise the time series signal, the multi-sensor time series deviation is eliminated through data alignment, and the infrared image quality is optimized using adaptive contrast enhancement; Based on the improved Transformer architecture, the temporal convolution operator is embedded to enhance the local feature perception ability of the self-attention mechanism and extract high-dimensional temporal features; Construct a dual-channel deep residual network to analyze electrical features and thermal-mechanical features respectively, output fault classification results after fusion, and update model parameters through online incremental learning; Based on the transfer learning controller, the pre-training parameters are transferred to adapt to the distribution characteristics of new fault samples, and the genetic algorithm is used to dynamically optimize the network depth and convolution kernel size to improve the generalization ability; Generate fault heat maps to locate abnormal areas, and combine decision tree interpreters to provide visual analysis and explainable reports based on diagnostic evidence.

Images