Variable environment fault diagnosis method, system and device based on depth subfield self-adaption and medium

Abstract

The invention relates to the technical field of fault diagnosis, and discloses a variable environment fault diagnosis method, system and device based on depth subfield self-adaption and a medium, and the method comprises the steps: collecting vibration data of a rotating machine, and carrying out the data preprocessing; a channel attention mechanism and a space attention mechanism are combined, sub-domain adaptive optimization based on local maximum mean value difference is introduced, and a fault diagnosis model is constructed through feature extraction, feature distribution difference minimization and classification loss optimization; and inputting the processed vibration data into a model for model training and fault diagnosis, and outputting the probability of various faults to which the vibration of the rotating machine belongs to obtain a fault diagnosis result. According to the fault diagnosis model, the feature information of each fault category is extracted by using the local maximum mean value difference, different sub-domains are quickly aligned, the learning of flow features between different sub-domains is accelerated by using a channel and space attention mechanism, and the fault state is quickly and accurately identified.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and in particular to a fault diagnosis method, system, device and medium for variable environments based on deep subdomain adaptive methods. Background Technology

[0002] As industry becomes increasingly modernized, complex and precise rotating machinery is becoming more prevalent. The development of deep learning technology has improved the classification performance of fault diagnosis for rotating components in rotating machinery. Traditional methods for fault diagnosis of rotating machinery generally use single-condition feature data through networks such as convolutional neural networks. However, in many practical applications, this assumption does not hold true. For example, the complexity of the working environment means that a single feature cannot reflect the complex damage process of rotating machinery, and the low computational efficiency of convolutional neural networks makes it difficult to accurately predict the fault damage state of rotating machinery in variable environments.

[0003] Existing methods either employ convolutional neural networks or their fusion networks, but these methods do not take into account the variable operating conditions in complex environments. There is an urgent need to research methods for rotating machinery fault diagnosis based on convolutional neural networks, to address the poor training performance of convolutional neural networks and their inability to meet practical operational requirements, thereby improving the rotating machinery fault diagnosis system. Summary of the Invention

[0004] In view of the aforementioned existing problems, this invention is proposed. Therefore, this invention provides a method, system, device, and medium for fault diagnosis in variable environments based on deep subdomain adaptation to solve the above problems.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a fault diagnosis method for variable environments based on deep subdomain adaptive methods, including: collecting vibration data of rotating machinery and performing data preprocessing; By combining channel attention and spatial attention mechanisms, and introducing subdomain adaptive optimization based on local maximum mean difference, a fault diagnosis model is constructed through feature extraction, minimizing feature distribution differences, and optimizing classification loss. Based on the fault diagnosis model, the processed vibration data is input into the model for model training and fault diagnosis, and the probability of each type of fault in the rotating machinery vibration is output to obtain the fault diagnosis result.

[0006] As a preferred embodiment of the multi-environment fault diagnosis method based on deep sub-domain adaptive design described in this invention, the method includes: collecting vibration data of rotating machinery and performing data preprocessing, which includes: The original vibration signals of rotating machinery under different operating conditions are collected by sensors at a fixed sampling frequency to form source domain and target domain data. The original vibration signals are then processed to convert the time domain signal into a frequency domain signal.

[0007] As a preferred embodiment of the multi-environment fault diagnosis method based on deep subdomain adaptiveness according to the present invention, the fault diagnosis model is a deep subdomain adaptive model that integrates convolutional block attention mechanism, including a one-dimensional convolution module, a convolutional block attention mechanism module, a local maximum mean difference module, and a connection layer decoding module. The one-dimensional convolutional module performs preliminary feature extraction using a pre-trained ResNet-50 network; The convolutional block attention mechanism module generates key region feature maps by combining spatial attention and channel attention; The local maximum mean difference module aligns the subdomain distribution of the same type of fault by the relationship between subdomains in different domains, thereby optimizing the distribution difference; The fully connected layer decoding module maps the extracted features to the predicted probability of the fault category and outputs the fault probability of each category. The beneficial effect of this preferred scheme is that the model can enhance its response to important feature channels. Combined with channel attention, the model can focus more on key regions. The local maximum mean difference is aligned by category, which improves cross-domain adaptability.

[0008] As a preferred embodiment of the multi-environment fault diagnosis method based on deep sub-domain adaptive design described in this invention, the local maximum mean difference module includes: We introduce subdomain adaptation based on local maximum mean difference to optimize the neighborhood adaptation loss. This is achieved by compressing the feature map into a feature vector and regenerating the feature kernel in the Hilbert space. The function performs convolution on the features, predicts the probability of the features in the target domain, obtains the optimization parameters of the local maximum mean difference function, optimizes the feature difference between each subdomain, and optimizes the model through stochastic gradient descent. The advantage of this preferred solution is that it calculates the difference in the characteristic distribution of the same type of fault in the source domain and the target domain, and by minimizing this difference, the model can accurately classify the fault in the target domain as well.

[0009] As a preferred embodiment of the multi-environment fault diagnosis method based on deep sub-domain adaptation described in this invention, the feature representation update function of the channel attention of the convolutional block attention mechanism module is expressed as: in, Feature representation of channel attention, Representation of features Average pooling, Representation of features Max pooling, , Indicates weight, This represents the sigmoid function.

[0010] As a preferred embodiment of the multi-environment fault diagnosis method based on deep sub-domain adaptation described in this invention, the feature representation update function of the spatial attention of the convolutional block attention mechanism module is expressed as: in, Feature representation of spatial attention, This represents the sigmoid function. The size of the convolution operation is The filter, , This represents the function that performs max pooling and average pooling along the channel dimension in spatial attention.

[0011] As a preferred embodiment of the multi-environment fault diagnosis method based on deep sub-domain adaptive design described in this invention, the local maximum mean difference module further includes: Measures of difference in local maximum mean Represented as: in, and This represents samples from the source and target domains. and It is the distribution of the source and target domain datasets. H represents the fault category, and H represents the feature core assigned to it. The regenerated Hilbert space; By measuring local distribution differences, the feedforward neural network optimizes samples of different categories using weight parameters. The optimization formula is expressed as: in, Indicates the introduction of optimization parameters The measure of the difference between the local maximum means. and Let c represent the weights of the categories c in the source and target domains, respectively. , This represents a data sample representing the source and target domains. Represents the feature mapping of the sample. , This represents the dataset containing the source and target domains.

[0012] Secondly, the present invention provides a fault diagnosis system for variable environments based on deep subdomain adaptive methods, comprising: The data acquisition module is used to collect vibration data of rotating machinery and perform data preprocessing. The model building module combines channel attention and spatial attention mechanisms and introduces subdomain adaptive optimization based on local maximum mean difference. Through feature extraction, minimizing feature distribution differences and optimizing classification loss, it constructs a deep subdomain adaptive model that integrates convolutional block attention mechanism. The output module is used to input the processed vibration data into the deep sub-domain adaptive model based on the fused convolutional block attention mechanism for model training and fault diagnosis, output the probability of various faults of rotating machinery vibration, and obtain the fault diagnosis result.

[0013] Thirdly, the present invention provides a computer device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the method for diagnosing faults in a variable environment based on deep sub-domain adaptation.

[0014] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the described method for fault diagnosis of a variable environment based on deep sub-domain adaptation.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: The fault diagnosis model of the present invention uses the local maximum mean difference to extract the feature information of each fault category and quickly aligns different subdomains. It also uses channel and spatial attention mechanisms to accelerate the learning of traffic features between different subdomains, which can make full use of transmission performance and achieve fast and accurate fault state identification on a small target domain training dataset. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1This is a schematic diagram of the overall process of a fault diagnosis method for variable environments based on deep subdomain adaptive methods according to an embodiment of the present invention.

[0018] Figure 2 This is a model network structure diagram of a fault diagnosis method for variable environments based on deep sub-domain adaptive methods according to an embodiment of the present invention.

[0019] Figure 3 This is a schematic diagram of the channel attention mechanism of a multi-environment fault diagnosis method based on deep sub-domain adaptive methods according to an embodiment of the present invention.

[0020] Figure 4 This is a schematic diagram of the spatial attention mechanism of a multi-environment fault diagnosis method based on deep sub-domain adaptation, as described in an embodiment of the present invention.

[0021] Figure 5 This figure shows the comparative experimental results of a fault diagnosis method for variable environments based on deep sub-domain adaptive methods according to an embodiment of the present invention.

[0022] Figure 6 This is a schematic diagram of the cross-device migration confusion matrix of a fault diagnosis method for variable environments based on deep sub-domain adaptation, as described in an embodiment of the present invention. Detailed Implementation

[0023] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0024] Example 1, referring to Figures 1-4 As one embodiment of the present invention, a fault diagnosis method for variable environments based on deep sub-domain adaptive methods is provided, such as... Figure 1 and Figure 2 As shown, it includes: S100: Collects vibration data of rotating machinery and performs data preprocessing; S200 combines channel attention and spatial attention mechanisms, and introduces subdomain adaptive optimization based on local maximum mean difference. Through feature extraction, minimizing feature distribution differences and optimizing classification loss, it constructs a deep subdomain adaptive model that integrates convolutional block attention mechanism. S300 is a deep subdomain adaptive model based on a fusion convolutional block attention mechanism. It inputs the processed vibration data into the model for model training and fault diagnosis, outputs the probability of various faults in rotating machinery vibration, and obtains the fault diagnosis results.

[0025] In a preferred embodiment, collecting vibration data of the rotating machinery and performing data preprocessing includes: The original vibration signals of rotating machinery under different operating conditions are collected by sensors at a fixed sampling frequency to form source domain and target domain data. The original vibration signals are then processed to convert the time domain signal into a frequency domain signal.

[0026] Specifically, vibration data collected by various sensors on rotating machinery is divided into source domain data and target domain data. The vibration data is the original vibration signal collected at a fixed sampling frequency. After fast Fourier transform, the collected time domain signal is converted into a frequency domain signal, which is used as the input sample of the model.

[0027] It should be noted that the original sampling data of existing rotating machinery vibration signals is too simplistic. This invention solves this problem by extracting shallow features in the time and frequency domains of the original vibration signals, which can fully express the degradation information of rotating machinery.

[0028] In one alternative implementation, the original vibration signal is processed to convert the time-domain signal into a frequency-domain signal. It can also be processed by wavelet transform, using scalable and translational basis functions (wavelet mother functions) to simultaneously provide the time and frequency information of the signal. By scaling the wavelet to match different frequency components, time-frequency domain information is obtained.

[0029] In another alternative implementation, the original vibration signal is processed to convert the time-domain signal into a frequency-domain signal. Alternatively, it can be processed using the Hilbert-Huang transform to adaptively decompose the complex original signal into a series of finite intrinsic mode functions with frequencies ranging from high to low, and a residual term. A Hilbert transform is then performed on each intrinsic mode function to obtain the instantaneous frequency and instantaneous amplitude of the signal, ultimately forming the Hilbert spectrum.

[0030] This invention diagnoses fault types by constructing a fault diagnosis model. Preferably, it outputs fault diagnosis results by constructing a deep subdomain adaptive model that integrates convolutional block attention mechanism.

[0031] In a preferred embodiment, the deep subdomain adaptive model that integrates the convolutional block attention mechanism includes a one-dimensional convolution module, a convolutional block attention mechanism module, a local maximum mean difference module, and a connection layer decoding module. The one-dimensional convolutional module performs initial feature extraction using a pre-trained ResNet-50 network; The convolutional block attention mechanism module combines spatial attention and channel attention to generate key region feature maps; The local maximum mean difference module aligns the subdomain distribution of the same type of fault by analyzing the relationships between subdomains in different domains, thereby optimizing the distribution difference. The fully connected layer decoding module maps the extracted features to the predicted probabilities of fault categories and outputs the fault probability for each category.

[0032] Specifically, the processed data is fed into a pre-trained ResNet50 convolutional neural network via a one-dimensional convolutional module. After feature extraction, the data is output to the convolutional block attention mechanism module for further feature extraction. This model uses the ResNet50 network as a foundation for initial signal feature extraction. Feature extraction is achieved by setting different parameters. Feature extraction is based on convolutional operations and backpropagation feedback of multiple convolutional and pooling layers to ultimately extract feature representations of different health states in the vibration signal. Then, the convolutional block attention mechanism and local maximum mean difference are used for further feature extraction and prediction.

[0033] The convolutional block attention mechanism module combines spatial and channel attention, including a channel attention module and a spatial attention module. The feature map output by the one-dimensional convolution module is input into the channel attention module, and a channel attention map is generated through average pooling and max pooling operations. The feature map weighted by channel attention is input into the spatial attention module, which generates a spatial attention map through average pooling and max pooling (along the channel axis) and then through a 7x7 convolutional layer, focusing on key regions in the spatial location of the feature map.

[0034] In a preferred embodiment, the channel attention module generates a channel attention map using the inter-channel relationships between features. Since each channel of the feature map is treated as a feature detector, channel attention focuses on capturing the most representative features. The internal structure of the channel attention mechanism is as follows: Figure 3 As shown, the feature representation update function for channel attention is expressed as: in, Feature representation of channel attention, Representation of features Average pooling, Representation of features Max pooling, , Indicates weight, This represents the sigmoid function.

[0035] Specifically, to compute spatial attention, the spatial attention module first performs MaxPool and AvgPool operations along the channel axis and concatenates them to generate effective feature descriptors. Then, convolutional layers are applied to generate the spatial attention map, represented as: ,in, The feature representation for spatial attention is shown below, where R represents all feature maps, and H and W represent the height and width of the feature maps. The internal structure of the channel attention mechanism is as follows: Figure 4 As shown.

[0036] In a preferred embodiment, the feature representation update function for spatial attention is expressed as: in, Feature representation of spatial attention, This represents the sigmoid function. The size of the convolution operation is The filter, , This represents the function that performs max pooling and average pooling along the channel dimension in spatial attention.

[0037] It should be noted that the spatial attention module generates a spatial attention map by utilizing the spatial relationships between features. The spatial attention mechanism focuses more on the location of the most representative features than channel attention. Combining spatial attention with channel attention allows the neural network to quickly obtain key information.

[0038] In a preferred embodiment, the local maximum mean difference module includes: We introduce subdomain adaptation based on local maximum mean difference to optimize the neighborhood adaptation loss. This is achieved by compressing the feature map into a feature vector and regenerating the feature kernel in the Hilbert space. The function performs convolution on the features, predicts the probability of the features in the target domain, obtains the optimization parameters of the local maximum mean difference function, optimizes the feature differences between each subdomain, and optimizes the model through stochastic gradient descent.

[0039] Specifically, the Local Maximum Mean Discrepancy (LMMD) module aligns the distributions of relevant subdomains by utilizing the relationships between subdomains in different domains to match global and local distributions. It introduces subdomain adaptation based on LMMD to optimize the domain adaptation loss.

[0040] The general maximum mean difference (MMD) algorithm defines the following difference metric, expressed as: in, It is assigned to the feature kernel The reproducing kernel Hilbert space (RKHS). Represents the expected value of the characteristic distribution. This represents mapping the original samples to the feature maps of the RKHS. The data distributions in the source and target domains are represented respectively. , Represents samples from the source and target domains, feature kernels The function is represented as: in, When representing the convolution operation between vectors, the subdomain relationship should be fully considered when aligning the distribution of subdomains of the same class in the source and target domains. This invention optimizes the difference by using an improved local maximum mean difference algorithm.

[0041] In a preferred embodiment, for local alignment, the difference measure of the local maximum mean difference Represented as: in, and This represents samples from the source and target domains. and It is the distribution of the source and target domain datasets. H represents the fault category, and H represents the feature core assigned to it. The regenerated Hilbert space; Measuring local distribution differences through weighting parameters (Weight parameters are distinguished according to different domains, for example, the source domain is s, the target domain is t, and generally it is represented by x.) In feedforward neural networks (such as ResNet, VGGNet, and AlexNet), the optimization formula for different classes of samples is expressed as: in, Indicates the introduction of optimization parameters The measure of the difference between the local maximum means. and Let c represent the weights of the categories c in the source and target domains, respectively. , This represents a data sample representing the source and target domains. Represents the feature mapping of the sample. , This represents the dataset containing the source and target domains.

[0042] Specifically, the weight parameters are calculated using the following function: in, This represents the weight of category c of data x. This indicates the data samples and labels for this field. , The labels represent the predicted and actual values, and D represents the dataset.

[0043] To better utilize feedforward neural networks to optimize the LMMD algorithm, we activate the aforementioned LMMD function in different layers. Given the activation function of layer 1, the feature kernel... The representation is converted to: ,in, This represents the feature sample of the l-th round. The first layer is the initial definition, and in subsequent training processes, the feature kernel k-function formula mentioned above is used as the mapping representation. The following equation is used as the loss function. , represented as: in, , This represents the number of samples in the source and target domains.

[0044] Furthermore, the fully connected layer of the fully connected layer decoding module is a decoder that maps features to predicted labels. The fully connected layer connects all nodes of the spatial attention module, that is, it integrates all extracted features, and then the output value is converted into class probability by activating the SoftMax function.

[0045] In one alternative implementation, the fault diagnosis model can also be constructed using traditional machine learning, such as support vector machines or random forests. A large number of features are manually extracted from the original vibration signal. These features include time-domain features, frequency-domain features, and time-frequency-domain features. Principal component analysis, linear discriminant analysis, and other methods are used to select the most effective parts from the large number of features. The selected features are then input into the traditional machine learning model for training and classification, and the fault results are output.

[0046] In another alternative implementation, the fault diagnosis model can also be constructed using a recurrent neural network. Vibration signals are essentially time series, and RNNs, LSTM long short-term memory networks, GRU gated recurrent units, etc., are used to process sequence data, which can capture the long-term dependencies of signals in the time dimension and improve the accuracy of fault diagnosis.

[0047] Furthermore, a deep subdomain adaptive model based on a fusion convolutional block attention mechanism is used to input the processed vibration data into the model for model training and fault diagnosis. The model outputs the probability of each type of fault in the rotating machinery vibration, yielding fault diagnosis results, including: First, a deep sub-domain adaptive model incorporating a convolutional block attention mechanism is constructed. The source and target domains of the original vibration signals are directly imported from the processed dataset. This model utilizes a ResNet-50 network, a pre-trained model with powerful feature extraction and processing capabilities. The multi-layer neural network pre-trained model uses the original features and epoch time to reduce classification errors caused by feature differences. To obtain feature maps with channel attention in the first convolutional layer, the original features are input into the channel attention module to extract feature maps. These are then compressed into feature vectors, and convolution calculations are performed in RKHS using the feature kernels in RKHS. Features are convolutionally processed, and then the probability in the target domain is predicted through a fully connected (FC) layer to obtain the optimized parameters of the LMMD function. Following the above steps, the feature differences between each subdomain are optimized within the LMMD module. The SGD algorithm is then used to train the LMMD module for gradient descent. To accelerate the construction of a deep subdomain adaptive model with a fused convolutional block attention mechanism for identifying fault diagnoses under different operating conditions, the differences between features in each subdomain of the source and target domains are fitted. A spatial attention module is added before the last convolutional layer of the pre-trained neural network. Similar to the feature extraction and processing steps of the channel attention module, a feature map containing spatial attention is computed and compressed into a feature vector. The transformed features are then used to output the fault state category. The output of the SoftMax layer is the probability of belonging to each diagnostic type; finally, the category with the highest predicted probability is selected as the identified fault state.

[0048] The above is an illustrative scheme of a fault diagnosis method for variable environments based on deep subdomain adaptation, as described in this embodiment. It should be noted that the technical solution of this fault diagnosis system based on deep subdomain adaptation for variable environments belongs to the same concept as the technical solution of the aforementioned fault diagnosis method based on deep subdomain adaptation for variable environments. Details not described in detail in this embodiment can be found in the description of the aforementioned fault diagnosis method based on deep subdomain adaptation for variable environments.

[0049] Example 2: This example provides a fault diagnosis system for variable environments based on deep subdomain adaptation, including: The data acquisition module is used to collect vibration data of rotating machinery and perform data preprocessing. The model building module combines channel attention and spatial attention mechanisms and introduces subdomain adaptive optimization based on local maximum mean difference. Through feature extraction, minimizing feature distribution differences and optimizing classification loss, it constructs a deep subdomain adaptive model that integrates convolutional block attention mechanism. The output module is used for deep subdomain adaptive models based on the fusion convolutional block attention mechanism. It inputs the processed vibration data into the model for model training and fault diagnosis, outputs the probability of each type of fault in the rotating machinery vibration, and obtains the fault diagnosis results.

[0050] This embodiment also provides a computer device suitable for fault diagnosis in variable environments based on deep subdomain adaptive methods, including: The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement a fault diagnosis method for a variable environment based on deep subdomain adaptation, as proposed in the above embodiments.

[0051] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements a fault diagnosis method for variable environments based on deep subdomain adaptive methods as proposed in the above embodiments.

[0052] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for diagnosing faults in a variable environment based on deep subdomain adaptation proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0053] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0054] Example 3, refer to Table 1 and Figures 5-6As an embodiment of the present invention, a fault diagnosis method for variable environments based on deep subdomain adaptive methods is provided. To verify its beneficial effects, it is scientifically demonstrated through economic benefit calculations and simulation experiments.

[0055] Experiments were conducted on the Western Reserve University bearing dataset and the Paderborn bearing dataset based on the deep subdomain adaptive network proposed in this example, which incorporates a convolutional block attention mechanism. The method presented in this example demonstrates that, compared to Support Vector Machines (SVM), Domain Adversarial Neural Networks (DANN), and Wide First-layer Kernel Convolutional Neural Networks (WDCNN), bearing fault diagnosis under varying target domains using the deep subdomain adaptive network can fully utilize transmission performance and achieve fast and accurate fault state identification on a smaller target domain training dataset. Further improvements to the deep subdomain adaptive network, which incorporates a convolutional block attention mechanism, are made by considering different signal endpoints, online diagnosis, and real-time diagnosis, to enable bearing diagnosis in practical applications.

[0056] In this example, the datasets used in the experiment are the Western Reserve University bearing dataset and the Paderborn bearing dataset. Four sets of cross-device experiments were conducted to achieve cross-device bearing fault diagnosis: D1→F1, D1→F2, D2→F1, and D2→F2. Taking D1→F1 as an example, D1 is the source domain data with labels, and F1 is the target domain data without labels. Transfer learning from D1 to F1 is performed, that is, the feature representation learned in D1 is mapped to F1 for classification prediction. The source domain data D1 consists of 6000 samples from the Western Reserve University bearings at 1hp and 1772rpm, representing three different bearing states at the fan end. D2 consists of 6000 samples from the Western Reserve University bearings at 2hp and 1750rpm, representing three different bearing states at the fan end. Considering the significant differences between the Paderborn bearing dataset and the Western Reserve University bearing dataset, a dataset created from artificial damage within the Paderborn bearing dataset was selected, as it is more similar to the Western Reserve University bearing data. Because the data on artificial damage includes both electrical discharge machining (EDM) and manual etching, target domain datasets F1 and F2 were created. F1 consists of 7800 samples from the bearing dataset covering three bearing states under EDM and case number 2. F2 consists of 7800 samples from the Paderborn bearing dataset covering three bearing states under case number 2 under manual etching, as shown in Table 1.

[0057] Table 1 Cross-device experimental data table

[0058] To test the fault diagnosis capability of this invention, the training and test sets were partitioned using the leave-one-out method for verification. The proposed model was compared with SVM, WDCNN, DANN, and Deep Subdomain Adaptation Net (DSAN). The final accuracy and label confusion matrix experimental results are as follows: Figure 5 and Figure 6 As shown ( Figure 6 (D1-F2 are on the left and D2-F1 are on the right). The prediction results show that the method of the present invention is superior to other methods in most rotating machinery.

[0059] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A fault diagnosis method for variable environments based on deep subdomain adaptation, characterized in that, include: Collect vibration data of rotating machinery and perform data preprocessing; By combining channel attention and spatial attention mechanisms, and introducing subdomain adaptive optimization based on local maximum mean difference, a fault diagnosis model is constructed through feature extraction, minimizing feature distribution differences, and optimizing classification loss. Based on the fault diagnosis model, the processed vibration data is input into the model for model training and fault diagnosis, and the probability of each type of fault in the rotating machinery vibration is output to obtain the fault diagnosis result.

2. The fault diagnosis method for variable environments based on deep sub-domain adaptation as described in claim 1, characterized in that, Collecting vibration data from rotating machinery and performing data preprocessing includes: The original vibration signals of rotating machinery under different operating conditions are collected by sensors at a fixed sampling frequency to form source domain and target domain data. The original vibration signals are then processed to convert the time domain signal into a frequency domain signal.

3. The fault diagnosis method for variable environments based on deep sub-domain adaptation as described in claim 1, characterized in that, The fault diagnosis model is a deep subdomain adaptive model that integrates a convolutional block attention mechanism, including a one-dimensional convolution module, a convolutional block attention mechanism module, a local maximum mean difference module, and a connection layer decoding module. The one-dimensional convolutional module performs preliminary feature extraction using a pre-trained ResNet-50 network; The convolutional block attention mechanism module generates key region feature maps by combining spatial attention and channel attention; The local maximum mean difference module aligns the subdomain distribution of the same type of fault by the relationship between subdomains in different domains, thereby optimizing the distribution difference; The fully connected layer decoding module maps the extracted features to the predicted probabilities of fault categories and outputs the fault probability for each category.

4. The fault diagnosis method for variable environments based on deep sub-domain adaptive methods as described in claim 3, characterized in that, The local maximum mean difference module includes: We introduce subdomain adaptation based on local maximum mean difference to optimize the neighborhood adaptation loss. This is achieved by compressing the feature map into a feature vector and regenerating the feature kernel in the Hilbert space. The function performs convolution on the features, predicts the probability of the features in the target domain, obtains the optimization parameters of the local maximum mean difference function, optimizes the feature differences between each subdomain, and optimizes the model through stochastic gradient descent.

5. The fault diagnosis method for variable environments based on deep sub-domain adaptation as described in claim 3, characterized in that, The feature representation update function for the channel attention of the convolutional block attention mechanism module is expressed as: in, Feature representation of channel attention, Representation of features Average pooling, Representation of features Max pooling, , Indicates weight, This represents the sigmoid function.

6. The fault diagnosis method for variable environments based on deep sub-domain adaptation as described in claim 5, characterized in that, The feature representation update function for the spatial attention of the convolutional block attention mechanism module is expressed as follows: in, Feature representation of spatial attention, This represents the sigmoid function. The size of the convolution operation is The filter, , This represents the function that performs max pooling and average pooling along the channel dimension in spatial attention.

7. The fault diagnosis method for variable environments based on deep sub-domain adaptive methods as described in claim 4, characterized in that, The local maximum mean difference module also includes: Measures of difference in local maximum mean Represented as: in, and This represents samples from the source and target domains. and It is the distribution of the source and target domain datasets. H represents the fault category, and H represents the feature core assigned to it. The regenerated Hilbert space; By measuring local distribution differences, the feedforward neural network optimizes samples of different categories using weight parameters. The optimization formula is expressed as: in, Indicates the introduction of optimization parameters The measure of the difference between the local maximum means. and Let c represent the weights of the categories c in the source and target domains, respectively. , This represents a data sample representing the source and target domains. Represents the feature mapping of the sample. , This represents the dataset containing the source and target domains.

8. A fault diagnosis system for variable environments based on deep subdomain adaptation, employing the fault diagnosis method for variable environments based on deep subdomain adaptation as described in any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to collect vibration data of rotating machinery and perform data preprocessing. The model building module combines channel attention and spatial attention mechanisms and introduces subdomain adaptive optimization based on local maximum mean difference. Through feature extraction, minimizing feature distribution differences and optimizing classification loss, it constructs a deep subdomain adaptive model that integrates convolutional block attention mechanism. The output module is used to input the processed vibration data into the deep sub-domain adaptive model based on the fused convolutional block attention mechanism for model training and fault diagnosis, output the probability of various faults of rotating machinery vibration, and obtain the fault diagnosis result.

9. A computer device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the multi-environment fault diagnosis method based on deep sub-domain adaptive as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores computer-executable instructions, which, when executed by a processor, implement the steps of the deep sub-domain adaptive multi-environment fault diagnosis method as described in any one of claims 1 to 7.

Images