Rolling bearing fault targeted migration diagnosis method and system across working conditions

Abstract

The application provides a kind of cross-condition rolling bearing fault targeting migration diagnosis method and system, solve the problem that traditional rolling bearing fault diagnosis algorithm is difficult to extract network deep feature information in source domain and target domain, cannot realize effective cross-domain fault diagnosis.This application uses feature encoder to accurately extract high-dimensional mapping features of the signal from the input rolling bearing vibration signal;Further input the feature to the graph construction layer, mine the deep features of the data, and model the instance graph using the multi-channel kernel graph convolution network;Use the training based on difference and confrontation to minimize the distance between the source domain and target domain distribution, and the classifier uses the extracted domain invariant feature to complete cross-domain fault recognition.Compared with other methods, under the cross-condition of rolling bearing, the deep features can be better extracted for cross-domain transmission, greatly improving the diagnosis accuracy.

Description

Technical Field

[0001] This invention relates to the field of bearing fault diagnosis technology, and in particular to a method and system for targeted migration diagnosis of rolling bearing faults across operating conditions. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Rolling bearings are critical components in most rotating machinery. Due to harsh working environments, changing operating conditions, and prolonged overload operation, they frequently fail, leading to overall machine failure and economic losses. Therefore, monitoring the operating status of rolling bearings under various operating conditions is of great significance.

[0004] In recent years, with the emergence of massive amounts of data, data-driven methods based on deep learning have attracted much attention. Unlike traditional shallow network architectures, they are implemented by stacking multiple layers of nonlinear processing units, providing an end-to-end solution, and have been the subject of numerous studies. Jia et al. proposed a deep neural network (DNN) based on SAE for identifying faults in motors and gearboxes. Ding et al. proposed using convolutional neural networks to mine multi-scale features of energy fluctuations from wavelet packet energy images for fault diagnosis of spindle bearings. Chen et al. proposed an automatic feature learning neural network that uses the original vibration signal as input and uses two CNNs with different kernel sizes to automatically extract signal features of different frequencies from the original data. Then, based on the learned features, LSTM is used to identify the fault type. While these deep learning algorithms have achieved good results in the field of rolling bearing fault diagnosis under constant operating conditions, they struggle to extract inter-domain differences under complex cross-operating conditions, leading to a decrease in the model's generalization performance.

[0005] Transfer learning, as a method to reduce cross-domain feature distribution differences, provides a new approach to establishing knowledge transfer from labeled data in the source domain to unlabeled data in the target domain, effectively solving the problem of inter-domain distribution differences under different working conditions. In the field of intelligent fault diagnosis, many transfer learning methods have been proposed to solve cross-domain diagnosis problems, which can be broadly categorized into instance-based, model-based, and feature-based methods. Xiao et al. used TrAdaBoost to adjust the weight factor of each training sample to enhance the diagnostic performance of the fault classifier. Wang et al. proposed a conditional MMD based on estimated pseudo-labels to shorten the distribution distance in bearing fault diagnosis. By minimizing the MMD loss, marginal and conditional distributions are aligned simultaneously in multiple layers. Han et al. proposed a deep adversarial convolutional neural network (DACNN) that utilizes an adversarial loss function to reduce inter-domain differences, improving the generalization performance of gearbox and motor fault diagnosis. Li et al. used graph data with topological structure as input and employed a graph convolutional network (GCN) with more effective data relationship mining and stronger feature representation for mechanical fault diagnosis, achieving excellent performance. It can be seen that category labels, domain labels, and data structure information play an important role in reducing the differences between the source and target domains, and they should complement and enhance each other.

[0006] However, the inventors discovered that existing methods only consider two types of information, source domain and target domain, and do not integrate data structure into deep neural network models. The data distribution varies greatly under different working conditions, resulting in low fault identification accuracy and insufficient generalization performance. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a method and system for targeted migration diagnosis of rolling bearing faults across operating conditions. Under the cross-operating conditions of rolling bearings, it can better extract deep features for cross-domain transmission, greatly improving the diagnostic accuracy.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] The first aspect of this invention provides a method for targeted migration diagnosis of rolling bearing faults across operating conditions.

[0010] A method for targeted migration diagnosis of rolling bearing faults across operating conditions includes the following steps:

[0011] Obtain the vibration signal of the rolling bearing;

[0012] Deep, high-dimensional features in vibration signals are extracted using a feature encoder, graph construction layer, and multi-channel kernel graph convolutional network.

[0013] Based on the extracted deep high-dimensional features and classifier, the rolling bearing fault diagnosis results are obtained;

[0014] Among them, classification loss is obtained based on the deep high-dimensional features obtained by combining the source domain training set with the classifier, structural difference loss is obtained based on the deep high-dimensional features obtained by the source domain training set and the target domain training set, and adversarial loss is obtained based on the deep high-dimensional features obtained by the source domain training set and the target domain training set and the adversarial network.

[0015] The overall loss function is the sum of the classification loss, the product of the first parameter and the structural difference loss, and the product of the second parameter and the adversarial loss. Based on the overall objective function, the parameters of the feature extractor, classifier, and discriminator are optimized through the backpropagation algorithm.

[0016] As an optional implementation of the first aspect of the present invention, the graph construction layer is used to obtain the adjacency matrix, including:

[0017] Based on the feature encoder network, a high-dimensional feature map, X = G(x), is obtained from the sample data;

[0018] The extracted high-dimensional feature maps are input into a linear layer and represented after passing through a softmax layer as follows:

[0019] The adjacency matrix A is obtained by performing matrix multiplication between the features of the linear layer and its transpose.

[0020] Construct edge relationships using the KNN algorithm:

[0021] As a further limitation of the first aspect of the present invention, the multi-channel kernel graph convolutional network includes:

[0022]

[0023]

[0024] Where X represents the input and A represents the adjacency matrix. G represents the trainable weight matrix, and G represents the multi-channel kernel map convolution operation. Representing the kth i The high-dimensional feature representation of each channel in the Lth layer, [·] represents feature concatenation, and H represents the output feature after passing through the multi-channel kernel graph convolutional network.

[0025] As one optional implementation of the first aspect of the present invention, the cross-entropy loss L C ,include:

[0026]

[0027] in, E represents the prediction result of the classifier, and E represents the expected value. For source domain samples, Tag it.

[0028] As one optional implementation of the first aspect of the present invention, the structural difference loss L s ,include:

[0029]

[0030] in, and Let φ and Ω represent the feature maps of the i-th source domain sample and the j-th target domain sample after passing through the feature extractor, respectively. Let φ represent the nonlinear feature map. k The distance metric from the extracted features to the regenerated kernel Hilbert space RKHS is represented by a convex combination of m kernels k. u To efficiently estimate the mapping: Where, α u These are weighted parameters for different cores, and E represents the expected value in mathematics. For source domain samples, For the target domain sample.

[0031] As one optional implementation of the first aspect of the present invention, combating loss L AD ,include:

[0032]

[0033] Where D(·) is the feature output after passing through the discriminator. and Let E represent the feature maps of the i-th source domain sample and the j-th target domain sample after passing through the feature extractor, respectively, and let E represent the expected value. For source domain samples, For the target domain sample.

[0034] As an optional implementation of the first aspect of the present invention, the parameters of the feature extractor, classifier, and discriminator are optimized using a backpropagation algorithm, including:

[0035]

[0036] in, Represents the partial differential operator, η represents the learning rate, and θ represents the partial differential operator. F θ represents the parameters of the feature extractor. C θ represents the parameters of the classifier. D L represents the parameters of the discriminator. C For cross-entropy loss, L s For structural difference loss, LAD To combat the losses.

[0037] A second aspect of the present invention provides a rolling bearing fault targeted migration diagnosis system that covers multiple operating conditions.

[0038] A cross-condition rolling bearing fault targeted migration diagnostic system, comprising:

[0039] The data acquisition module is configured to acquire the vibration signal of the rolling bearing;

[0040] The feature extraction module is configured to extract deep, high-dimensional features from vibration signals through a feature encoder, a graph construction layer, and a multi-channel kernel graph convolutional network.

[0041] The fault diagnosis module is configured to obtain rolling bearing fault diagnosis results based on the extracted deep high-dimensional features and classifier.

[0042] Among them, classification loss is obtained based on the deep high-dimensional features obtained by combining the source domain training set with the classifier, structural difference loss is obtained based on the deep high-dimensional features obtained by the source domain training set and the target domain training set, and adversarial loss is obtained based on the deep high-dimensional features obtained by the source domain training set and the target domain training set and the adversarial network.

[0043] The overall loss function is the sum of the classification loss, the product of the first parameter and the structural difference loss, and the product of the second parameter and the adversarial loss. Based on the overall objective function, the parameters of the feature extractor, classifier, and discriminator are optimized through the backpropagation algorithm.

[0044] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of the cross-condition rolling bearing fault targeted migration diagnosis method as described in the first aspect of the present invention.

[0045] A fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the cross-condition rolling bearing fault targeted migration diagnosis method as described in the first aspect of the present invention.

[0046] Compared with the prior art, the beneficial effects of the present invention are:

[0047] 1. This invention uses a feature encoder to adaptively extract signal features from the input signal, and uses a graph construction layer to obtain the data structure in the features captured by the feature encoder, thereby constructing an instance graph. Furthermore, a multi-channel kernel graph convolutional network is applied to model it, further mining the high-dimensional features of the signal, thus solving the problem of difficult deep feature extraction in the source and target domains under different working conditions.

[0048] 2. In view of the problem of large data differences between the source domain and target domain of rolling bearing vibration signal under cross-operating conditions, the present invention adopts a loss function based on data structure difference and an adversarial inter-domain alignment loss function to jointly reduce the inter-domain difference. At the same time, the classifier uses the extracted domain-invariant features to complete cross-domain fault identification.

[0049] 3. This invention solves the problems of difficulty in obtaining labeled data in industrial scenarios, large differences in data distribution under different working conditions, resulting in low fault identification accuracy and insufficient generalization performance. This invention can not only extract deep high-dimensional features to reduce inter-domain differences, but also obtain better diagnostic performance. Attached Figure Description

[0050] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0051] Figure 1 This is a flowchart of the method for targeted migration diagnosis of rolling bearing faults under multiple operating conditions provided in Embodiment 1 of the present invention;

[0052] Figure 2 This is the KNN graph construction process based on sample data features provided in Embodiment 1 of the present invention;

[0053] Figure 3 This is a schematic diagram of the HFZZ rotating machinery fault simulation platform provided in Embodiment 1 of the present invention;

[0054] Figure 4 These are experimental results for different cross-domain diagnostic tasks provided in Embodiment 1 of the present invention. Detailed Implementation

[0055] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0056] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0057] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0058] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0059] Example 1:

[0060] like Figure 1 As shown, Embodiment 1 of the present invention provides a method for targeted migration diagnosis of rolling bearing faults across operating conditions, comprising the following steps:

[0061] S1: Signal Acquisition and Dataset Division

[0062] Data under different operating conditions was collected using a mechanical failure testing platform, including labeled source domain datasets. and unlabeled target domain dataset Furthermore, the sample data was further divided into training data and test data;

[0063] S2: Network Model Construction and High-Dimensional Feature Extraction

[0064] A model is constructed using a feature encoder, a graph construction layer, and a multi-channel kernel graph convolutional network to extract deep, high-dimensional features from the sample data.

[0065] S3: Overall Objective Function and Model Optimization

[0066] By using a classification loss function, the classification error is reduced. Transfer learning methods are used to jointly reduce inter-domain differences using a loss function based on data structure differences and an adversarial inter-domain alignment loss function. The parameters in the overall objective function are optimized using a backpropagation algorithm.

[0067] S4: Model testing and diagnostic results output

[0068] During the testing phase, the target domain test data can be used to perform deep feature extraction using the optimized network in S2, and the classifier can be directly used for fault classification.

[0069] In S2, the construction process of the network model and high-dimensional feature extraction includes:

[0070] (1) Feature Encoder

[0071] It mainly consists of a generator G and a classifier C. The generator G is used to encode the input data to obtain a high-dimensional feature representation, and the classifier C will perform the final classification of the source and target tasks.

[0072] In this embodiment, a one-dimensional convolutional neural network (CNN) is constructed for feature extraction and fault classification. The hidden layer of the CNN corresponds to G and is used to perform non-linear feature mapping to obtain deep feature representation. The Softmax of the CNN output layer is used as a classifier C to obtain the probability output of each fault type.

[0073] The high-dimensional features in CNNs can be represented as:

[0074] G(x)=G L (G L-1 (…G 2 (G 1 (x,w 1 )))) (1)

[0075] Where x is the input, G is the feature map of the CNN, and w i is the learning weight of the i-th layer in the CNN, and L is the number of layers in the CNN.

[0076] (2) Graph construction layer

[0077] After the samples pass through the feature encoder, the KNN algorithm can be used to generate a graph construction layer, defining the adjacency relationships between all outputs and reflecting the local characteristics between samples. The construction process is as follows: Figure 2 As shown.

[0078] The construction of edges can be represented by the following formula:

[0079] A ij =KNN(k,L) ij ,Ω i ),A ij ∈A (2)

[0080] Among them, Ω i ={L i1 ,L i2 ,…,L in} represents node h i The set of distances to all other nodes, where k is a hyperparameter.

[0081] The graph construction layer is used to obtain the adjacency matrix A and to obtain the instance graph from the min-batch input matrix. It mainly consists of the following steps:

[0082] (A) Using the feature encoder network in (1), a high-dimensional feature map, i.e., X = G(x), is obtained from the sample data;

[0083] (B) The extracted high-dimensional feature maps, input into a linear layer and after passing through Softmax, can be represented as:

[0084] (C) Obtain the adjacency matrix A by performing matrix multiplication between the features of the linear layer and its transpose, i.e.

[0085] (D) Construct edge relationships using the KNN algorithm, i.e. Therefore, the construction of the adjacency matrix can be summarized as follows:

[0086]

[0087] Where A is the constructed adjacency matrix. It is the output of high-dimensional features after passing through a linear layer and Softmax, and normlize(·) represents the normalization function. It is a sparse adjacency matrix, and KNN(·) returns the indices of the k largest values ​​in the row direction of the adjacency matrix A.

[0088] (3) Multichannel kernel graph convolutional network

[0089] Graph convolutional networks (GCNNs) are networks that can be represented by the geometry and structure of data, providing more information and learning based on the connections between nodes. They can generally be simplified to G = (A, X), where A is the adjacency matrix reflecting the connections between nodes, and X represents the node features. L = I N -D -1 / 2 AD -1 / 2 It is a Laplacian matrix, where D can be obtained from the adjacency matrix, i.e. I N It is the identity matrix; graph convolution uses a filter g. θ =diag(θ) to smooth the input signal can be expressed as:

[0090] g θ * G x = Ug θ U T x (4)

[0091] Where θ is a learnable parameter, * G This is a graph convolution operation, where U is the eigenvector of the Laplacian matrix. T x represents the Fourier transform of the signal on the graph.

[0092] The graph convolution operation defined in Equation (4) is not localized and has a high computational cost. Equation (5) is used to restrict the convolution kernel to a polynomial expansion:

[0093]

[0094] Where K is the order of the polynomial. λ represents the eigenvalue of the Laplacian matrix.

[0095] A multi-kernel graph convolutional network is used to obtain feature representations with a wider receptive field. Its convolution operation can be defined as:

[0096]

[0097] in, These are learnable parameters. Representing the kth i High-dimensional feature representation of each convolutional kernel.

[0098] Suppose that the number of multi-channel convolutional kernels in the network is k. i The number of channels is L = (1,2,…,l), which is the network depth. A multi-kernel graph convolutional network with different receptive fields is constructed by performing graph convolution on different channels.

[0099]

[0100]

[0101] Where X represents the input and A represents the adjacency matrix. G represents the trainable weight matrix, and G represents the multi-channel kernel map convolution operation. Representing the kth i The high-dimensional feature representation of each channel in the Lth layer, [·] represents feature concatenation, and H represents the output feature after passing through the multi-channel kernel graph convolutional network.

[0102] In S3, the process of constructing the overall objective function and model optimization includes:

[0103] To fully utilize the characteristics of the data and the deep network structure to reduce the feature differences between the source and target domains after data mapping, a three-part loss function was adopted, namely classification loss, structural difference loss, and adversarial loss, to form the overall objective function.

[0104] (1) Classification loss

[0105] Cross-entropy loss is used to estimate the classification loss between the true label and the predicted label. Cross-entropy loss can be defined as:

[0106]

[0107] in, L represents the prediction result of the label classifier. C Let represent the cross-entropy loss function. E represents the expected value.

[0108] (2) Structural difference loss

[0109] The following metric is used to reduce the structural difference loss between the source and target domains, defined as:

[0110]

[0111] in, and Let represent the feature maps of the i-th source domain sample and the j-th target domain sample after passing through the feature extractor (in this invention, the feature encoder and the multi-channel kernel graph convolutional network), respectively. φ represents the nonlinear feature map, and Ω represents the nonlinear feature map. k This represents the distance metric from the extracted features to the regenerated kernel Hilbert space RKHS. In this embodiment, a convex combination k of m kernels is used. u To efficiently estimate the mapping:

[0112]

[0113] Where, α u These are weighted parameters for different cores, and In this invention, it is evenly distributed in [0,1].

[0114] (3) Combating losses

[0115] Adversarial training is employed to address the issues of domain skewness and domain shift. A domain discriminator determines whether the extracted features originate from the source or target domain, and a feature extractor is trained to deceive the discriminator. When the two reach a minmax game equilibrium, domain invariance is achieved. The loss function in formula (12) is used as the adversarial loss.

[0116]

[0117] Where D(·) is the feature output after passing through the discriminator, with a value between 0 and 1, which can distinguish whether it comes from the source domain or the target domain.

[0118] (4) Overall objective function

[0119] The overall objective function can be defined as:

[0120]

[0121] Where τ and It is an adjustable parameter.

[0122] (5) Parameter optimization

[0123] For the overall objective function in formula (13), the parameters of each part are optimized using the backpropagation algorithm as follows:

[0124]

[0125]

[0126]

[0127] in, Represents the partial differential operator, η represents the learning rate, and θ represents the partial differential operator.F θ represents the parameters of the feature extractor. C θ represents the parameters of the classifier. D The parameters represent the discriminator.

[0128] The detailed network structure of the rolling bearing fault-targeted migration diagnosis method under cross-operating conditions proposed in this embodiment is shown in Table 1, where C represents the fault type:

[0129] Table 1: Detailed Network Structure of the Targeted Migration Diagnosis Method for Rolling Bearing Faults under Multiple Operating Conditions

[0130]

[0131] This embodiment provides the following specific example:

[0132] Build the HFZZ rotating machinery fault simulation platform, such as Figure 3 As shown, the platform consists of a motor, control system, radial loading device, and accelerometer. Raw data was acquired via the accelerometer at a frequency of 12.8 kHz. Three different speeds were used to simulate bearing operation under different conditions: Condition A (1750 rpm), Condition B (2000 rpm), and Condition C (2250 rpm). Bearing faults were processed to generate nine health conditions, including normal (NM), inner ring fault (IR, IRU), outer ring fault (OR), rolling element fault (BA), and various mixed faults. These included single and compound faults, detailed in Table 2. In the experiment, 100 data sets were collected from each health condition at each operating speed, with each set containing 1024 data points. A total of 900 samples were obtained for each condition (100 * 9 health conditions), resulting in a dataset of 2700 samples (900 * 3 different speeds), as detailed in Table 3. The training and test sets were divided in a 6:4 ratio.

[0133] Table 2: Detailed Description of Rolling Bearing Health Status

[0134]

[0135] Table 3: Detailed Description of Rolling Bearing Migration Task

[0136]

[0137] To verify the superiority of the proposed targeted migration diagnosis method for rolling bearing faults under multiple operating conditions, several state-of-the-art deep neural network algorithms, such as CNN (Baseline), MKMMD, JAN, DANN, and CDAN, were used to compare the cross-domain diagnosis task under multiple operating conditions. Ten experiments were conducted in the fault diagnosis experiment, and the average value was taken as the experimental results, as shown in Table 4. Figure 4 As shown in the results, the proposed method in this embodiment achieves superior performance, with an improvement in overall average accuracy. Even when the difference between the source and target domains becomes greater, and cross-domain tasks become more difficult, the proposed algorithm can still achieve performance improvements.

[0138] Table 4: Experimental Results of Rolling Bearing Dataset

[0139]

[0140]

[0141] Example 2:

[0142] Embodiment 2 of the present invention provides a rolling bearing fault targeted migration diagnosis system across operating conditions, comprising:

[0143] The data acquisition module is configured to acquire the vibration signal of the rolling bearing;

[0144] The feature extraction module is configured to extract deep, high-dimensional features from vibration signals through a feature encoder, a graph construction layer, and a multi-channel kernel graph convolutional network.

[0145] The fault diagnosis module is configured to obtain rolling bearing fault diagnosis results based on the extracted deep high-dimensional features and classifier.

[0146] Among them, classification loss is obtained based on the deep high-dimensional features obtained by combining the source domain training set with the classifier, structural difference loss is obtained based on the deep high-dimensional features obtained by the source domain training set and the target domain training set, and adversarial loss is obtained based on the deep high-dimensional features obtained by the source domain training set and the target domain training set and the adversarial network.

[0147] The overall loss function is the sum of the classification loss, the product of the first parameter and the structural difference loss, and the product of the second parameter and the adversarial loss. Based on the overall objective function, the parameters of the feature extractor, classifier, and discriminator are optimized through the backpropagation algorithm.

[0148] The working method of the system is the same as the cross-condition rolling bearing fault targeted migration diagnosis method provided in Example 1, and will not be repeated here.

[0149] Example 3:

[0150] Embodiment 3 of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the cross-condition rolling bearing fault targeted migration diagnosis method described in Embodiment 1 of the present invention.

[0151] Example 4:

[0152] Embodiment 4 of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the cross-working-condition rolling bearing fault targeted migration diagnosis method described in Embodiment 1 of the present invention.

[0153] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0154] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0155] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0156] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0157] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0158] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for targeted migration diagnosis of rolling bearing faults across operating conditions, characterized in that, The process includes the following: Obtain the vibration signal of the rolling bearing; Deep, high-dimensional features in vibration signals are extracted using a feature encoder, graph construction layer, and multi-channel kernel graph convolutional network. Based on the extracted deep high-dimensional features and classifier, the rolling bearing fault diagnosis results are obtained; Among them, classification loss is obtained based on the deep high-dimensional features obtained by combining the source domain training set with the classifier; structural difference loss is obtained based on the deep high-dimensional features obtained by the source domain training set and the target domain training set; and adversarial loss is obtained based on the deep high-dimensional features obtained by the source domain training set and the target domain training set, as well as the adversarial network. The overall loss function is the sum of the classification loss, the product of the first parameter and the structural difference loss, and the product of the second parameter and the adversarial loss. Based on the overall objective function, the parameters of the feature extractor, classifier, and discriminator are optimized through the backpropagation algorithm.

2. The method for targeted migration diagnosis of rolling bearing faults across operating conditions as described in claim 1, characterized in that, The graph construction layer is used to obtain the adjacency matrix, including: Based on the feature encoder network, a high-dimensional feature map, X = G(x), is obtained from the sample data; The extracted high-dimensional feature maps are input into a linear layer and represented after passing through a softmax layer as follows: The adjacency matrix A is obtained by performing matrix multiplication between the features of the linear layer and its transpose. The edge relationships are constructed using the KNN algorithm, i.e.

3. The method for targeted migration diagnosis of rolling bearing faults across operating conditions as described in claim 2, characterized in that, Multi-channel kernel graph convolutional networks, including: Where X represents the input and A represents the adjacency matrix. G represents the trainable weight matrix, and G represents the multi-channel kernel map convolution operation. Representing the kth i The high-dimensional feature representation of each channel in the Lth layer, [·] represents feature concatenation, and H represents the output feature after passing through the multi-channel kernel graph convolutional network.

4. The method for targeted migration diagnosis of rolling bearing faults across operating conditions as described in claim 1, characterized in that, Cross-entropy loss L C ,include: in, E represents the prediction result of the classifier, and E represents the expected value. For source domain samples, Tag it.

5. The method for targeted migration diagnosis of rolling bearing faults across operating conditions as described in claim 1, characterized in that, Structural difference loss L s ,include: in, and Let φ and Ω represent the feature maps of the i-th source domain sample and the j-th target domain sample after passing through the feature extractor, respectively. Let φ represent the nonlinear feature map. k The distance metric from the extracted features to the regenerated kernel Hilbert space RKHS is represented by a convex combination of m kernels k. u To efficiently estimate the mapping: Where, α u These are weighted parameters for different cores, and E represents the expected value in mathematics. For source domain samples, For the target domain sample.

6. The method for targeted migration diagnosis of rolling bearing faults across operating conditions as described in claim 1, characterized in that, Combat loss L AD ,include: Where D(·) is the feature output after passing through the discriminator. and Let E represent the feature maps of the i-th source domain sample and the j-th target domain sample after passing through the feature extractor, respectively, and let E represent the expected value. For source domain samples, For the target domain sample.

7. The method for targeted migration diagnosis of rolling bearing faults across operating conditions as described in claim 1, characterized in that, The parameters of the feature extractor, classifier, and discriminator are optimized using the backpropagation algorithm, including: in, Represents the partial differential operator, η represents the learning rate, and θ represents the partial differential operator. F θ represents the parameters of the feature extractor. C θ represents the parameters of the classifier. D L represents the parameters of the discriminator. C For cross-entropy loss, L s For structural difference loss, L AD To combat the losses.

8. A cross-condition rolling bearing fault targeted migration diagnostic system, characterized in that, include: The data acquisition module is configured to acquire the vibration signal of the rolling bearing; The feature extraction module is configured to extract deep, high-dimensional features from vibration signals through a feature encoder, a graph construction layer, and a multi-channel kernel graph convolutional network. The fault diagnosis module is configured to obtain rolling bearing fault diagnosis results based on the extracted deep high-dimensional features and classifier. Among them, classification loss is obtained based on the deep high-dimensional features obtained by combining the source domain training set with the classifier; structural difference loss is obtained based on the deep high-dimensional features obtained by the source domain training set and the target domain training set; and adversarial loss is obtained based on the deep high-dimensional features obtained by the source domain training set and the target domain training set, as well as the adversarial network. The overall loss function is the sum of the classification loss, the product of the first parameter and the structural difference loss, and the product of the second parameter and the adversarial loss. Based on the overall objective function, the parameters of the feature extractor, classifier, and discriminator are optimized through the backpropagation algorithm.

9. A computer-readable storage medium having a program stored thereon, characterized in that, When executed by the processor, the program implements the steps in the cross-condition rolling bearing fault targeted migration diagnosis method as described in any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the cross-condition rolling bearing fault targeted migration diagnosis method as described in any one of claims 1-7.

Images