CN116026593A - Cross-working-condition rolling bearing fault targeted migration diagnosis method and system - Google Patents

Abstract

The present invention provides a rolling bearing fault targeted migration diagnosis method and system across working conditions, which solves the problem that the traditional rolling bearing fault diagnosis algorithm is difficult to extract deep network feature information in the source domain and the target domain, and cannot realize effective cross-domain fault diagnosis. . The present invention uses a feature encoder to accurately extract the high-dimensional mapping feature of the signal from the input rolling bearing vibration signal; this feature is further input into the graph construction layer, the deep features of the data are mined, and the multi-channel kernel image convolution network is used to analyze Graph modeling; using difference-based and adversarial training to minimize the distance between the source and target domain distributions, the classifier uses the extracted domain-invariant features to complete cross-domain fault identification. Compared with other methods, the present invention can better extract deep features for cross-domain transmission under the condition of cross-working conditions of rolling bearings, and greatly improve the diagnosis accuracy.

Description

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

Technical Field

The present invention relates to the technical field of bearing fault diagnosis, and in particular to a rolling bearing fault targeted migration diagnosis method and system across working conditions.

Background Art

The statements in this section merely provide background art related to the present invention and do not necessarily constitute prior art.

Rolling bearings are the key parts of most rotating mechanical equipment. Due to the characteristics of harsh working environment, changing working conditions, long overload operation time, etc., they often fail, which leads to mechanical failure of the whole machine and causes certain economic losses. Therefore, it is of great significance to monitor the operating status of rolling bearings under different working conditions.

In recent years, with the emergence of massive data, data-driven methods based on deep learning have attracted much attention. It is different from the traditional shallow network architecture. It is realized by stacking multiple layers of nonlinear processing units and provides an end-to-end solution. Many scholars have also conducted corresponding research. Jia et al. proposed a deep neural network DNN based on SAE to identify faults in motors and gearboxes. Ding et al. proposed the use of 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 using raw vibration signals as input, and used two CNNs with different kernel sizes to automatically extract different frequency signal features from raw data, and then used LSTM to identify fault types based on the learned features. The above deep learning algorithms have achieved good results in the field of rolling bearing fault diagnosis under constant working conditions. However, it is difficult to extract inter-domain difference characteristics under complex cross-working conditions, which reduces the generalization performance of the model.

As a method to reduce the difference in cross-domain feature distribution, transfer learning provides a new idea for establishing knowledge transfer from source domain labeled data to target domain unlabeled data, and effectively solves the problem of inter-domain distribution differences under cross-working conditions. In the field of intelligent fault diagnosis, many transfer learning methods have been proposed to solve the cross-domain diagnosis problem, which can be basically divided into instance-based, model-based and feature-based. 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 of bearing fault diagnosis. By minimizing the MMD loss, the marginal distribution and conditional distribution are aligned simultaneously in multiple layers. Han et al. proposed a deep adversarial convolutional neural network (DACNN) that uses adversarial loss functions to reduce inter-domain differences and improve the generalization performance of gearbox and motor fault diagnosis. Li et al. used graph data with topological structure as input and adopted graph convolutional network (GCN) modeling, which is more effective in data relationship mining and more powerful in feature representation, for mechanical fault diagnosis, and achieved excellent performance. It can be seen that the information of category labels, domain labels, and data structures plays an important role in reducing the differences between the source domain and the target domain, and they should complement and enhance each other.

However, the inventors found that the existing methods only considered two types of information, the source domain and the target domain, and did not integrate the data structure into the deep neural network model. The data distribution under different working conditions was quite different, resulting in low fault identification accuracy and insufficient generalization performance.

Summary of the invention

In order to address the deficiencies in the prior art, the present invention provides a rolling bearing fault targeted migration diagnosis method and system across operating conditions. Under the rolling bearing cross-operating conditions, deep features can be better extracted for cross-domain transfer, greatly improving the diagnostic accuracy.

In order to achieve the above object, the present invention adopts the following technical solution:

A first aspect of the present invention provides a rolling bearing fault targeted migration diagnosis method across operating conditions.

A rolling bearing fault targeted migration diagnosis method across working conditions includes the following processes:

Obtain vibration signals of rolling bearings;

Extract deep high-dimensional features from vibration signals through feature encoders, graph construction layers, and multi-channel kernel graph convolutional networks;

According to the extracted deep high-dimensional features and classifiers, the rolling bearing fault diagnosis results are obtained;

Among them, the classification loss is obtained according to the deep high-dimensional features obtained by combining the source domain training set with the classifier, the structural difference loss is obtained according to the deep high-dimensional features obtained by the source domain training set and the target domain training set, and the adversarial loss is obtained according to the deep high-dimensional features obtained by the source domain training set and the target domain training set and 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. According to the overall objective function, the parameters of the feature extractor, classifier, and discriminator are optimized through the back propagation algorithm.

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

According to the feature encoder network, a high-dimensional feature map is obtained from the sample data, that is, X = G(x);

The extracted high-dimensional feature map is input into the linear layer and expressed as

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

Through the KNN algorithm, construct edge relationships:

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

代表可训练权值矩阵，G代表多通道核图卷积操作，

代表第k_i个通道在第L层的高维特征表示，[·]代表特征拼接，H表示经过多通道核图卷积网络后的输出特征。Among them, X represents the input, A represents the adjacency matrix,

represents the trainable weight matrix, G represents the multi-channel kernel graph convolution operation,

represents the high-dimensional feature representation of the k _i -th channel at the L layer, [ ] represents feature concatenation, and H represents the output feature after passing through the multi-channel kernel graph convolutional network.

As an optional implementation of the first aspect of the present invention, the cross entropy loss _LC includes:

表示分类器的预测结果，E表示数学期望值，

为源域样本，

为其标签。in,

represents the prediction result of the classifier, E represents the mathematical expectation,

is the source domain sample,

Label it.

As an optional implementation of the first aspect of the present invention, the structural difference loss L _s includes:

和

分别表示第i个源域样本和第j个目标域样本经过特征提取器后的特征映射，φ表示非线性特征映射，Ω_k表示嵌入提取的特征到再生核希尔伯特空间RKHS中的距离度量，采用m个核的凸组合k_u来对映射进行有效地估计：

其中，α_u是不同核的加权参数，且

E表示数学期望值，

为源域样本，

为目标域样本。in,

and

They represent the feature maps of the i-th source domain sample and the j-th target domain sample after the feature extractor, φ represents the nonlinear feature map, Ω _k represents the distance metric embedded in the extracted features to the reproducing kernel Hilbert space RKHS, and the convex combination of m kernels k _u is used to effectively estimate the mapping:

where _αu is the weighting parameter of different kernels, and

E represents the mathematical expectation,

is the source domain sample,

is the target domain sample.

As an optional implementation of the first aspect of the present invention, counteracting the loss L _AD includes:

和

分别表示第i个源域样本和第j个目标域样本经过特征提取器后的特征映射，E表示数学期望值，

为源域样本，

为目标域样本。Among them, D(·) is the feature output after the discriminator,

and

They represent the feature maps of the i-th source domain sample and the j-th target domain sample after passing through the feature extractor, E represents the mathematical expectation value,

is the source domain sample,

is the target domain sample.

As an optional implementation of the first aspect of the present invention, the parameters of the feature extractor, the classifier and the discriminator are optimized by a back propagation algorithm, including:

代表偏微分算子，η代表学习率，θ_F代表特征提取器的参数，θ_C代表分类器的参数，θ_D代表判别器的参数，L_C为交叉熵损失，L_s为结构差异损失，L_AD为对抗损失。in,

represents the partial differential operator, η represents the learning rate, θ _F represents the parameters of the feature extractor, θ _C represents the parameters of the classifier, θ _D represents the parameters of the discriminator, _LC is the cross entropy loss, _Ls is the structural difference loss, and _LAD is the adversarial loss.

A second aspect of the present invention provides a rolling bearing fault targeted migration diagnosis system across operating conditions.

A rolling bearing fault targeted migration diagnosis system across working conditions, comprising:

The data acquisition module is configured to: acquire a vibration signal of the rolling bearing;

A feature extraction module, configured to: extract deep high-dimensional features from the vibration signal through a feature encoder, a graph construction layer, and a multi-channel kernel graph convolutional network;

The fault diagnosis module is configured to: obtain a rolling bearing fault diagnosis result based on the extracted deep high-dimensional features and the classifier;

Among them, the classification loss is obtained according to the deep high-dimensional features obtained by combining the source domain training set with the classifier, the structural difference loss is obtained according to the deep high-dimensional features obtained by the source domain training set and the target domain training set, and the adversarial loss is obtained according to the deep high-dimensional features obtained by the source domain training set and the target domain training set and 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. According to the overall objective function, the parameters of the feature extractor, classifier, and discriminator are optimized through the back propagation algorithm.

A third aspect of the present invention is a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the method for targeted migration diagnosis of rolling bearing faults across operating conditions as described in the first aspect of the present invention.

The fourth aspect of the present invention provides an electronic device, comprising a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the method for targeted migration diagnosis of rolling bearing faults across working conditions as described in the first aspect of the present invention are implemented.

Compared with the prior art, the present invention has the following beneficial effects:

1. The present invention adopts 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 and applying a multi-channel kernel graph convolutional network to model it, further mining the high-dimensional features of the signal, and solving the problem of difficulty in extracting deep features in the source domain and the target domain under cross-working conditions.

2. To address the problem of large data differences between the source domain and target domain of rolling bearing vibration signals under different working conditions, the present invention adopts a loss function based on data structure differences and a loss function based on adversarial domain-to-domain comparison to jointly reduce the inter-domain differences. At the same time, the classifier uses the extracted domain-invariant features to complete cross-domain fault identification.

3. The present invention solves the problems that labeled data is difficult to obtain in industrial scenarios, data distribution varies greatly under different working conditions, resulting in low fault identification accuracy and insufficient generalization performance. The present invention can not only extract deep high-dimensional features to narrow the differences between domains, but also obtain better diagnostic performance.

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

FIG1 is a flow chart of a rolling bearing fault targeted migration diagnosis method under cross-operating conditions provided by Embodiment 1 of the present invention;

FIG2 is a KNN graph construction process based on sample data features provided in Example 1 of the present invention;

FIG3 is a schematic diagram of the HFZZ rotating machinery fault simulation platform provided in Example 1 of the present invention;

FIG. 4 shows the experimental results of different cross-domain diagnosis tasks provided by Example 1 of the present invention.

DETAILED DESCRIPTION

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and/or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and/or combinations thereof.

In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

Embodiment 1:

As shown in FIG1 , Embodiment 1 of the present invention provides a rolling bearing fault targeted migration diagnosis method across working conditions, including the following process:

S1: Signal acquisition and data set division

和不带标签的目标域数据集

并且样本数据被进一步划分为训练数据和测试数据；Data under different working conditions are collected through the mechanical failure test platform, including labeled source domain datasets

and unlabeled target domain dataset

And the sample data is further divided into training data and test data;

S2: Network model construction and high-dimensional feature extraction

The model is constructed through feature encoder, graph construction layer, and multi-channel kernel graph convolutional network to extract deep high-dimensional features of sample data;

S3: Overall objective function and model optimization

Through the classification loss function, the classification error is reduced, and the loss function based on data structure difference and the loss function based on adversarial domain alignment are used to jointly reduce the domain difference, and the back propagation algorithm is used to optimize the parameters in the overall objective function;

S4: Model testing and diagnostic results output

During the testing phase, the target domain test data can use the optimized network in S2 for deep feature extraction, and the classifier can be directly used for fault classification.

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

(1) Feature Encoder

It mainly includes generator G and classifier C. Generator G is used to encode the input data to obtain high-dimensional feature representation, and classifier C will perform the final classification of source and target tasks;

In this embodiment, a one-dimensional convolutional neural network CNN is constructed for feature extraction and fault classification. The hidden layer of CNN corresponds to G and is used to perform nonlinear 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.

The high-dimensional feature representation in CNN can be expressed as:

G(x)＝G ^L (G ^L-1 (…G ² (G ¹ (x, w ¹ )))) (1)

Where x is the input, G is the feature map of CNN, ^wi is the learned weight of the i-th layer in CNN, and L is the number of layers of CNN.

(2) Graph Construction Layer

After the samples pass through the feature encoder, the KNN algorithm can be used to generate a graph construction layer to define the adjacency relationship between all outputs and reflect the local characteristics between samples. The construction process is shown in Figure 2.

The construction of the edge can be expressed as follows:

A _ij =KNN(k,L _ij ,Ω _i ),A _ij ∈A (2)

Wherein, Ω _i ={L _i1 ,L _i2 ,…,L _in } represents the distance set between node h _i and all other nodes, and k is a hyperparameter.

The graph construction layer is used to obtain the adjacency matrix A and obtain the instance graph from the minimum batch input matrix. It is mainly divided into the following steps:

(A) Using the feature encoder network in (1), we obtain a high-dimensional feature map from the sample data, i.e., X = G(x);

(B) The extracted high-dimensional feature map is input into the linear layer and can be expressed as follows after Softmax:

(C) By performing matrix multiplication between the features of the linear layer and its transpose, the adjacency matrix A is obtained, that is,

因此，邻接矩阵的构建总结如下：(D) Through the KNN algorithm, the edge relationship is constructed, that is,

Therefore, the construction of the adjacency matrix can be summarized as follows:

是高维特征经过线性层和Softmax后的输出，normlize(·)表示归一化函数，

是稀疏邻接矩阵，KNN(·)返回邻接矩阵A在行方向上的前k个最大值的索引。Among them, A is the constructed adjacency matrix,

is the output of the high-dimensional feature after the linear layer and Softmax, normlize(·) represents the normalization function,

is a sparse adjacency matrix, KNN(·) returns the indices of the first k largest values in the row direction of the adjacency matrix A.

(3) Multi-channel kernel graph convolutional network

I_N是单位矩阵，图卷积使用滤波器g_θ＝diag(θ)来平滑输入信号，可以表示为：Graph convolutional network is a network that can be represented by the geometric shape and structure of data, which can provide more information representation and learn the network based on the connection relationship between nodes. It can be generally simplified to G = (A, X), where A is the adjacency matrix, which can reflect the connection relationship between nodes, and X is the feature of the node. L = I _N -D ^-1/2 AD ^-1/2 is the Laplacian matrix, where D can be obtained from the adjacency matrix, that is,

I _N is the identity matrix. Graph convolution uses a filter g _θ = diag(θ) to smooth the input signal, which can be expressed as:

g _θ * _G x＝Ug _θ U ^T x (4)

Among them, θ is a learnable parameter, * _G is the graph convolution operation, U is the eigenvector of the Laplacian matrix, and U ^T x represents the Fourier transform of the signal on the graph.

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

λ表示Laplacian矩阵的特征值。Where K is the order of the polynomial,

λ represents the eigenvalue of the Laplacian matrix.

A multi-core graph convolutional network is used to obtain feature representation with a wider receptive field. Its convolution operation can be defined as:

是可学习的参数，

代表第k_i个卷积核的高维特征表示。in,

is a learnable parameter,

Represents the high-dimensional feature representation of the k _i -th convolution kernel.

Assume that the number of multi-channel convolution kernels in the network is k _i , that is, the number of channels, and the number of network layers is L = (1, 2, …, l), that is, the network depth. Use a multi-core graph convolution network with different receptive fields to perform graph convolution on different channels and construct a multi-channel kernel graph convolution network:

代表可训练权值矩阵，G代表多通道核图卷积操作，

代表第k_i个通道在第L层的高维特征表示，[·]代表特征拼接，H表示经过多通道核图卷积网络后的输出特征。Among them, X represents the input, A represents the adjacency matrix,

represents the trainable weight matrix, G represents the multi-channel kernel graph convolution operation,

represents the high-dimensional feature representation of the k _i -th channel at the L layer, [ ] represents feature concatenation, and H represents the output feature after passing through the multi-channel kernel graph convolutional network.

In S3, the overall objective function and model optimization construction process includes:

In order to make full use of the characteristics of the data and the deep network structure to narrow the feature differences between the source domain and the target domain after data mapping, a three-part loss function is adopted, namely classification loss, structural difference loss, and adversarial loss to constitute the overall objective function.

(1) Classification loss

The classification loss between the true label and the predicted label is estimated by the cross entropy loss, which can be defined as:

表示标签分类器的预测结果，L_C表示交叉熵损失函数。E表示数学期望值。in,

represents the prediction result of the label classifier, L _C represents the cross entropy loss function, and E represents the mathematical expectation.

(2) Structural difference loss

The following metric is used to reduce the structural difference loss between the source domain and the target domain, which is defined as:

和

分别表示第i个源域样本和第j个目标域样本经过特征提取器(本发明中指特征编码器和多通道核图卷积网络)后的特征映射。φ表示非线性特征映射，Ω_k表示嵌入提取的特征到再生核希尔伯特空间RKHS中的距离度量，本实施例中，采用m个核的凸组合k_u来对映射进行有效地估计：in,

and

Respectively represent the feature mapping of the i-th source domain sample and the j-th target domain sample after passing through the feature extractor (the feature encoder and the multi-channel kernel graph convolution network in the present invention). φ represents a nonlinear feature mapping, Ω _k represents the distance metric embedded in the extracted features into the reproducing kernel Hilbert space RKHS. In this embodiment, a convex combination of m kernels k _u is used to effectively estimate the mapping:

本发明中对其在[0,1]中进行了平均分配。where _αu is the weighting parameter of different kernels, and

In the present invention, it is evenly distributed in [0,1].

(3) Fighting Losses

The adversarial training method is used to deal with the problems of domain skew and domain shift. The domain discriminator is used to determine whether the extracted features are from the source domain or the target domain, and the feature extractor is trained to deceive the discriminator. When the two reach the minmax game equilibrium, the domain invariance feature can be obtained, and the loss function in formula (12) is used as the adversarial loss:

Among them, D(·) is the feature output after the discriminator, and its value is between 0 and 1, which can distinguish whether it comes from the source domain or the target domain.

(4) Overall objective function

The overall objective function can be defined as:

是可调参数。Among them, τ and

It is an adjustable parameter.

(5) Parameter Optimization

For the overall objective function in formula (13), the parameters of each part are optimized by back propagation algorithm as follows:

代表偏微分算子，η代表学习率，θ_F代表特征提取器的参数，θ_C代表分类器的参数，θ_D代表判别器的参数。in,

represents the partial differential operator, η represents the learning rate, θ _F represents the parameters of the feature extractor, θ _C represents the parameters of the classifier, and θ _D represents the parameters of the discriminator.

The detailed network structure of the rolling bearing fault targeted migration diagnosis method under different working conditions proposed in this embodiment is shown in Table 1, where C represents the fault type:

Table 1: Detailed network structure of rolling bearing fault targeted migration diagnosis method under cross-operating conditions

This embodiment provides the following specific cases:

The HFZZ rotating machinery fault simulation platform was built, as shown in Figure 3. The platform consists of a motor, a control system, a radial loading device, an accelerometer, etc. The original data was collected by an accelerometer with a collection frequency of 12.8kHz. Three different speeds were used to simulate the operating conditions of the bearings under different working conditions, namely, working condition A (1750rpm), working condition B (2000rpm) and working condition C (2250rmp). The bearing fault was processed to generate 9 health conditions including normal (NM), inner ring fault (IR, IRU), outer ring fault (OR), rolling element fault (BA) and multiple mixed faults, including single faults and compound faults. The detailed information is shown in Table 2. In the experiment, 100 sets of data were collected from each health condition at each operating speed, and each data contained 1024 data points. Under each working condition, a total of 900 samples (100*9 health conditions) were obtained, so the data set contains a total of 2700 samples (900*3 different speeds). The specific description is shown in Table 3, and the training set and test set are divided according to the ratio of 6:4.

Table 2: Detailed description of rolling bearing health status

Table 3: Detailed description of rolling bearing migration tasks

In order to verify the superiority of the proposed method for targeted migration diagnosis of rolling bearing faults under cross-working conditions, several state-of-the-art deep neural network algorithms, such as CNN (Baseline), MKMMD, JAN, DANN, and CDAN, were used to compare cross-domain diagnosis tasks under cross-working conditions. Ten tests were conducted in the fault diagnosis experiment, and the average values were taken as the experimental results as shown in Table 4 and Figure 4. It can be seen from the results that the method proposed in this embodiment has achieved superior performance and has improved in terms of overall average accuracy. When the difference between the source domain and the target domain becomes larger and the cross-domain task becomes more difficult, the proposed algorithm can still achieve performance improvement.

Table 4: Experimental results of rolling bearing dataset

Embodiment 2:

Embodiment 2 of the present invention provides a rolling bearing fault targeted migration diagnosis system across working conditions, including:

The data acquisition module is configured to: acquire a vibration signal of the rolling bearing;

A feature extraction module, configured to: extract deep high-dimensional features from the vibration signal through a feature encoder, a graph construction layer, and a multi-channel kernel graph convolutional network;

The fault diagnosis module is configured to: obtain a rolling bearing fault diagnosis result based on the extracted deep high-dimensional features and the classifier;

Among them, the classification loss is obtained according to the deep high-dimensional features obtained by combining the source domain training set with the classifier, the structural difference loss is obtained according to the deep high-dimensional features obtained by the source domain training set and the target domain training set, and the adversarial loss is obtained according to the deep high-dimensional features obtained by the source domain training set and the target domain training set and 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. According to the overall objective function, the parameters of the feature extractor, classifier, and discriminator are optimized through the back propagation algorithm.

The working method of the system is the same as the rolling bearing fault targeted migration diagnosis method across working conditions provided in Example 1, and will not be repeated here.

Embodiment 3:

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 method for targeted migration diagnosis of rolling bearing faults across operating conditions as described in Embodiment 1 of the present invention.

Embodiment 4:

Embodiment 4 of the present invention provides an electronic device, comprising a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in the method for targeted migration diagnosis of rolling bearing faults across working conditions as described in Embodiment 1 of the present invention are implemented.

Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program codes.

The present invention is described with reference to the flowchart and/or block diagram of the method, device (system), and computer program product according to the embodiment of the present invention. It should be understood that each process and/or box in the flowchart and/or block diagram, as well as the combination of the process and/or box in the flowchart and/or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for implementing the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims (10)

1. A cross-working-condition rolling bearing fault targeted migration diagnosis method is characterized by comprising the following steps:

acquiring a vibration signal of a rolling bearing;

extracting deep high-dimensional features in the vibration signals through a feature encoder, a graph construction layer and a multi-channel kernel graph convolution network;

obtaining a rolling bearing fault diagnosis result according to the extracted deep high-dimensional characteristics and the classifier;

the method comprises the steps of obtaining classification loss according to deep high-dimensional characteristics obtained by combining a classifier with a source domain training set, obtaining structural difference loss according to the deep high-dimensional characteristics obtained by combining the source domain training set and a target domain training set, and obtaining countermeasures loss according to the deep high-dimensional characteristics obtained by combining the source domain training set and the target domain training set and a countermeasures network;

and optimizing parameters of the feature extractor, the classifier and the discriminator through a back propagation algorithm according to the overall objective function by taking 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 counterattack loss as an overall loss function.

2. The method for diagnosing the fault-targeted migration of the rolling bearing under the cross-working condition of claim 1, wherein the method comprises the following steps of,

the graph construction layer is used for acquiring an adjacency matrix and comprises the following steps:

obtaining a high-dimensional feature map, i.e., x=g (X), from the sample data according to the feature encoder network;

the extracted high-dimensional characteristic map is input into a linear layer and expressed as after Softmax

By performing matrix multiplication calculation between the features of the linear layer and its transpose, an adjacency matrix a is obtained,

by KNN algorithm, edge relations are constructed, i.e

3. A cross-operating mode rolling bearing fault targeted migration diagnostic method as claimed in claim 2, wherein,

a multi-channel kernel graph rolling network comprising:

wherein X represents the input, A represents the adjacency matrix,

represents a trainable weight matrix, G represents a multi-channel kernel graph convolution operation, ++>

Represents the kth _i High-dimensional characterization of individual channels at layer L, [. Cndot.]Representing feature stitching, H represents the output features after passing through the multi-channel kernel graph convolution network.

4. The method for diagnosing the fault-targeted migration of the rolling bearing under the cross-working condition of claim 1, wherein the method comprises the following steps of,

cross entropy loss L _C Comprising:

wherein ,

representing the prediction result of the classifier, E representing the mathematical expectation,/->

For the source domain sample, ++>

For its tag.

5. The method for diagnosing the fault-targeted migration of the rolling bearing under the cross-working condition of claim 1, wherein the method comprises the following steps of,

structural differential loss L _s Comprising:

wherein ,

and

Respectively representing the feature mapping of the ith source domain sample and the jth target domain sample after the ith source domain sample passes through a feature extractor, phi represents the nonlinear feature mapping, omega _k Representing distance metrics of embedded extracted features into regenerated kernel Hilbert space RKHS, employing convex combinations k of m kernels _u To effectively estimate the mapping:

wherein ,α_u Is a weighting parameter of different kernels and +.>

E represents a mathematical expectation value, +.>

For the source domain sample, ++>

Is a target domain sample.

6. The method for diagnosing the fault-targeted migration of the rolling bearing under the cross-working condition of claim 1, wherein the method comprises the following steps of,

countering loss L _AD Comprising:

wherein D (·) is the feature output after passing through the discriminator,

and

Respectively representing the feature mapping of the ith source domain sample and the jth target domain sample after the feature extractor, E represents the mathematical expectation value,>

for the source domain sample, ++>

Is a target domain sample.

7. The method for diagnosing the fault-targeted migration of the rolling bearing under the cross-working condition of claim 1, wherein the method comprises the following steps of,

optimizing parameters of the feature extractor, classifier and arbiter by a back propagation algorithm, comprising:

wherein ,

represents partial differential operator, eta represents learning rate, theta _F Parameters representative of feature extractor, θ _C Representing the parameters of the classifier, θ _D Representing parameters of the arbiter, L _C For cross entropy loss, L _s For structural differential loss, L _AD To combat losses.

8. A cross-condition rolling bearing fault targeted migration diagnostic system, comprising:

a data acquisition module configured to: acquiring a vibration signal of a rolling bearing;

a feature extraction module configured to: extracting deep high-dimensional features in the vibration signals through a feature encoder, a graph construction layer and a multi-channel kernel graph convolution network;

a fault diagnosis module configured to: obtaining a rolling bearing fault diagnosis result according to the extracted deep high-dimensional characteristics and the classifier;

the method comprises the steps of obtaining classification loss according to deep high-dimensional characteristics obtained by combining a classifier with a source domain training set, obtaining structural difference loss according to the deep high-dimensional characteristics obtained by combining the source domain training set and a target domain training set, and obtaining countermeasures loss according to the deep high-dimensional characteristics obtained by combining the source domain training set and the target domain training set and a countermeasures network;

and optimizing parameters of the feature extractor, the classifier and the discriminator through a back propagation algorithm according to the overall objective function by taking 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 counterattack loss as an overall loss function.

9. A computer readable storage medium having stored thereon a program, which when executed by a processor, implements the steps of the cross-regime rolling bearing fault targeted migration diagnostic method according to any one of claims 1 to 7.

10. An electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, wherein the processor performs the steps in the cross-regime rolling bearing fault targeted migration diagnostic method of any one of claims 1-7 when the program is executed.

Images