Rolling bearing migration fault diagnosis method and system under domain asymmetry

Abstract

The present invention provides a rolling bearing migration fault diagnosis method and system under the condition of domain asymmetry, and obtains a fault diagnosis result based on a preprocessed acceleration vibration signal and a pretrained rolling bearing migration fault diagnosis model; wherein the rolling bearing migration fault diagnosis model includes: a domain sharing module, a domain discrimination module and an adaptive module; the domain discrimination module trains the domain discrimination module by maximizing the Wasserstein distance, and generates a domain asymmetry factor based on the domain discrimination module; the adaptive module combines the domain asymmetry factor and calculates the weighted overall distribution distance and weighted local distribution distance of the source domain and the target domain by the maximum mean difference, thereby obtaining the loss function of the adaptive module; the domain sharing module is trained by minimizing the cross entropy loss of the domain sharing module and the migration loss of the adaptive module; the present invention effectively improves the accuracy of the fault diagnosis model under the condition of domain asymmetry.

Description

Technical Field

[0001] The present invention relates to the technical field of bearing fault diagnosis, and in particular to a rolling bearing migration fault diagnosis method and system under domain asymmetry conditions. Background Art

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

[0003] Rotating machinery is widely used in modern industry. The health of rolling bearings, key components of rotating machinery, significantly impacts their performance and service life. Due to harsh operating environments and constantly changing loads, the health of rolling bearings deteriorates over time. Therefore, research on rotating machinery condition monitoring and fault diagnosis is crucial to detecting, locating, and diagnosing potential rolling bearing failures during operation. This research is crucial for ensuring the safe and stable operation of rotating machinery, reducing economic losses and operating costs, and preventing major accidents.

[0004] In recent years, with the rapid development of artificial intelligence (AI), intelligent fault diagnosis methods based on deep learning have reduced the reliance of traditional diagnostic methods on physical knowledge and mathematical models, and have been widely used in the field of rolling bearing fault diagnosis. However, since deep learning-based fault diagnosis methods require extensive data support, the available data from actual rotating machinery operation is insufficient to train reliable intelligent diagnostic models. This has led to the emergence of deep transfer learning. Deep transfer learning-based fault diagnosis methods can apply knowledge acquired from laboratory rolling bearing diagnosis tasks to rolling bearing diagnosis tasks in actual rotating machinery, effectively overcoming the problem of insufficient available data in real-world scenarios.

[0005] The inventors found that existing deep transfer learning methods usually have the following assumptions: 1) the signal data of the rotating machinery (target domain) under actual working conditions is balanced in all states; 2) the fault type of the rotating machinery (target domain) under actual working conditions is consistent with the laboratory simulated rotating machinery (source domain); however, in reality, the rotating machinery operates normally most of the time, and the fault frequency is low, resulting in insufficient fault signal data, that is, data imbalance; in addition, some fault states simulated in the laboratory simulated rotating machinery may never occur in the rotating machinery under actual working conditions, that is, domain asymmetry; therefore, when the existing deep transfer fault diagnosis method is applied to the rotating machinery under actual working conditions, it is difficult to obtain satisfactory diagnostic results. Summary of the Invention

[0006] In order to address the shortcomings of the existing technology, the present invention provides a rolling bearing migration fault diagnosis method and system under domain asymmetry conditions. By adopting a domain asymmetry factor to assign weights to the data features of the source domain, the domain asymmetry problem existing in actual scenarios is solved, and the accuracy of the fault diagnosis model under domain asymmetry conditions is effectively improved.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A first aspect of the present invention provides a rolling bearing migration fault diagnosis method under domain asymmetry.

[0009] A rolling bearing migration fault diagnosis method under domain asymmetry includes the following steps:

[0010] Collect the acceleration vibration signal of the rolling bearing and perform preprocessing;

[0011] Obtain the fault diagnosis result based on the pre-processed acceleration vibration signal and the pre-trained rolling bearing migration fault diagnosis model;

[0012] Among them, the rolling bearing migration fault diagnosis model includes: domain sharing module, domain discrimination module and adaptive module;

[0013] The domain sharing module at least includes a convolutional layer, a maximum pooling layer, a first fully connected layer, a second fully connected layer, a third fully connected layer and a SoftMax activation function connected in sequence;

[0014] The domain discrimination module takes the output of the second fully connected layer of the domain sharing module as input, trains the domain discrimination module by maximizing the Wasserstein distance, and generates a domain asymmetry factor based on the domain discrimination module;

[0015] The adaptive module takes the output of the second fully connected layer of the domain sharing module as input, combines the domain asymmetry factor, and calculates the weighted overall distribution distance and weighted local distribution distance between the source and target domains through the maximum mean difference. The loss function of the adaptive module is derived based on the weighted overall distribution distance and weighted local distribution distance.

[0016] The domain sharing module is trained by minimizing the cross entropy loss of the domain sharing module and the migration loss of the adaptive module, and the output of the trained domain sharing module is used as the fault diagnosis result.

[0017] As an optional implementation of the first aspect of the present invention, the domain sharing module includes a first convolutional layer, a first maximum pooling layer, a second convolutional layer, a second maximum pooling layer, a first fully connected layer, a second fully connected layer, a third fully connected layer and a Softmax activation function connected in sequence.

[0018] As an optional implementation of the first aspect of the present invention, the Wasserstein distance L adv ,include:

[0019]

[0020] in, represents the output value of the time domain discrimination module of the i-th data feature in the input source domain, represents the output value of the time domain discrimination module of the jth data feature in the input target domain; n s Indicates the number of data in the source domain, n t Indicates the amount of data in the target domain.

[0021] As a further limitation of the first aspect of the present invention, the domain asymmetry factor includes:

[0022]

[0023] Among them, W i s represents the domain asymmetry factor of the i-th source domain data; σ s Represents the Sigmoid function.

[0024] As an optional implementation of the first aspect of the present invention, the weighted overall distribution distance L g ,include:

[0025]

[0026] Where φ(·) is the mapping from the original data space to the reproducible kernel Hilbert space, and are the features of the source domain and the target domain respectively, is an operator for different data distribution distances in the reproducible kernel Hilbert space, W i s represents the domain asymmetry factor of the i-th source domain data.

[0027] As an optional implementation of the first aspect of the present invention, the weighted local distribution distance L l ,include:

[0028]

[0029] in, and are the category weights of source domain data and target domain data respectively, C is the number of categories of source domain data, φ(·) is the mapping from the original data space to the reproducible kernel Hilbert space, and are the features of the source domain and the target domain respectively, is an operator for different data distribution distances in the reproducible kernel Hilbert space, W i s represents the domain asymmetry factor of the i-th source domain data.

[0030] As an optional implementation of the first aspect of the present invention, the migration loss function L of the adaptive module DA ,include:

[0031] L DA =L g +μL l ;

[0032] in, p is the training progress of the model, and γ is the parameter used to control the change of μ.

[0033] As an optional implementation of the first aspect of the present invention, minimizing the cross entropy loss of the domain sharing module and the migration loss of the adaptive module includes:

[0034]

[0035] Among them, L C is the cross entropy loss of the domain sharing module, L DA is the migration loss of the adaptive module, and λ is the migration weight.

[0036] A second aspect of the present invention provides a rolling bearing migration fault diagnosis system under domain asymmetry.

[0037] A rolling bearing migration fault diagnosis system under domain asymmetry, comprising:

[0038] The data acquisition module is configured to: collect acceleration vibration signals of the rolling bearing and perform preprocessing;

[0039] The fault diagnosis module is configured to obtain a fault diagnosis result based on the preprocessed acceleration vibration signal and the pre-trained rolling bearing migration fault diagnosis model;

[0040] Among them, the rolling bearing migration fault diagnosis model in the fault diagnosis module includes: a domain sharing module, a domain discrimination module and an adaptive module;

[0041] The domain sharing module at least includes a convolutional layer, a maximum pooling layer, a first fully connected layer, a second fully connected layer, a third fully connected layer and a SoftMax activation function connected in sequence;

[0042] The domain discrimination module takes the output of the second fully connected layer of the domain sharing module as input, trains the domain discrimination module by maximizing the Wasserstein distance, and generates a domain asymmetry factor based on the domain discrimination module;

[0043] The adaptive module takes the output of the second fully connected layer of the domain sharing module as input, combines the domain asymmetry factor, and calculates the weighted overall distribution distance and weighted local distribution distance between the source and target domains through the maximum mean difference. The loss function of the adaptive module is derived based on the weighted overall distribution distance and weighted local distribution distance.

[0044] The domain sharing module is trained by minimizing the cross entropy loss of the domain sharing module and the migration loss of the adaptive module, and the output of the trained domain sharing module is used as the fault diagnosis result.

[0045] As an optional implementation of the second aspect of the present invention, minimizing the cross entropy loss of the domain sharing module and the migration loss of the adaptive module includes:

[0046]

[0047] Among them, L C is the cross entropy loss of the domain sharing module, L DA is the migration loss of the adaptive module, and λ is the migration weight.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. The rolling bearing migration fault diagnosis method and system under domain asymmetry conditions described in the present invention solves the domain asymmetry problem existing in actual scenarios by using a domain asymmetry factor to assign weights to the data features of the source domain, and effectively improves the accuracy of the fault diagnosis model under domain asymmetry conditions.

[0050] 2. The rolling bearing migration fault diagnosis method and system under domain asymmetry described in the present invention take into account the local distribution differences of data of the same category, align the source domain data and the target domain data more carefully, solve the data imbalance problem existing in actual scenarios, and obtain a fault diagnosis model with high accuracy and strong robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings, 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.

[0052] Figure 1 A schematic flow chart of a rolling bearing migration fault diagnosis method under domain asymmetry provided in Example 1 of the present invention;

[0053] Figure 2 A schematic diagram of the overall structure of the rolling bearing migration fault diagnosis model under domain asymmetry provided by Example 1 of the present invention;

[0054] Figure 3A detailed structural diagram of the domain sharing module provided in Example 1 of the present invention;

[0055] Figure 4 This is a detailed structural diagram of the domain discrimination module provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0056] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0057] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0058] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0059] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0060] Example 1:

[0061] like Figure 1 As shown, embodiment 1 of the present invention provides a rolling bearing migration fault diagnosis method under domain asymmetry, including the following process:

[0062] S1: Data collection and dataset creation

[0063] The acceleration sensor is used to collect the acceleration vibration signals of the rolling bearings under normal conditions, inner ring fault, outer ring fault, and rolling element fault at a sampling frequency of 12.8kHz. N ); where x represents an acceleration vibration signal of the bearing, x i represents the i-th sampling point of the vibration signal x, 1≤i≤N, N is the total number of sampling points of signal x;

[0064] The source domain dataset contains 4,000 signals, generated by the rotating machinery fault simulation platform HFZZ-II. Signals of different fault types are constructed in a ratio of 1:1:1:1:1:normal:inner race fault:outer race fault:rolling element fault.

[0065] The target domain dataset consists of 1000 signals, which are obtained by collecting data on high-speed train axle box bearings in working state using acceleration sensors. The data is constructed in a ratio of 3:1:1 for normal, inner race fault, and outer race fault.

[0066] S2: Data Preprocessing

[0067] The data is standardized and the standardization formula is as follows:

[0068]

[0069] Wherein, μ represents the mean value of a vibration signal x, and σ represents the standard deviation of a vibration signal x.

[0070] S3: Construction of rolling bearing migration fault diagnosis model

[0071] The rolling bearing migration fault diagnosis model consists of three parts: domain sharing module, domain discrimination module and adaptive module. The overall structure of the rolling bearing migration fault diagnosis model under domain asymmetry is as follows: Figure 2 shown.

[0072] S3.1: Domain Sharing Module

[0073] The domain sharing module consists of a convolutional layer, a maximum pooling layer, a first fully connected layer, a second fully connected layer, a third fully connected layer, and a SoftMax activation function. Its detailed structure is as follows: Figure 3 As shown in the figure, the third fully connected layer of the domain sharing module contains 4 neurons. Its output can be used to obtain the predicted label of the bearing through the SoftMax function. The predicted label and the true label form the cross entropy loss, which is used to train the domain sharing module. The cross entropy loss formula is as follows:

[0074]

[0075] Among them, L C represents the cross entropy loss, y i represents the true label of the i-th source domain data, represents the predicted label of the i-th source domain data, n s It represents the number of samples in the source domain, and log represents the logarithm with base e.

[0076] S3.2: Domain discrimination module

[0077] The domain discrimination module consists of a three-layer fully connected network and a ReLU activation function. Its detailed structure is as follows Figure 4 As shown in Figure 1, the input of the domain discrimination module is the high-level features extracted by the second fully connected layer of the domain sharing module, which maps these high-level features to the domain label space; the output (i.e., the third layer of the fully connected network of the domain discrimination module) is a single neuron node.

[0078] Wasserstein distance can represent the similarity between two data distributions: the smaller the distance, the greater the similarity between the two data distributions; the larger the distance, the less similar the two data distributions. Wasserstein distance is used as the loss function of the domain discrimination module, and the domain discrimination module is trained by maximizing the Wasserstein distance.

[0079] The loss function of the domain discrimination module is as follows:

[0080]

[0081] Among them, L adv Represents the loss function of the domain discrimination module, namely Wasserstein distance; represents the output value of the time domain discrimination module of the i-th data feature in the input source domain, represents the output value of the time domain discrimination module of the jth data feature in the input target domain; n t Indicates the amount of data in the target domain.

[0082] After the three-layer fully connected network training of the domain discrimination module is completed, the domain asymmetry factor is generated by formula (4):

[0083]

[0084] Among them, W i s represents the domain asymmetry factor of the i-th source domain data; σ s Represents the Sigmoid function.

[0085] A larger domain asymmetry factor is assigned to source domain samples that share a fault state, while a smaller factor is assigned to source domain samples that do not have a fault state in the target domain. In addition, the degree of data imbalance in the target domain also results in a smaller domain asymmetry factor corresponding to samples in the source domain that have the same fault state as the target domain and a smaller number of samples.

[0086] S3.3: Adaptive Module

[0087] The input of the adaptation module is also the output of the second fully connected layer of the domain sharing module. The weighted overall distribution distance and weighted local distribution distance of the source domain and the target domain are calculated by the maximum mean difference.

[0088] The maximum mean difference is a nonparametric distance metric that measures the distance between two different data distributions in a reproducible kernel Hilbert space. Its expression is as follows:

[0089]

[0090] Where L represents the distribution distance between the source domain and the target domain, φ(·) represents the mapping from the original data space to the reproducible kernel Hilbert space, and represent the characteristics of the source domain and the target domain respectively, Operator representing the distance between different data distributions in a reproducible kernel Hilbert space.

[0091] For the weighted overall distribution distance, the domain asymmetry factor is used to assign corresponding weights to the source domain data. The shared bearing state contributes more to the data distribution difference measurement, so it is assigned a larger weight. The bearing state unique to the source domain contributes less to the data distribution measurement, so it is assigned a smaller weight. Then, the maximum mean difference is used to calculate the overall distribution distance between the weighted source domain data and the target domain data. The measurement formula is as follows:

[0092]

[0093] Among them, L g Represents the weighted overall distribution distance between the source domain and target domain data.

[0094] For the weighted local distribution distance, based on the domain asymmetry factor weight, the distribution distance between each type of fault data in the source domain and the target domain is calculated separately. The weighted local distribution distance can measure the distribution distance between the source domain and the target domain in a more detailed manner, effectively avoiding the confusion of different fault data in the source domain and the target domain, and facilitating better migration from the source domain to the target domain. In order to conveniently calculate the weighted local distribution distance, the category weight is defined, and its calculation formula is shown in Equation (7):

[0095]

[0096] in, Indicates the category weight of the i-th data belonging to category c, y ic represents the probability that the i-th data belongs to category c. For source domain data, the unique encoding of the source domain label can be directly used as y ic ; For the target domain data, although there is no label, the output information of the domain sharing module is a probability distribution after the SoftMax function, which can be used as y ic .

[0097] Therefore, the local distribution difference measurement formula can be obtained, as shown in formula (8):

[0098]

[0099] Among them, L l Represents the weighted local distribution distance between the source domain and the target domain data, and They represent the category weights of the source domain and target domain data respectively, and C represents the number of categories of the source domain data.

[0100] Therefore, based on the above description, the migration loss function of the adaptive module is as follows:

[0101] L DA =L g +μL l (9)

[0102] Among them, μ is the weight factor for calculating the local distribution difference, and its value range is [0, 1]. Since the prediction information of the domain sharing module is not reliable in the early stage of model training, the measurement of local distribution difference is not accurate enough, so the initial value of μ is set to 0. With the iteration of model training, the credibility of the prediction information gradually increases, and the weight of the local distribution difference also gradually increases, eventually reaching 1. The calculation formula of μ is shown in formula (10):

[0103]

[0104] Among them, p is the training progress of the model, and its value range is [0,1]. γ is a parameter used to control the change of the weight factor μ, which is generally set to 10.

[0105] S4: Rolling bearing migration fault diagnosis model training

[0106] S4.1: Training of domain discrimination module

[0107] The three-layer fully connected network of the domain discrimination module is trained by maximizing the Wasserstein distance, namely:

[0108]

[0109] Among them, θ adv Represents the training parameters of the domain discrimination module.

[0110] After training, the domain asymmetry factor is generated through the output (the third layer of the fully connected network of the domain discrimination module).

[0111] S4.2: Training of Domain Sharing Module

[0112] The domain sharing module is trained by minimizing the classification loss of the domain sharing module and the transfer loss of the adaptation module, i.e.:

[0113]

[0114] Among them, θ represents the training parameters of the domain sharing module; λ is the migration weight, which represents the proportion of migration loss in the total loss. It needs to be set to a moderate value. When λ is very small, the data of the source domain and the target domain cannot be well migrated, resulting in low accuracy of the model in the target domain. When λ is very large, the parameters of the domain sharing module cannot be well trained, resulting in a decrease in the fault classification ability of the model. The default value of λ can be set to 5 and needs to be adjusted appropriately according to the model training process.

[0115] S5: Troubleshooting

[0116] After training, the rolling bearing migration fault diagnosis model acquires the ability to perform fault diagnosis on the data. After obtaining the test data, the fault data is standardized and fed into the domain sharing module of the rolling bearing migration fault diagnosis model, which then outputs the predicted status of the fault data.

[0117] Example 2:

[0118] Embodiment 2 of the present invention provides a rolling bearing migration fault diagnosis system under domain asymmetry, including:

[0119] The data acquisition module is configured to: collect acceleration vibration signals of the rolling bearing and perform preprocessing;

[0120] The fault diagnosis module is configured to obtain a fault diagnosis result based on the preprocessed acceleration vibration signal and the pre-trained rolling bearing migration fault diagnosis model;

[0121] Among them, the rolling bearing migration fault diagnosis model in the fault diagnosis module includes: a domain sharing module, a domain discrimination module and an adaptive module;

[0122] The domain sharing module at least includes a convolutional layer, a maximum pooling layer, a first fully connected layer, a second fully connected layer, a third fully connected layer and a SoftMax activation function connected in sequence;

[0123] The domain discrimination module takes the output of the second fully connected layer of the domain sharing module as input, trains the domain discrimination module by maximizing the Wasserstein distance, and generates a domain asymmetry factor based on the domain discrimination module;

[0124] The adaptive module takes the output of the second fully connected layer of the domain sharing module as input, combines the domain asymmetry factor, and calculates the weighted overall distribution distance and weighted local distribution distance between the source and target domains through the maximum mean difference. The loss function of the adaptive module is derived based on the weighted overall distribution distance and weighted local distribution distance.

[0125] The domain sharing module is trained by minimizing the cross entropy loss of the domain sharing module and the migration loss of the adaptive module, and the output of the trained domain sharing module is used as the fault diagnosis result.

[0126] The working method of the system is the same as the rolling bearing migration fault diagnosis method under domain asymmetry provided in Example 1, and will not be repeated here.

[0127] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, 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 magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0128] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts 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, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0129] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0130] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0131] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0132] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A rolling bearing migration fault diagnosis method under domain asymmetry, characterized in that: The following processes are included: Collect the acceleration vibration signal of the rolling bearing and perform preprocessing; Obtain the fault diagnosis result based on the pre-processed acceleration vibration signal and the pre-trained rolling bearing migration fault diagnosis model; Among them, the rolling bearing migration fault diagnosis model includes: domain sharing module, domain discrimination module and adaptive module; The domain sharing module at least includes a convolutional layer, a maximum pooling layer, a first fully connected layer, a second fully connected layer, a third fully connected layer and a SoftMax activation function connected in sequence; The domain discrimination module takes the output of the second fully connected layer of the domain sharing module as input, trains the domain discrimination module by maximizing the Wasserstein distance, and generates a domain asymmetry factor based on the domain discrimination module; The adaptive module takes the output of the second fully connected layer of the domain sharing module as input, combines the domain asymmetry factor, and calculates the weighted overall distribution distance and weighted local distribution distance between the source and target domains through the maximum mean difference. The loss function of the adaptive module is derived based on the weighted overall distribution distance and weighted local distribution distance. The domain sharing module is trained by minimizing the cross entropy loss of the domain sharing module and the migration loss of the adaptive module, and the output of the trained domain sharing module is used as the fault diagnosis result; Wasserstein distance ,include: in, Indicates the input source domain The output value of the time domain discrimination module of the data feature, Indicates the input target domain The output value of the time domain discrimination module of the data feature; Indicates the amount of data in the source domain, Indicates the amount of data in the target domain; Domain asymmetry factors, including: in, Indicates the The domain asymmetry factor of the source domain data; Represents the Sigmoid function; Weighted local distribution distance ,include: in, and are the category weights of source domain data and target domain data respectively, is the number of categories of source domain data, is the mapping from the original data space to the reproducible kernel Hilbert space, and are the features of the source domain and the target domain respectively, is an operator for different data distribution distances in the reproducible kernel Hilbert space, Indicates the The domain asymmetry factor of the source domain data.

2. The rolling bearing migration fault diagnosis method under domain asymmetry according to claim 1, characterized in that: The domain sharing module includes at least a convolutional layer, a maximum pooling layer, a first fully connected layer, a second fully connected layer, a third fully connected layer and a SoftMax activation function connected in sequence.

3. The rolling bearing migration fault diagnosis method under domain asymmetry according to claim 1, characterized in that: Weighted overall distribution distance ,include: in, is the mapping from the original data space to the reproducible kernel Hilbert space, and are the features of the source domain and the target domain respectively, is an operator for different data distribution distances in the reproducible kernel Hilbert space, Indicates the The domain asymmetry factor of the source domain data.

4. The rolling bearing migration fault diagnosis method under domain asymmetry according to claim 3, characterized in that: Transfer loss function of adaptive module ,include: ； in, , is the training progress of the model, To control Changing parameters.

5. The rolling bearing migration fault diagnosis method under domain asymmetry according to claim 1, characterized in that: Minimize the cross entropy loss of the domain-sharing module and the transfer loss of the adaptive module, including: in, is the cross entropy loss of the domain sharing module, is the migration loss of the adaptive module, is the migration weight.

6. A rolling bearing migration fault diagnosis system under domain asymmetry, characterized in that: The rolling bearing migration fault diagnosis method under domain asymmetry conditions according to any one of claims 1 to 5 comprises: The data acquisition module is configured to: collect acceleration vibration signals of the rolling bearing and perform preprocessing; The fault diagnosis module is configured to obtain a fault diagnosis result based on the preprocessed acceleration vibration signal and the pre-trained rolling bearing migration fault diagnosis model; Among them, the rolling bearing migration fault diagnosis model in the fault diagnosis module includes: a domain sharing module, a domain discrimination module and an adaptive module; The domain sharing module at least includes a convolutional layer, a maximum pooling layer, a first fully connected layer, a second fully connected layer, a third fully connected layer and a SoftMax activation function connected in sequence; The domain discrimination module takes the output of the second fully connected layer of the domain sharing module as input, trains the domain discrimination module by maximizing the Wasserstein distance, and generates a domain asymmetry factor based on the domain discrimination module; The adaptive module takes the output of the second fully connected layer of the domain sharing module as input, combines the domain asymmetry factor, and calculates the weighted overall distribution distance and weighted local distribution distance between the source and target domains through the maximum mean difference. The loss function of the adaptive module is derived based on the weighted overall distribution distance and weighted local distribution distance. The domain sharing module is trained by minimizing the cross entropy loss of the domain sharing module and the migration loss of the adaptive module, and the output of the trained domain sharing module is used as the fault diagnosis result.

7. The rolling bearing migration fault diagnosis system under domain asymmetry according to claim 6, characterized in that: Minimize the cross entropy loss of the domain-sharing module and the transfer loss of the adaptive module, including: in, is the cross entropy loss of the domain sharing module, is the migration loss of the adaptive module, is the migration weight.

Images