Automobile motor bearing fault diagnosis model construction method and device

Abstract

The invention provides an automobile motor bearing fault diagnosis model construction method and device. The method comprises the following steps: acquiring a data sample and a pre-constructed multi-adversarial multi-domain adaptive network; performing data preprocessing on the data sample to obtain preprocessed data; inputting the preprocessed data into a multi-adversarial multi-domain adaptive network, and calculating LMMD loss and subnet self-adversarial loss of each source-target domain adaptive subnet classifier; optimizing network parameters of the multi-adversarial multi-domain adaptive network according to the LMMD loss and the adversarial loss to obtain an optimized diagnosis network; and when a training termination condition is met, determining the finally optimized diagnosis network as a trained motor bearing fault diagnosis model. According to the method and the device, a multi-adversarial multi-domain self-adaptive network can be constructed, effective utilization of unmarked signals is realized, the data acquisition cost and difficulty are reduced, and thus the accuracy and the reliability of fault diagnosis of a label-free cross-working-condition motor bearing are improved.

Description

Technical Field

[0001] The present application relates to the technical field of vehicle fault diagnosis, and in particular to a method and device for constructing a fault diagnosis model for an automobile motor bearing. Background Art

[0002] With the rapid development of the new energy vehicle industry, vehicle safety has become the focus of consumer attention. As the core component of new energy vehicles, the performance of electric motors directly affects the safety and reliability of the entire vehicle. As a key component of electric motors, rolling bearing fault diagnosis is particularly critical. In the prior art, by collecting vibration and sound signals from the main drive reducer of the rolling mill, continuous wavelet transform and variational mode decomposition technology are used, combined with a two-dimensional convolutional neural network, to achieve high-precision diagnosis of unlabeled cross-operating motor bearing faults. However, in practice, it is found that existing methods mainly rely on high-quality, labeled data sets, but the data collection cost is high and the labeling difficulty is great, resulting in limited diagnostic accuracy. Summary of the invention

[0003] The purpose of the embodiments of the present application is to provide a method and device for constructing an automobile motor bearing fault diagnosis model. By constructing a multi-adversarial multi-domain adaptive network, the effective use of unlabeled signals is achieved, the cost and difficulty of data collection are reduced, and the accuracy and reliability of unlabeled cross-operating condition motor bearing fault diagnosis are improved.

[0004] The first aspect of the present application provides a method for constructing a fault diagnosis model for an automobile motor bearing, comprising: Acquire data samples and a pre-built multi-adversarial multi-domain adaptive network; wherein the data samples include a labeled source domain data set and an unlabeled target domain data set; the multi-adversarial multi-domain adaptive network includes a source target domain adaptive subnet, a multi-source domain adversarial network, multiple source target domain adaptive subnet classifiers, and a multi-subnet joint decision module; Performing data preprocessing on the data samples to obtain preprocessed data; Input the preprocessed data into a multi-adversarial multi-domain adaptive network, and calculate the LMMD loss and the subnet self-adversarial loss of each of the source target domain adaptive subnet classifiers; According to the LMMD loss and the adversarial loss, the network parameters of the multi-adversarial multi-domain adaptive network are optimized to obtain an optimized diagnostic network; When the training termination condition is reached, the final optimized diagnosis network is determined as the trained motor bearing fault diagnosis model.

[0005] Based on the above implementation process, this method can construct a multi-adversarial multi-domain adaptive network to achieve effective utilization of unlabeled signals, reduce the cost and difficulty of data acquisition, and thus improve the accuracy and reliability of unlabeled cross-operating condition motor bearing fault diagnosis.

[0006] Further, the source domain data set includes sensor collected data of laboratory bearing operation, and the target domain data set includes collected data of vehicle built-in sensors; The source-target domain adaptation subnetwork includes four attention encoder layers and a fully connected network; The multi-source domain adversarial network includes three fully connected layers.

[0007] Furthermore, the data samples are preprocessed to obtain preprocessed data, including: Performing missing value processing on the data sample to obtain missing value processed data; The missing value processed data is subjected to data denoising processing based on an attention mechanism to obtain preprocessed data.

[0008] Furthermore, performing missing value processing on the data sample to obtain missing value processed data includes: When the sample size of the data sample is greater than a preset sample size threshold, performing missing value deletion processing on the data sample to obtain missing value processed data; When the sample size of the data sample is not greater than a preset sample size threshold, mean interpolation processing is performed on the missing values ​​of the data sample to obtain missing value processed data.

[0009] Further, the inputting the preprocessed data into the multi-adversarial multi-domain adaptive network and calculating the LMMD loss and the subnet self-adversarial loss of each source target domain adaptive subnet classifier includes: The pre-processed data is input into the source-target domain adaptive subnet for processing to obtain a plurality of source domain feature information in a unified feature space; Calculating the LMMD loss according to the source domain feature information; wherein the LMMD loss includes a first LMMD loss of each of the source target domain adaptive subnet classifiers on the source domain and a second LMMD loss of each of the source target domain adaptive subnet classifiers on the target domain; According to the multi-source domain adversarial network and the source domain feature information, the subnet self-adversarial loss of each source target domain adaptive subnet classifier is calculated.

[0010] Furthermore, the network parameters of the multi-adversarial multi-domain adaptive network are optimized according to the LMMD loss and the adversarial loss to obtain an optimized diagnostic network, including: According to the LMMD loss and the subnet's own adversarial loss, adjusting the confidence score of each of the source-target domain adaptive subnet classifiers to obtain an optimized classifier; Optimizing the parameters of the source-target domain adaptive subnet according to the LMMD loss to obtain an optimized adaptive subnet; The multi-adversarial multi-domain adaptive network is updated according to the optimized classifier and the optimized adaptive subnet to obtain an optimized diagnosis network.

[0011] Furthermore, the method further comprises: Collect real-time bearing operation data through sensors built into the vehicle to be diagnosed; Performing feature extraction on the real-time bearing operation data through the source-target domain adaptive subnet to obtain extracted features; Classifying the extracted features by the source-target domain adaptive subnet classifier to obtain a classification result; According to the confidence scores corresponding to the multi-subnet joint decision module and the source-target domain adaptive subnet classifier, the classification results are collaboratively decided to obtain a fault diagnosis result.

[0012] A second aspect of the present application provides a vehicle motor bearing fault diagnosis model construction device, the vehicle motor bearing fault diagnosis model construction device comprising: A first acquisition unit is used to acquire data samples and a pre-built multi-adversarial multi-domain adaptive network; wherein the data samples include a labeled source domain data set and an unlabeled target domain data set; the multi-adversarial multi-domain adaptive network includes a source target domain adaptive subnet, a multi-source domain adversarial network, a plurality of source target domain adaptive subnet classifiers and a multi-subnet joint decision module; A preprocessing unit, used for performing data preprocessing on the data sample to obtain preprocessed data; A loss calculation unit, used for inputting the preprocessed data into a multi-adversarial multi-domain adaptive network, and calculating the LMMD loss and the subnet self-adversarial loss of each of the source target domain adaptive subnet classifiers; A parameter optimization unit, used for optimizing the network parameters of the multi-adversarial multi-domain adaptive network according to the LMMD loss and the adversarial loss to obtain an optimized diagnostic network; The determination unit is used to determine the final optimized diagnosis network as the trained motor bearing fault diagnosis model when the training termination condition is reached.

[0013] Further, the source domain data set includes sensor collected data of laboratory bearing operation, and the target domain data set includes collected data of vehicle built-in sensors; The source-target domain adaptation subnetwork includes four attention encoder layers and a fully connected network; The multi-source domain adversarial network includes three fully connected layers.

[0014] Furthermore, the preprocessing unit comprises: A first processing subunit is used to perform missing value processing on the data sample to obtain missing value processed data; The denoising subunit is used to perform data denoising processing based on the attention mechanism on the missing value processing data to obtain preprocessed data.

[0015] Furthermore, the first processing subunit includes: A deletion module is used to perform missing value deletion processing on the data sample when the sample size of the data sample is greater than a preset sample size threshold to obtain missing value processed data; The interpolation module is used to perform mean interpolation processing on the missing values ​​of the data samples to obtain missing value processed data when the sample size of the data samples is not greater than a preset sample size threshold.

[0016] Furthermore, the loss calculation unit includes: A second processing sub-unit is used to input the pre-processed data into a source-target domain adaptive sub-network for processing, so as to obtain a plurality of source domain feature information in a unified feature space; A calculation subunit, configured to calculate the LMMD loss according to the source domain feature information; wherein the LMMD loss includes a first LMMD loss of each of the source-target domain adaptive subnet classifiers on the source domain and a second LMMD loss of each of the source-target domain adaptive subnet classifiers on the target domain; The calculation subunit is used to calculate the subnet self-adversarial loss of each source target domain adaptive subnet classifier according to the multi-source domain adversarial network and the source domain feature information.

[0017] Furthermore, the parameter optimization unit includes: An adjustment subunit, used for adjusting the confidence score of each of the source-target domain adaptive subnetwork classifiers according to the LMMD loss and the subnetwork's own adversarial loss to obtain an optimized classifier; An optimization subunit, configured to optimize the parameters of the source-target domain adaptive subnetwork according to the LMMD loss to obtain an optimized adaptive subnetwork; An updating subunit is configured to update the multi-adversarial multi-domain adaptive network according to the optimized classifier and the optimized adaptive subnet to obtain an optimized diagnosis network.

[0018] Furthermore, the automobile motor bearing fault diagnosis model building device also includes: A collection unit, used to collect real-time bearing operation data through sensors built into the vehicle to be diagnosed; An extraction unit, configured to extract features from the real-time bearing operation data through the source-target domain adaptive subnet to obtain extracted features; A classification unit, used for classifying the extracted features through the source-target domain adaptive subnet classifier to obtain a classification result; The diagnosis unit is used to perform collaborative decision processing on the classification results according to the confidence scores corresponding to the multi-subnet joint decision module and the source target domain adaptive subnet classifier to obtain a fault diagnosis result.

[0019] The third aspect of the present application provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the automobile motor bearing fault diagnosis model construction method described in any one of the first aspect of the present application.

[0020] The fourth aspect of the present application provides a computer-readable storage medium storing computer program instructions. When the computer program instructions are read and executed by a processor, the method for constructing an automobile motor bearing fault diagnosis model described in any one of the first aspect of the present application is executed.

[0021] The beneficial effects of the present application are as follows: the method and device can achieve better and simpler noise reduction, reducing the need for prior knowledge; at the same time, it can also retain the original distribution of bearing operation data to the maximum extent, improve the accuracy of downstream classification tasks, and minimize the computing cost and improve computing efficiency; then, based on the source-target domain adaptive subnet, it can solve the problem of large differences between source domain data and target domain data, so that it is more in line with the daily operation and maintenance of automobiles, while improving the diagnostic accuracy in the unlabeled target domain, the safety and reliability of new energy vehicles are greatly improved; then, the robustness of the network is improved through the multi-source domain adversarial model and generalization ability. Among them, through the calculation of adversarial loss, the model can automatically adapt to the data characteristics of different source domains during training, which can reduce the occurrence of overfitting, thereby improving the model's ability to cope with various complex situations, and can more effectively utilize existing data, reduce the cost of data acquisition, and improve the efficiency of fault diagnosis; finally, the multi-subnet collaborative decision-making mechanism can integrate the outputs of multiple sub-networks and make full use of the complementarity between different sub-networks, thereby improving the accuracy and robustness of diagnosis, so that it can better cope with complex and changeable motor bearing fault conditions and reduce the risks of misjudgment and missed judgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0023] Figure 1 A schematic diagram of a flow chart of a method for constructing a fault diagnosis model for an automobile motor bearing provided in an embodiment of the present application; Figure 2 A schematic flow chart of another method for constructing a fault diagnosis model for an automobile motor bearing provided in an embodiment of the present application; Figure 3 A schematic diagram of a multi-adversarial multi-domain adaptive network structure provided in an embodiment of the present application; Figure 4 A schematic diagram of a signal preprocessing process provided in an embodiment of the present application; Figure 5 A schematic diagram of the structure of a source-target domain adaptive subnet provided in an embodiment of the present application; Figure 6 A schematic diagram of a multi-source domain adversarial network and a source-target domain adaptive subnet learning common features of multiple domains provided in an embodiment of the present application; Figure 7 A schematic diagram of the structure of a vehicle motor bearing fault diagnosis model building device provided in an embodiment of the present application; Figure 8 A schematic structural diagram of another automobile motor bearing fault diagnosis model building device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0025] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0026] Example 1 Please see Figure 1 , Figure 1 A flow chart of a method for constructing a fault diagnosis model for an automobile motor bearing provided in this embodiment. The method for constructing a fault diagnosis model for an automobile motor bearing includes: S101, obtaining data samples and a pre-built multi-adversarial multi-domain adaptive network.

[0027] In this embodiment, the data samples include a labeled source domain data set and an unlabeled target domain data set; the multi-adversarial multi-domain adaptive network includes a source-target domain adaptive subnet, a multi-source domain adversarial network, multiple source-target domain adaptive subnet classifiers and a multi-subnet joint decision module.

[0028] In this embodiment, the source domain data set includes sensor data collected by laboratory bearing operation, and the target domain data set includes data collected by vehicle built-in sensors; The source-target domain adaptation subnetwork consists of four attention encoder layers and a fully connected network; The multi-source domain adversarial network consists of three fully connected layers.

[0029] S102: Perform data preprocessing on the data samples to obtain preprocessed data.

[0030] S103: Input the preprocessed data into the multi-adversarial multi-domain adaptive network, and calculate the LMMD loss and the subnet self-adversarial loss of each source target domain adaptive subnet classifier.

[0031] S104. According to the LMMD loss and the adversarial loss, the network parameters of the multi-adversarial multi-domain adaptive network are optimized to obtain an optimized diagnostic network.

[0032] S105. When the training termination condition is reached, the finally optimized diagnosis network is determined as the trained motor bearing fault diagnosis model.

[0033] In this embodiment, the method aims at the problem of unlabeled cross-operating motor bearing fault diagnosis, and proposes a method for constructing an automobile motor bearing fault diagnosis model. The model is constructed based on a multi-adversarial multi-domain adaptive network (MMAN for short), and can be used to effectively diagnose cross-operating motor bearing faults. Among them, the multi-adversarial multi-domain adaptive network can realize the effective use of unlabeled signals, thereby improving the accuracy and reliability of unlabeled cross-operating motor bearing fault diagnosis.

[0034] In this embodiment, the method first designs a shared feature extractor and adopts a multi-head self-attention encoder to capture the temporal correlation in the vibration signal. Then, by introducing an adversarial module and calculating the adversarial loss of multiple source domains, the robustness and generalization ability of the network are enhanced. In addition, the method also constructs a source-target domain adaptive subnetwork, which can use the Local Maximum Mean Discrepancy (LMMD) loss function to achieve category alignment between the source domain and the target domain. At the same time, at the decision-making level, the method proposes a collaborative decision-making strategy that can achieve more accurate fault diagnosis through the fusion of multiple subnetworks. Among them, the strategy uses LMMD loss to calculate the confidence score, which can help multi-subnetwork classifiers make more intelligent fusion decisions, thereby helping to improve the overall accuracy of fault diagnosis.

[0035] In this embodiment, the execution subject of the method may be a computing device such as a computer or a server, and no limitation is made in this embodiment.

[0036] In this embodiment, the execution subject of the method may also be a smart device such as a smart phone, a tablet computer, etc., which is not limited in this embodiment.

[0037] It can be seen that the implementation of the automobile motor bearing fault diagnosis model construction method described in this embodiment can construct a multi-adversarial multi-domain adaptive network, realize the effective use of unlabeled signals, reduce the cost and difficulty of data collection, and thus improve the accuracy and reliability of unlabeled cross-operating condition motor bearing fault diagnosis.

[0038] Example 2 Please see Figure 2 , Figure 2 A flow chart of a method for constructing a fault diagnosis model for an automobile motor bearing provided in this embodiment. The method for constructing a fault diagnosis model for an automobile motor bearing includes: S201, obtaining data samples and a pre-built multi-adversarial multi-domain adaptive network.

[0039] In this embodiment, the data samples include a labeled source domain data set and an unlabeled target domain data set; the multi-adversarial multi-domain adaptive network includes a source-target domain adaptive subnet, a multi-source domain adversarial network, multiple source-target domain adaptive subnet classifiers and a multi-subnet joint decision module.

[0040] In this embodiment, the source domain data set includes sensor data collected by laboratory bearing operation, and the target domain data set includes data collected by vehicle built-in sensors; The source-target domain adaptation subnetwork consists of four attention encoder layers and a fully connected network; The multi-source domain adversarial network consists of three fully connected layers.

[0041] In this embodiment, the data sample can be the data collected by the transmission laboratory according to the transmission system category and load (such as the data collected and recorded by the sensor of the laboratory bearing operation). Among them, the method uses this data as the source domain data set, which makes the method no longer need to be designed separately for the operation conditions of different individual bearings of new energy vehicles, thereby greatly reducing the data collection cost and diagnosis cost. As for the target domain data, the data is the actual new energy vehicle bearing operation data (such as data automatically collected by the vehicle's built-in sensor system, which is unlabeled data). After data processing in the target domain and the source domain, it can be passed into the model training to obtain the automobile motor bearing fault diagnosis model (that is, the new energy vehicle bearing health diagnosis model).

[0042] In this embodiment, the pre-built multi-adversarial multi-domain adaptive network structure is as follows: Figure 3 As shown in FIG. 1 , the public feature extractor is a source-target domain adaptive subnet, and the state classifier is a source-target domain adaptive subnet classifier.

[0043] S202: Perform missing value processing on the data samples to obtain missing value processed data.

[0044] As an optional implementation, performing missing value processing on the data sample to obtain missing value processed data includes: When the sample size of the data sample is greater than the preset sample size threshold, the data sample is processed by missing value deletion to obtain missing value processed data; When the sample size of the data sample is not greater than the preset sample size threshold, the missing values ​​of the data sample are interpolated by the mean to obtain missing value processed data.

[0045] In this embodiment, due to the limitation of data scale and the complexity of sensor acquisition and transmission, there are indeed cases of missing data and obvious anomalies. These missing or abnormal data cannot accurately represent the characteristics of the bearing, thus affecting the accuracy of fault diagnosis.

[0046] Therefore, in order to overcome these difficulties and improve the accuracy and stability of fault diagnosis, this method provides two effective solutions. The first is the deletion method, that is, when the data at a certain moment is abnormal or missing, the data at that moment is directly deleted, and the information collected at the next moment is filled in the missing position in sequence. This method is simple and easy to implement, and can effectively avoid the training difficulties or network fluctuations caused by the inability of the neural network to handle missing data, thereby improving the network training speed. However, it should be noted that the deletion method may lead to a reduction in the amount of data, which in turn affects the generalization ability of the neural network.

[0047] Therefore, this method also provides a second solution: the filling method. When the amount of data is too small or the feature data is insufficient, directly deleting the data may cause the network generalization ability to decrease, thereby affecting the diagnostic accuracy. In order to solve this problem, this method uses the mean interpolation method to interpolate the data. Based on the basic assumption that time series data have similar distributions, missing data can be estimated by taking the mean of the data at the previous moment and the next moment. Similarly, for batch data, the interpolation value can also be obtained by taking the mean of the corresponding data of the previous batch and the next batch. This method can effectively make up for the problems caused by missing data and improve the integrity and accuracy of the data.

[0048] S203, performing data denoising processing based on the attention mechanism on the missing value processing data to obtain preprocessed data.

[0049] In this embodiment, the bearing will inevitably be affected by background environments such as road bumps and street noise during operation. These interference factors will generate a lot of noise, further interfering with the accuracy of the data. If these problems are not properly handled, the performance of the neural network will be greatly weakened and the reliability of fault diagnosis will be reduced. In order to solve the problem of data denoising, the method proposes to introduce an attention mechanism. By extracting the associated information in the bearing timing information, the attention mechanism can effectively suppress the interference of noise and highlight the characteristics of the bearing operation law. Based on the basic assumption that the noise distribution is a Gaussian distribution and the observation that the bearing operation conforms to certain laws, the black box model characteristics of the neural network can be fully utilized to extract the regular characteristics of the bearing operation in the high-dimensional space, thereby achieving effective suppression of noise and accurate data extraction.

[0050] Please see Figure 4 , Figure 4 A schematic diagram of the signal preprocessing process is shown. Based on the schematic diagram, it can be seen that this method can achieve comprehensive optimization processing of the collected signals by comprehensively using methods such as deletion, mean interpolation and attention mechanism, improve the integrity and accuracy of the data, and provide reliable data support for subsequent fault diagnosis. It can be seen that this helps to improve the accuracy and efficiency of bearing fault diagnosis of new energy vehicles and provide strong guarantee for the safe operation of new energy vehicles.

[0051] S204: input the preprocessed data into the source-target domain adaptive subnet for processing, and obtain a plurality of source domain feature information in a unified feature space.

[0052] In this embodiment, the method inputs the processed data into the source-target domain adaptive subnet. Its core purpose is to integrate different source domain and target domain data into a unified feature space, and to achieve accurate alignment of different categories in these domains by introducing the LMMD (local maximum mean difference) loss function.

[0053] Please see Figure 5 , Figure 5 The schematic diagram of the structure of the source-target domain adaptive subnet (i.e., source-target domain adaptive subnet) is shown. The source-target domain adaptive subnet mainly includes four attention encoder layers (encoder layer 1 to encoder layer 4) and a fully connected network F1. Each encoder layer has powerful feature extraction and conversion capabilities, and can extract valuable feature information from the original data layer by layer. The introduction of the attention mechanism enables the subnet to adaptively focus on the part of the data that is most relevant to fault diagnosis, thereby improving the expressiveness of the features. The fully connected network F1 is responsible for further integrating and converting the features extracted by the encoder layer, and finally outputting them to a unified feature space. In this process, the use of the LMMD loss function plays a key role. It achieves precise alignment of these categories in the feature space by comparing and measuring the feature distributions of different categories in the source and target domains.

[0054] In this embodiment, the specific parameter settings of the source-target domain adaptive subnet are shown in Table 1, where the detailed parameters of each encoder layer and the fully connected network F1 are carefully designed and optimized to ensure that the subnet can efficiently and accurately complete the alignment task of the source target domain.

[0055] Table 1 Specific parameters of the source-target domain adaptive subnet

[0056] In this embodiment, the method can successfully realize the integration and alignment of different source domain and target domain data in the feature space based on the source-target domain adaptive subnet, thereby providing strong technical support for the cross-domain diagnosis of new energy vehicle bearing faults, which in turn is conducive to improving the accuracy and efficiency of fault diagnosis. The coast can provide new ideas and directions for research and application in related fields.

[0057] S205: Calculate LMMD loss according to source domain feature information.

[0058] In this embodiment, the LMMD loss includes a first LMMD loss of each source-target domain adaptive subnet classifier on the source domain and a second LMMD loss of each source-target domain adaptive subnet classifier on the target domain.

[0059] S206. Calculate the subnet self-adversarial loss of each source target domain adaptive subnet classifier according to the multi-source domain adversarial network and source domain feature information.

[0060] In this embodiment, the multi-source domain adversarial network is combined with the source-target domain adaptive subnet to improve the performance of the model. The multi-source domain adversarial network is designed to accurately identify the unique attributes of multi-source domain features, and jointly mine the shared features between multi-source domains and target domains through collaborative work with the source-target domain adaptive subnet.

[0061] In this embodiment, the multi-source domain adversarial network can be composed of three fully connected layers. The multi-source domain adversarial network uses the principle of adversarial loss to fuse and compare the feature information of different source domains, thereby achieving in-depth analysis of source domain features. At the same time, the source-target domain adaptive subnet focuses on capturing the shared features between the source domain and the target domain, providing more comprehensive and accurate information for bearing fault diagnosis.

[0062] Please see Figure 6 , Figure 6 The figure shows a schematic diagram of the common features of multi-domains learned by the multi-source domain adversarial network and the source-target domain adaptive subnet. The method can build an intelligent and flexible bearing fault diagnosis model by combining these two networks, so that the model can not only make full use of the data resources of multi-source domains, but also effectively deal with complex situations in the target domain, and improve the accuracy and generalization ability of diagnosis.

[0063] S207. According to the LMMD loss and the subnet's own adversarial loss, the confidence score of each source-target domain adaptive subnet classifier is adjusted to obtain an optimized classifier.

[0064] S208. Optimize the parameters of the source-target domain adaptive subnet according to the LMMD loss to obtain an optimized adaptive subnet.

[0065] S209, updating the multi-adversarial multi-domain adaptive network according to the optimized classifier and the optimized adaptive subnet to obtain an optimized diagnosis network.

[0066] In this embodiment, during the training and updating process, the method also adopts a multi-source-target domain adaptive subnet collaborative decision-making strategy. Among them, the strategy comprehensively considers the LMMD loss of the source domain and the target domain and the adversarial loss of each subnet, and uses it as an evaluation indicator. Based on these indicators, confidence scores can be further calculated to evaluate and compare the effectiveness of multiple source-target domain adaptive subnet sets in feature mapping. It can be seen that this process contributes to the collaborative decision-making of multi-domain diagnostic information.

[0067] Specifically, the strategy can first calculate the LMMD loss of each source-target domain adaptive subnetwork classifier on the source domain and target domain as well as the adversarial loss of the subnetwork itself. These loss values ​​can reflect the performance of the subnetwork in feature extraction and classification tasks. Subsequently, a confidence score is assigned to each source-target domain adaptive subnetwork classifier based on these loss values. When calculating the confidence score, the LMMD loss and adversarial loss should be considered comprehensively to ensure a comprehensive evaluation of the performance of the source-target domain adaptive subnetwork on the source domain and target domain. Finally, these confidence scores are used for collaborative decision-making. The results of the source-target domain adaptive subnetwork classifier are weighted averaged or voted according to the confidence score to obtain the final fault diagnosis result.

[0068] This implementation method can make full use of multi-source domain information to improve the accuracy and reliability of diagnosis results. The calculation formula is as follows:

[0069] In the formula, represents the classification probability of the target domain samples provided by the i-th group of source-target domain adaptive subnetwork classifiers.

[0070] S210. When the training termination condition is reached, the finally optimized diagnosis network is determined as the trained motor bearing fault diagnosis model.

[0071] As an optional implementation, real-time bearing operation data is collected through sensors built into the vehicle to be diagnosed; The feature extraction of real-time bearing operation data is performed through the source-target domain adaptive subnet to obtain the extracted features; The extracted features are classified by the source-target domain adaptive subnet classifier to obtain the classification result; According to the confidence scores corresponding to the multi-subnet joint decision module and the source-target domain adaptive subnet classifier, the classification results are processed through collaborative decision-making to obtain the fault diagnosis results.

[0072] In this embodiment, the method can use these confidence scores to make collaborative decisions at the end. The results of the source-target domain adaptive subnet classifier are weighted averaged or voted according to the confidence scores to obtain the final fault diagnosis result.

[0073] In this embodiment, the execution subject of the method may be a computing device such as a computer or a server, and no limitation is made in this embodiment.

[0074] In this embodiment, the execution subject of the method may also be a smart device such as a smart phone, a tablet computer, etc., which is not limited in this embodiment.

[0075] It can be seen that the implementation of the automobile motor bearing fault diagnosis model construction method described in this embodiment can significantly improve the ability to capture key features in the motor bearing timing signal through a feature extraction network based on a multi-head self-attention encoder; using a neural network to replace the original signal processing-based method, and using a data-driven method to replace the transmission system fault diagnosis method based on a large amount of prior knowledge, can improve the diagnostic accuracy and efficiency while reducing the cost of fault diagnosis; at the same time, the source-target domain alignment adaptive subnet can solve the problem of inconsistent data distribution in the cross-domain diagnosis of new energy vehicle motor bearing faults. Through this mechanism, data from different source domains and target domains can be effectively integrated and aligned in the feature space, so that the model can better adapt to the operating environment of new energy vehicles under different working conditions, and improve the accuracy and reliability of diagnosis; then, the introduction of multi-source domain adversarial networks can significantly enhance the robustness and generalization ability of the network by calculating the adversarial loss of multiple source domains, so that the model can better adapt to various complex data characteristics during the training process, improve the model's adaptability to data from different source domains, and further enhance the stability and accuracy of fault diagnosis; finally, the multi-subnet collaborative decision-making mechanism can make full use of the complementarity between different sub-networks to realize the fusion of the outputs of multiple sub-networks, thereby reducing the errors and uncertainties that may be caused by a single classifier, and achieving more comprehensive and accurate fault diagnosis, thereby improving the accuracy and robustness of diagnosis, so that the model can better cope with complex and changeable bearing fault conditions.

[0076] Example 3 Please see Figure 7 , Figure 7 The schematic diagram of the structure of a vehicle motor bearing fault diagnosis model building device provided in this embodiment. Figure 7 As shown, the automobile motor bearing fault diagnosis model building device includes: The first acquisition unit 310 is used to acquire data samples and a pre-built multi-adversarial multi-domain adaptive network; wherein the data samples include a labeled source domain data set and an unlabeled target domain data set; the multi-adversarial multi-domain adaptive network includes a source target domain adaptive subnet, a multi-source domain adversarial network, multiple source target domain adaptive subnet classifiers and a multi-subnet joint decision module; The preprocessing unit 320 is used to perform data preprocessing on the data sample to obtain preprocessed data; A loss calculation unit 330 is used to input the preprocessed data into the multi-adversarial multi-domain adaptive network and calculate the LMMD loss and the subnet self-adversarial loss of each source target domain adaptive subnet classifier; A parameter optimization unit 340 is used to optimize the network parameters of the multi-adversarial multi-domain adaptive network according to the LMMD loss and the adversarial loss to obtain an optimized diagnosis network; The determination unit 350 is used to determine the finally optimized diagnosis network as the trained motor bearing fault diagnosis model when the training termination condition is reached.

[0077] In this embodiment, the explanation of the automobile motor bearing fault diagnosis model construction device can refer to the description in Example 1 or Example 2, and will not be further elaborated in this embodiment.

[0078] It can be seen that the automobile motor bearing fault diagnosis model construction device described in this embodiment can construct a multi-adversarial multi-domain adaptive network, realize the effective use of unlabeled signals, reduce the cost and difficulty of data collection, and thus improve the accuracy and reliability of unlabeled cross-operating condition motor bearing fault diagnosis.

[0079] Example 4 Please see Figure 8 , Figure 8 The schematic diagram of the structure of a vehicle motor bearing fault diagnosis model building device provided in this embodiment. Figure 8 As shown, the automobile motor bearing fault diagnosis model building device includes: The first acquisition unit 310 is used to acquire data samples and a pre-built multi-adversarial multi-domain adaptive network; wherein the data samples include a labeled source domain data set and an unlabeled target domain data set; the multi-adversarial multi-domain adaptive network includes a source target domain adaptive subnet, a multi-source domain adversarial network, multiple source target domain adaptive subnet classifiers and a multi-subnet joint decision module; The preprocessing unit 320 is used to perform data preprocessing on the data sample to obtain preprocessed data; A loss calculation unit 330 is used to input the preprocessed data into the multi-adversarial multi-domain adaptive network and calculate the LMMD loss and the subnet self-adversarial loss of each source target domain adaptive subnet classifier; A parameter optimization unit 340 is used to optimize the network parameters of the multi-adversarial multi-domain adaptive network according to the LMMD loss and the adversarial loss to obtain an optimized diagnosis network; The determination unit 350 is used to determine the finally optimized diagnosis network as the trained motor bearing fault diagnosis model when the training termination condition is reached.

[0080] In this embodiment, the source domain data set includes sensor data collected by laboratory bearing operation, and the target domain data set includes data collected by vehicle built-in sensors; The source-target domain adaptation subnetwork consists of four attention encoder layers and a fully connected network; The multi-source domain adversarial network consists of three fully connected layers.

[0081] As an optional implementation, the preprocessing unit 320 includes: The first processing subunit 321 is used to perform missing value processing on the data sample to obtain missing value processed data; The denoising subunit 322 is used to perform data denoising processing based on the attention mechanism on the missing value processing data to obtain preprocessed data.

[0082] As an optional implementation, the first processing subunit 321 includes: A deletion module is used to delete missing values ​​from data samples when the sample size of the data samples is greater than a preset sample size threshold, so as to obtain missing value processed data; The interpolation module is used to perform mean interpolation processing on the missing values ​​of the data samples when the sample size of the data samples is not greater than the preset sample size threshold to obtain missing value processed data.

[0083] As an optional implementation, the loss calculation unit 330 includes: The second processing sub-unit 331 is used to input the pre-processed data into the source-target domain adaptive sub-network for processing, and obtain a plurality of source domain feature information in a unified feature space; A calculation subunit 332 is used to calculate the LMMD loss according to the source domain feature information; wherein the LMMD loss includes a first LMMD loss of each source-target domain adaptive subnet classifier on the source domain and a second LMMD loss of each source-target domain adaptive subnet classifier on the target domain; The calculation subunit 332 is used to calculate the subnet self-adversarial loss of each source target domain adaptive subnet classifier according to the multi-source domain adversarial network and source domain feature information.

[0084] As an optional implementation, the parameter optimization unit 340 includes: An adjustment subunit 341 is used to adjust the confidence score of each source target domain adaptive subnetwork classifier according to the LMMD loss and the subnetwork's own adversarial loss to obtain an optimized classifier; An optimization subunit 342 is used to optimize the parameters of the source-target domain adaptive subnetwork according to the LMMD loss to obtain an optimized adaptive subnetwork; The updating subunit 343 updates the multi-adversarial multi-domain adaptive network according to the optimized classifier and the optimized adaptive subnet to obtain an optimized diagnosis network.

[0085] As an optional implementation, the automobile motor bearing fault diagnosis model building device further includes: A collection unit 360, used to collect real-time bearing operation data through sensors built into the vehicle to be diagnosed; An extraction unit 370 is used to extract features from real-time bearing operation data through a source-target domain adaptive subnet to obtain extracted features; A classification unit 380 is used to classify the extracted features through a source-target domain adaptive subnet classifier to obtain a classification result; The diagnosis unit 390 is used to perform collaborative decision processing on the classification results according to the confidence scores corresponding to the multi-subnet joint decision module and the source target domain adaptive subnet classifier to obtain a fault diagnosis result.

[0086] In this embodiment, the explanation of the automobile motor bearing fault diagnosis model construction device can refer to the description in Example 1 or Example 2, and will not be further elaborated in this embodiment.

[0087] It can be seen that the automobile motor bearing fault diagnosis model construction device described in this embodiment can construct a multi-adversarial multi-domain adaptive network, realize the effective use of unlabeled signals, reduce the cost and difficulty of data collection, and thus improve the accuracy and reliability of unlabeled cross-operating condition motor bearing fault diagnosis.

[0088] An embodiment of the present application provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the automobile motor bearing fault diagnosis model construction method in embodiment 1 or embodiment 2 of the present application.

[0089] An embodiment of the present application provides a computer-readable storage medium storing computer program instructions. When the computer program instructions are read and executed by a processor, the method for constructing a fault diagnosis model for an automobile motor bearing in Embodiment 1 or Embodiment 2 of the present application is executed.

[0090] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0091] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0092] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0093] The above description is only an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0094] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0095] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

Claims

1. A method for constructing a fault diagnosis model for an automobile motor bearing, characterized in that: include: Acquire data samples and a pre-built multi-adversarial multi-domain adaptive network; wherein the data samples include a labeled source domain data set and an unlabeled target domain data set; the multi-adversarial multi-domain adaptive network includes a source target domain adaptive subnet, a multi-source domain adversarial network, multiple source target domain adaptive subnet classifiers, and a multi-subnet joint decision module; Performing data preprocessing on the data samples to obtain preprocessed data; Input the preprocessed data into a multi-adversarial multi-domain adaptive network, and calculate the LMMD loss and the subnet self-adversarial loss of each of the source target domain adaptive subnet classifiers; Optimizing network parameters of the multi-adversarial multi-domain adaptive network according to the LMMD loss and the adversarial loss to obtain an optimized diagnostic network; When the training termination condition is reached, the final optimized diagnosis network is determined as the trained motor bearing fault diagnosis model.

2. The method for constructing a fault diagnosis model for an automobile motor bearing according to claim 1, characterized in that: The source domain data set includes sensor data collected during laboratory bearing operation, and the target domain data set includes data collected by vehicle built-in sensors; The source-target domain adaptation subnetwork includes four attention encoder layers and a fully connected network; The multi-source domain adversarial network includes three fully connected layers.

3. The method for constructing a fault diagnosis model for an automobile motor bearing according to claim 1, characterized in that: The performing data preprocessing on the data sample to obtain preprocessed data includes: Performing missing value processing on the data sample to obtain missing value processed data; The missing value processed data is subjected to data denoising processing based on an attention mechanism to obtain preprocessed data.

4. The method for constructing a fault diagnosis model for an automobile motor bearing according to claim 3, characterized in that: The performing missing value processing on the data sample to obtain missing value processed data includes: When the sample size of the data sample is greater than a preset sample size threshold, performing missing value deletion processing on the data sample to obtain missing value processed data; When the sample size of the data sample is not greater than a preset sample size threshold, mean interpolation processing is performed on the missing values ​​of the data sample to obtain missing value processed data.

5. The method for constructing a fault diagnosis model for an automobile motor bearing according to claim 1, characterized in that: The step of inputting the preprocessed data into the multi-adversarial multi-domain adaptive network and calculating the LMMD loss and the subnet self-adversarial loss of each of the source target domain adaptive subnet classifiers includes: The pre-processed data is input into the source-target domain adaptive subnet for processing to obtain a plurality of source domain feature information in a unified feature space; Calculating the LMMD loss according to the source domain feature information; wherein the LMMD loss includes a first LMMD loss of each of the source target domain adaptive subnet classifiers on the source domain and a second LMMD loss of each of the source target domain adaptive subnet classifiers on the target domain; According to the multi-source domain adversarial network and the source domain feature information, the subnet self-adversarial loss of each source target domain adaptive subnet classifier is calculated.

6. The method for constructing a fault diagnosis model for an automobile motor bearing according to claim 1, characterized in that: The step of optimizing the network parameters of the multi-adversarial multi-domain adaptive network according to the LMMD loss and the adversarial loss to obtain an optimized diagnostic network includes: According to the LMMD loss and the subnet's own adversarial loss, adjusting the confidence score of each of the source-target domain adaptive subnet classifiers to obtain an optimized classifier; Optimizing the parameters of the source-target domain adaptive subnet according to the LMMD loss to obtain an optimized adaptive subnet; The multi-adversarial multi-domain adaptive network is updated according to the optimized classifier and the optimized adaptive subnet to obtain an optimized diagnosis network.

7. The method for constructing a fault diagnosis model for an automobile motor bearing according to claim 1, characterized in that: The method further comprises: Collect real-time bearing operation data through sensors built into the vehicle to be diagnosed; Performing feature extraction on the real-time bearing operation data through the source-target domain adaptive subnet to obtain extracted features; Classifying the extracted features by the source-target domain adaptive subnet classifier to obtain a classification result; According to the confidence scores corresponding to the multi-subnet joint decision module and the source-target domain adaptive subnet classifier, the classification results are collaboratively decided to obtain a fault diagnosis result.

8. A device for constructing a fault diagnosis model for an automobile motor bearing, characterized in that: The automobile motor bearing fault diagnosis model building device comprises: A first acquisition unit is used to acquire data samples and a pre-built multi-adversarial multi-domain adaptive network; wherein the data samples include a labeled source domain data set and an unlabeled target domain data set; the multi-adversarial multi-domain adaptive network includes a source target domain adaptive subnet, a multi-source domain adversarial network, a plurality of source target domain adaptive subnet classifiers and a multi-subnet joint decision module; A preprocessing unit, used for performing data preprocessing on the data sample to obtain preprocessed data; A loss calculation unit, used for inputting the preprocessed data into a multi-adversarial multi-domain adaptive network, and calculating the LMMD loss and the subnet self-adversarial loss of each of the source target domain adaptive subnet classifiers; A parameter optimization unit, used for optimizing the network parameters of the multi-adversarial multi-domain adaptive network according to the LMMD loss and the adversarial loss to obtain an optimized diagnostic network; The determination unit is used to determine the final optimized diagnosis network as the trained motor bearing fault diagnosis model when the training termination condition is reached.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the automotive motor bearing fault diagnosis model construction method according to any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium stores computer program instructions, and when the computer program instructions are read and executed by a processor, the method for constructing a fault diagnosis model for an automobile motor bearing according to any one of claims 1 to 7 is executed.

Images