Motor current signal fault diagnosis method, device, equipment and storage medium

Abstract

The application discloses a motor current signal fault diagnosis method, device and equipment and a storage medium. The motor current signal fault diagnosis method comprises the following steps: acquiring a motor current signal; inputting the motor current signal into a preset signal classification model; performing fault identification processing on the motor current signal based on the signal classification model to obtain fault information; wherein the signal classification model comprises a feature extractor and a fault discriminator, and the feature extractor and the fault discriminator are obtained through training based on a semantic alignment grouping confrontation method. Through the feature extractor and the fault discriminator in the pre-trained signal classification model, the motor current signal is subjected to semantic alignment grouping confrontation, so that samples from different fields but having the same category label are mapped in the feature space in the vicinity, so that the feature extraction capability of the model is improved, and the accuracy of fault diagnosis is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault diagnosis, in particular to a motor current signal fault diagnosis method, device, equipment and storage medium. BACKGROUND

[0002] SCARA robots are important members in industrial production automation, with 20 functions such as high-precision accessory processing, carrying, assembly and sorting. SCARA robots are favored by automation enterprises due to their high speed and high precision processing advantages. However, the screw rod of the SCARA robot is prone to jamming, lack of steel balls and other faults under the condition of high speed and long time operation. Therefore, accurate fault diagnosis of the SCARA robot is a difficult problem to be solved at present.

[0003] The main method for engineering practice in the related art is the transfer learning method. The diagnosis method of transfer learning mainly includes feature distance measurement based method, domain adaptation based on adversarial learning and other methods. However, due to the existence of flexible link between the motor and the screw rod, the fault features contained in the motor current are extremely weak, making it difficult for the feature distance measurement and the adversarial domain adaptation method to extract fault features, resulting in low accuracy of fault diagnosis. SUMMARY

[0004] The main purpose of the present application is to provide a motor current signal fault diagnosis method, device, equipment and storage medium, aiming at solving the technical problem of low accuracy of fault diagnosis in the prior art.

[0005] To achieve the above purpose, the present application provides a motor current signal fault diagnosis method, which comprises:

[0006] obtaining a motor current signal;

[0007] inputting the motor current signal into a preset signal classification model, based on the signal classification model, performing fault recognition processing on the motor current signal to obtain fault information;

[0008] The signal classification model comprises a feature extractor and a fault discriminator, and the feature extractor and the fault discriminator are trained based on a semantic alignment grouping adversarial method.

[0009] Optionally, before the step of obtaining the motor current signal, the method comprises:

[0010] obtaining source domain current signal samples and fault result labels of the source domain current signal samples;

[0011] based on the source domain current signal samples, pre-training an initial feature extractor;

[0012] grouping the source domain current signal samples into a preset first number of semantic alignment groups to obtain grouped source domain current signal samples;

[0013] pre-training an initial fault discriminator based on the grouped source domain current signal samples;

[0014] iteratively training the initial feature extractor and the initial fault discriminator based on the source domain current signal samples and fault result labels of the source domain current signal samples to obtain a signal classification model meeting an accuracy condition.

[0015] Optionally, the step of iteratively training the initial feature extractor and the initial fault discriminator based on the source domain current signal samples and fault result labels of the source domain current signal samples to obtain a signal classification model meeting an accuracy condition comprises:

[0016] inputting the source domain current signal samples in the semantic alignment groups into the initial feature extractor, extracting first fault features of the source domain current signal samples based on the initial feature extractor, and determining predicted fault information based on the first fault features;

[0017] determining a fault classification loss based on the predicted fault information and the fault result labels;

[0018] inputting the first fault features into the initial fault discriminator to obtain a group adversarial loss;

[0019] determining a target loss based on the fault classification loss and the group adversarial loss;

[0020] determining whether the target loss meets a loss standard indicated by a preset loss threshold range;

[0021] If the target loss does not meet the loss standard indicated by the preset loss threshold range, returning to the step of inputting the source domain current signal samples in the semantic alignment groups into the initial feature extractor, extracting first fault features of the source domain current signal samples based on the initial feature extractor, and obtaining predicted fault information, until the target loss meets the loss standard indicated by the preset loss threshold range to stop training and obtain a signal classification model meeting an accuracy condition.

[0022] Optionally, the step of iteratively training the initial feature extractor and the initial fault discriminator based on the source domain current signal samples and fault result labels of the source domain current signal samples to obtain a signal classification model meeting an accuracy condition comprises:

[0023] obtaining target domain current signal samples;

[0024] Based on the source domain current signal sample, the fault result label of the source domain current signal sample and the target domain current signal sample, the initial feature extractor and the initial fault discriminator are iteratively trained in combination with a maximum mean discrepancy (MMD) distance measurement method to obtain a signal classification model that meets an accuracy condition.

[0025] Optionally, the step of obtaining the signal classification model that meets the accuracy condition based on the source domain current signal sample, the fault result label of the source domain current signal sample and the target domain current signal sample in combination with the maximum mean discrepancy (MMD) distance measurement method comprises:

[0026] The source domain current signal sample in the semantic alignment group is input to the initial feature extractor, first fault features of the source domain current signal sample are extracted based on the initial feature extractor, and predicted fault information is determined based on the first fault features.

[0027] A fault classification loss is determined based on the predicted fault information and the fault result label.

[0028] The first fault features are input to the initial fault discriminator to obtain a group adversarial loss.

[0029] The target domain current signal sample is input to the initial feature extractor to obtain second fault features of the target domain.

[0030] A maximum mean discrepancy (MMD) distance measurement function is used to calculate a distance loss of the first fault features and the second fault features.

[0031] A target loss is obtained by calculating a sum of the fault classification loss, the group adversarial loss and the distance loss.

[0032] It is determined whether the target loss meets a loss standard indicated by a preset loss threshold range.

[0033] If the target loss does not meet the loss standard indicated by the preset loss threshold range, the step of inputting the source domain current signal sample in the semantic alignment group to the initial feature extractor, extracting first fault features of the source domain current signal sample based on the initial feature extractor and obtaining predicted fault information is returned until the target loss meets the loss standard indicated by the preset loss threshold range, and the training is stopped to obtain the signal classification model that meets the accuracy condition.

[0034] Optionally, the step of inputting the target domain current signal sample to the initial feature extractor to obtain second fault features of the target domain comprises:

[0035] performing fast Fourier transform on the target domain current signal sample to obtain a transformed target domain current signal sample;

[0036] normalizing the transformed target domain current signal sample to obtain a normalized target domain current signal sample;

[0037] inputting the normalized target domain current signal sample into the initial feature extractor to obtain a second fault feature of the target domain.

[0038] Optionally, the signal classification model adopts a lightweight convolutional network, wherein the lightweight convolutional network is constructed by a depthwise convolution, a pointwise convolution and a group convolution.

[0039] The application further provides a motor current signal fault diagnosis device, characterized in that the motor current signal fault diagnosis device comprises:

[0040] an acquisition module configured to acquire a motor current signal;

[0041] a recognition module configured to input the motor current signal into a preset signal classification model, perform fault recognition processing on the motor current signal based on the signal classification model, and obtain fault information; wherein the signal classification model comprises a feature extractor and a fault discriminator, and the feature extractor and the fault discriminator are trained based on a semantic alignment grouping adversarial method.

[0042] The application further provides a motor current signal fault diagnosis device, which comprises a memory, a processor and a program stored in the memory and used for implementing a motor current signal fault diagnosis method,

[0043] The memory is configured to store the program for implementing the motor current signal fault diagnosis method.

[0044] The processor is configured to execute the program for implementing the motor current signal fault diagnosis method to implement the steps of the motor current signal fault diagnosis method.

[0045] The application further provides a storage medium, which stores a program for implementing a motor current signal fault diagnosis method, and the program for implementing the motor current signal fault diagnosis method is executed by a processor to implement the steps of the motor current signal fault diagnosis method.

[0046] This application provides a method, apparatus, device, and storage medium for fault diagnosis of motor current signals. Compared to related technologies where the motor current contains very weak fault features due to the flexible link between the motor and the lead screw, making it difficult to extract fault features using feature distance measurement and adversarial adaptive methods, resulting in low accuracy in fault diagnosis, this application acquires the motor current signal; inputs the motor current signal into a preset signal classification model; and performs fault identification processing on the motor current signal based on the signal classification model to obtain fault information. The signal classification model includes a feature extractor and a fault discriminator, which are trained using a semantic alignment grouping adversarial method. That is, in this application, the feature extractor and fault discriminator in the pre-trained signal classification model perform semantic alignment grouping adversarial processing on the motor current signal, ensuring that samples from different domains but with the same category label are mapped close together in the feature space, thereby improving the model's feature extraction capability and thus improving the accuracy of fault diagnosis. Attached Figure Description

[0047] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0048] Figure 1 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of this application;

[0049] Figure 2 This is a flowchart illustrating the first embodiment of the fault diagnosis method for motor current signals according to this application.

[0050] Figure 3 This is a schematic diagram of the module of the fault diagnosis device for motor current signal of this application;

[0051] Figure 4 This is a schematic diagram of the semantic alignment grouping adversarial setting in the first embodiment of the fault diagnosis method for motor current signals of this application;

[0052] Figure 5 This is a schematic diagram of the signal classification model of the first embodiment of the fault diagnosis method for motor current signals in this application;

[0053] Figure 6 This is a schematic diagram of the diagnostic process of the second embodiment of the fault diagnosis method for motor current signal of this application;

[0054] Figure 7 FIG. 3 is a flowchart of a third embodiment of a motor current signal fault diagnosis method according to the present application.

[0055] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0056] It should be understood that the specific embodiments described herein merely exemplify the application and do not limit the application.

[0057] As shown in FIG. 1, a terminal structure schematic diagram of a hardware running environment involved in an embodiment of the present application is shown. Figure 1 Figure 1 The terminal of the embodiment of the present application can be a PC, or a smart phone, a tablet computer, an e-book reader, an MP3 (Moving Picture Experts Group Audio Layer III) player, an MP4 (Moving Picture Experts Group Audio Layer IV) player, a portable computer, or other movable terminal devices with display functions.

[0058] As shown in FIG. 2, the terminal can include a processor 1001, for example, a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002. The communication bus 1002 is used to realize the connection communication between the components. The user interface 1003 can include a display screen (Display) and an input unit such as a keyboard (Keyboard). Optionally, the user interface 1003 can further include a standard wired interface and a wireless interface. The network interface 1004 can optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 can be a high-speed RAM memory, or a stable memory (non-volatile memory) such as a magnetic disk memory. Optionally, the memory 1005 can also be a storage device independent of the aforementioned processor 1001.

[0059] As shown in FIG. 2, the terminal can include a processor 1001, for example, a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002. The communication bus 1002 is used to realize the connection communication between the components. The user interface 1003 can include a display screen (Display) and an input unit such as a keyboard (Keyboard). Optionally, the user interface 1003 can further include a standard wired interface and a wireless interface. The network interface 1004 can optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 can be a high-speed RAM memory, or a stable memory (non-volatile memory) such as a magnetic disk memory. Optionally, the memory 1005 can also be a storage device independent of the aforementioned processor 1001. Figure 1

[0060] ​​Optionally, the terminal can further include a camera, RF (Radio Frequency) circuit, sensors, audio circuit, WiFi module, and the like. The sensors can include, for example, a light sensor, a motion sensor, and other sensors. Specifically, the light sensor can include an ambient light sensor and a proximity sensor, wherein the ambient light sensor can adjust the brightness of the display screen according to the brightness of ambient light, and the proximity sensor can turn off the display screen and / or backlight when the mobile terminal is moved to the ear. As one of the motion sensors, the gravity acceleration sensor can detect the size of acceleration in each direction (generally three axes), and when at rest, it can detect the size and direction of gravity, which can be used for identifying mobile terminal posture applications (such as landscape / portrait screen switching, related games, magnetometer posture calibration), vibration recognition related functions (such as pedometer, tapping), and the like. Of course, the mobile terminal can also be configured with a gyroscope, barometer, hygrometer, thermometer, infrared sensor, and other sensors, which will not be described here.

[0061] Those skilled in the art can understand that Figure 1 The terminal structure shown in the above embodiments does not constitute a limitation on the terminal, and can include more or fewer components than shown, or combine certain components, or different component arrangements.

[0062] As shown in Figure 1 The memory 1005 as a computer storage medium can include an operation device, a network communication module, a user interface module, and a motor current signal fault diagnosis program.

[0063] In the terminal shown in Figure 1 The network interface 1004 is mainly used for connecting to a background server and communicating data with the background server; the user interface 1003 is mainly used for connecting to a client (user end) and communicating data with the client; and the processor 1001 can be used to call the motor current signal fault diagnosis program stored in the memory 1005.

[0064] Referring to Figure 2 , the embodiment of the present application provides a motor current signal fault diagnosis method, which comprises:

[0065] Step S100, acquiring a motor current signal;

[0066] Step S200, inputting the motor current signal into a preset signal classification model, performing fault identification processing on the motor current signal based on the signal classification model, and obtaining fault information;

[0067] The signal classification model includes a feature extractor and a fault discriminator, and the feature extractor and the fault discriminator are trained based on a semantic alignment grouping adversarial method.

[0068] In this embodiment, the application scenario is:

[0069] As an example, the fault diagnosis scenario of the motor current signal can be that the screw rod of the SCARA robot is prone to faults such as jamming and lack of steel balls under the condition of high speed and long time operation, and it is necessary to accurately identify the fault category. The main method used in engineering practice in the related art is the transfer learning method. The main methods of the transfer learning diagnosis method include feature distance measurement and domain adaptation based on adversarial learning. However, due to the flexible link between the motor and the screw rod, the fault features contained in the motor current are extremely weak, making it difficult for the feature distance measurement and the adversarial domain adaptation method to extract fault features, resulting in low accuracy of fault diagnosis. For this scenario, the motor current signal fault diagnosis method of the embodiment performs semantic alignment grouping and adversarial training on the motor current signal through the feature extractor and the fault discriminator in the pre-trained signal classification model, so that samples from different domains but with the same category label are mapped in the feature space. Close to, in this way, the feature extraction capability of the model is improved, and the accuracy of fault diagnosis is improved.

[0070] As an example, the application scenario of the motor current signal fault diagnosis not only includes the fault diagnosis of the screw rod of the SCARA robot as described above, but also includes various motor fault diagnosis scenarios, which are not specifically limited here.

[0071] The embodiment aims to improve the accuracy of fault diagnosis.

[0072] In this embodiment, the motor current signal fault diagnosis method is applied to a motor current signal fault diagnosis device.

[0073] The specific steps are as follows:

[0074] Step S100, acquiring a motor current signal;

[0075] In this embodiment, the motor current signal is the current signal output by the motor of the SCARA robot. Since existing diagnosis methods all rely on vibration signals as diagnosis data, but adding vibration sensors to each SCARA robot in engineering applications will increase huge economic costs, therefore, the motor current signal is selected as the diagnosis data in this application, which is convenient to implement and saves hardware costs.

[0076] In this embodiment, the device can acquire the motor current signal in real time by receiving the motor current signal sent by the motor, or the device can acquire the motor current signal by receiving the motor current signal transmitted by the motor at regular intervals, and determining whether the motor current signal contains fault information.

[0077] Step S200, input the motor current signal to a preset signal classification model, perform fault identification processing on the motor current signal based on the signal classification model, and obtain fault information;

[0078] The signal classification model comprises a feature extractor and a fault discriminator, and the feature extractor and the fault discriminator are trained based on a semantic alignment grouping adversarial method.

[0079] In the embodiment, the signal classification model comprises a feature extractor and a fault discriminator, and the feature extractor and the fault discriminator are trained based on a semantic alignment grouping adversarial method, wherein the semantic alignment grouping adversarial method is used to ensure that samples from different domains but with the same class label are mapped in the feature space, find the domain-invariant characteristics of the feature samples in different domains, and realize semantic alignment between different labels. Figure 4 Referring to FIG. 2, a certain number of semantic alignment groups are set, and the current data of different machines are included in each group. 0, 1 and 2 represent different types of state labels, for example, fault labels such as jamming and missing steel balls, and a normal state label. Adversarial relationship between G1 and G2, G3 and G4 can not only find the domain-invariant characteristics of different domains, but also realize semantic alignment between different labels. The same applies to G5 and G6, G7 and G8. The number of groups can be set as needed.

[0080] Before the step S100 of obtaining the motor current signal, the method comprises the following steps A100-A400:

[0081] Step A100, obtaining source domain current signal samples and fault result labels of the source domain current signal samples;

[0082] In the embodiment, the source domain represents a different domain from the test sample, but has rich supervised information; the target domain represents the domain where the test sample is located, and has no label or only a small amount of label. The source domain and the target domain often belong to the same type of task, but have different distributions. The source domain current signal samples are obtained based on laboratory current signal samples, and have corresponding labels, i.e., fault result labels of the source domain current signal samples.

[0083] Step A200, pre-training an initial feature extractor based on the source domain current signal samples;

[0084] In the embodiment, the device pre-trains an initial feature extractor based on the source domain current signal samples, wherein the initial feature extractor is a model having the function of extracting current signal samples, and continuously converges in iterative training until a feature extractor meeting the accuracy condition is obtained.

[0085] Step A300, grouping the source domain current signal samples into a preset first number of semantic alignment groups to obtain grouped source domain current signal samples.

[0086] In the embodiment, the device groups the source domain current signal samples into a preset first number of semantic alignment groups to obtain grouped source domain current signal samples. Specifically, the device randomly samples from the source domain current signal samples and groups them into a preset first number of semantic alignment groups. For example, the device randomly samples from the source domain data set and divides them into G1 to G8 groups.

[0087] Step A400, pre-training an initial fault discriminator based on the grouped source domain current signal samples.

[0088] In the embodiment, the device pre-trains an initial fault discriminator based on the grouped source domain current signal samples, wherein the initial fault discriminator is a model having the function of judging the fault of current signal samples, and continuously converges in iterative training until a fault discriminator meeting the accuracy condition is obtained, wherein the fault discriminator is used to assist in training the feature extractor.

[0089] Step A500, iteratively training the initial feature extractor and the initial fault discriminator based on the source domain current signal samples and the fault result labels of the source domain current signal samples to obtain a signal classification model meeting the accuracy condition.

[0090] In the embodiment, the device iteratively trains the initial feature extractor and the initial fault discriminator based on the source domain current signal samples and the fault result labels of the source domain current signal samples to obtain a signal classification model meeting the accuracy condition, wherein the reference Figure 5In the present application, lightweight convolution networks are used to build the model, including depthwise convolution (DW), pointwise convolution (PW) and group convolution (GC). The feature fusion block is composed of PW, DW and average pooling layer, and the average pooling layer is used to reuse the upper layer features to improve the generalization performance of the model. Channel splicing can avoid using convolution kernels to increase the number of channels, so as to further reduce the model parameter amount. Channel shuffling is to fuse the information of the two branches of the feature fusion block to improve the feature extraction capability of the model. The present application designs a lightweight diagnostic model based on DW, PW and GC, which creates conditions for deploying the model on mobile terminals.

[0091] Specifically, the step A500 includes the following steps A510-A560:

[0092] Step A510, inputting the source domain current signal sample in the semantic alignment group into the initial feature extractor, extracting the first fault feature of the source domain current signal sample based on the initial feature extractor, and determining the predicted fault information based on the first fault feature.

[0093] In the present embodiment, the device inputs the source domain current signal sample in the semantic alignment group into the initial feature extractor, extracts the first fault feature of the source domain current signal sample based on the initial feature extractor, and determines the predicted fault information based on the first fault feature.

[0094] Specifically, the step A510 includes the following steps A511-A513:

[0095] Step A511, performing feature extraction on the source domain current signal sample based on the initial feature extractor to obtain the third fault feature.

[0096] In the present embodiment, the device performs feature extraction on the source domain current signal sample based on the initial feature extractor to obtain the third fault feature, wherein the feature extraction method includes but is not limited to HOG feature extraction algorithm, LBP feature extraction algorithm and Haar feature extraction algorithm.

[0097] Step A512, performing convolution and average pooling on the third fault feature to obtain the fourth fault feature.

[0098] Step A513, fusing feature information by channel splicing and channel shuffling on the fourth fault feature to obtain the first fault feature of the source domain current signal sample.

[0099] In the present embodiment, with reference to Figure 5The device uses a lightweight diagnostic model based on DW, PW and GC to perform convolution and average pooling on the third fault feature to obtain the fourth fault feature. The fourth fault feature is then processed by channel splicing and channel shuffling to fuse the feature information and obtain the first fault feature of the source domain current signal sample.

[0100] Step A520: Based on the predicted fault information and the fault result label, determine the fault classification loss;

[0101] In this embodiment, the device determines the fault classification loss based on the predicted fault information and the fault result label. Specifically, the device calculates the fault classification loss using mean square error based on the predicted fault information and the fault result label.

[0102] Step A530: Input the first fault feature into the initial fault discriminator to obtain the group adversarial loss;

[0103] In this embodiment, the device inputs the first fault feature to the initial fault discriminator to obtain the group adversarial loss.

[0104] Step A540: Based on the fault classification loss and the group adversarial loss, determine the target loss;

[0105] In this embodiment, the device determines the target loss based on the fault classification loss and the group confrontation loss. Specifically, the device adds the fault classification loss and the group confrontation loss together as the target loss.

[0106] Step A550: Determine whether the target loss meets the loss standard indicated by the preset loss threshold range;

[0107] In this embodiment, the device determines whether the target loss meets the loss standard indicated by a preset loss threshold range. The loss standard indicated by the loss threshold range is set by the device itself and is used to determine whether the target loss of the current training cycle has converged. If the target loss meets the loss standard indicated by the preset loss threshold range, it means that the target loss of the current training cycle has converged, the accuracy requirement has been met, and the training of the signal classification model has been completed. If the target loss does not meet the loss standard indicated by the preset loss threshold range, it means that the target loss of the current training cycle has not converged, the accuracy requirement has not been met, and iterative training for the next cycle is required.

[0108] If the target loss does not satisfy the loss standard indicated by the preset loss threshold range, the step of inputting the source domain current signal sample in the semantic alignment group to the initial feature extractor, extracting the first fault feature of the source domain current signal sample based on the initial feature extractor, and obtaining the predicted fault information is returned until the target loss satisfies the loss standard indicated by the preset loss threshold range, and the training is stopped to obtain the signal classification model with the accuracy condition satisfied.

[0109] In the embodiment, if the target loss does not satisfy the loss standard indicated by the preset loss threshold range, it indicates that the target loss of the current training period does not converge, the accuracy requirement has not been met, and the next period of iterative training is needed. Then, the device returns to the step of inputting the source domain current signal sample in the semantic alignment group to the initial feature extractor, extracting the first fault feature of the source domain current signal sample based on the initial feature extractor, and obtaining the predicted fault information until the target loss satisfies the loss standard indicated by the preset loss threshold range, and the training is stopped to obtain the signal classification model with the accuracy condition satisfied.

[0110] The motor current signal fault diagnosis method provided in the application has the advantages that: in the related art, due to the existence of a flexible link between the motor and the screw rod, the fault features contained in the motor current are extremely weak, so that the feature distance metric and the method of domain self-adaptation are difficult to extract fault features, resulting in low accuracy of fault diagnosis. In the application, the motor current signal is obtained; the motor current signal is input to a preset signal classification model, and the motor current signal is subjected to fault recognition processing based on the signal classification model to obtain fault information; wherein the signal classification model includes a feature extractor and a fault discriminator, and the feature extractor and the fault discriminator are trained based on a semantic alignment grouping confrontation method. That is, in the application, the feature extractor and the fault discriminator in the pre-trained signal classification model are used for semantic alignment grouping confrontation of the motor current signal, so that samples from different domains but with the same category label are mapped in the feature space. In this way, the feature extraction capability of the model is improved, and the accuracy of fault diagnosis is improved.

[0111] Based on the above first embodiment, the application further provides another embodiment, and the motor current signal fault diagnosis method includes:

[0112] Referring to Figure 6 , the step A500 of performing iterative training on the initial feature extractor and the initial fault discriminator based on the source domain current signal sample and the fault result label of the source domain current signal sample to obtain the signal classification model with the accuracy condition satisfied further includes the following steps B100-B200:

[0113] Step B100, obtain a target domain current signal sample;

[0114] In this embodiment, the target domain current signal sample is an industrial field obtained current signal sample, and there is no corresponding label, that is, an unknown label current signal sample.

[0115] Step B200, based on the source domain current signal sample, the fault result label of the source domain current signal sample, and the target domain current signal sample, the initial feature extractor and the initial fault discriminator are iteratively trained by combining a maximum mean difference MMD distance measurement method, to obtain a signal classification model with a precision condition.

[0116] In this embodiment, the device iteratively trains the initial feature extractor and the initial fault discriminator based on the source domain current signal sample, the fault result label of the source domain current signal sample, and the target domain current signal sample, and combines a maximum mean difference MMD distance measurement method, to obtain a signal classification model with a precision condition, that is, a signal classification model based on semantic alignment grouping confrontation and maximum mean difference MMD feature distance combination method, to improve the feature extraction capability of the model, and further improve the generalization performance of the signal classification model.

[0117] Specifically, the step B200 includes the following steps B210-B280:

[0118] Step B210, input the source domain current signal sample in the semantic alignment group to the initial feature extractor, extract the first fault feature of the source domain current signal sample based on the initial feature extractor, and determine the predicted fault information based on the first fault feature;

[0119] Step B220, determine the fault classification loss based on the predicted fault information and the fault result label;

[0120] Step B230, input the first fault feature to the initial fault discriminator to obtain a group confrontation loss;

[0121] In this embodiment, the steps B210-B230 are the same as the steps A510-A530 described above, and will not be described here.

[0122] Step B240, input the target domain current signal sample to the initial feature extractor to obtain a second fault feature of the target domain;

[0123] In this embodiment, the device inputs the target domain current signal sample to the initial feature extractor to extract the feature of the target domain current signal sample, to obtain a second fault feature of the target domain.

[0124] Specifically, the step B240 comprises steps B241-B243.

[0125] The step B241 performs fast Fourier transform on the target domain current signal sample to obtain a transformed target domain current signal sample.

[0126] The step B242 normalizes the transformed target domain current signal sample to obtain a normalized target domain current signal sample.

[0127] The step B243 inputs the normalized target domain current signal sample into the initial feature extractor to obtain a second fault feature of the target domain.

[0128] In this embodiment, the device collects screw current data from an industrial site and performs fast Fourier transform, then normalizes the data, and finally inputs the data into the feature extractor to obtain the fault feature of the target domain. Fast Fourier transform (fast Fourier transform) is a general term for efficient and fast calculation methods for calculating discrete Fourier transform (DFT) using a computer, and is abbreviated as FFT.

[0129] The step B250 calculates the distance loss of the first fault feature and the second fault feature by using a maximum mean discrepancy (MMD) distance measurement function.

[0130] In this embodiment, the device calculates the distance loss of the first fault feature and the second fault feature by using a maximum mean discrepancy (MMD) distance measurement function. The maximum mean discrepancy (MMD) distance measurement function is used to measure the difference between two distributions. The maximum distance between the two distributions should be used as the standard for measuring the two distributions to calculate the distance loss of the first fault feature and the second fault feature.

[0131] The step B260 calculates the sum of the fault classification loss, the group adversarial loss, and the distance loss to obtain a target loss.

[0132] In this embodiment, the device calculates the sum of the fault classification loss, the group adversarial loss, and the distance loss to obtain a target loss, i.e., the loss is accumulated and used to optimize the feature extractor.

[0133] The step B270 judges whether the target loss meets a loss standard indicated by a preset loss threshold range.

[0134] Step B280, if the target loss does not meet the loss standard indicated by the preset loss threshold range, return to the step of inputting the source domain current signal sample in the semantic alignment group into the initial feature extractor, extracting the first fault feature of the source domain current signal sample based on the initial feature extractor to obtain the predicted fault information, until the target loss meets the loss standard indicated by the preset loss threshold range to stop training, and obtain a signal classification model with a precision condition.

[0135] In the embodiment, steps B270 and B280 refer to steps A550-A560 above, which will not be repeated here.

[0136] In the embodiment, the application improves the feature extraction capability of the model by designing a method of combining multi-source domain grouping confrontation and feature distance measurement, which can solve the problem that the fault feature of the current signal is weak. The fault diagnosis model designed by the application can complete real-time diagnosis of the screw end to end, which can reduce the output of unqualified products and improve economic efficiency.

[0137] Based on the first and second embodiments described above, the application also provides another embodiment, which refers to Figure 7 , the motor current signal fault diagnosis method comprises:

[0138] Step 1, load the source domain data set collected in the laboratory.

[0139] Step 2, use the source domain data set to pre-train the feature extractor.

[0140] Step 3, randomly sample from the source domain data set and divide it into G1 to G8 groups.

[0141] Step 4, use G1 to G8 group data to train the discriminator.

[0142] Step 5, input G1 to G8 group data into the feature extractor and extract the fault features of the source domain, and then calculate the fault classification loss of the source domain according to the true label.

[0143] Step 6, input the fault features extracted in step 4 into the discriminator and calculate the group confrontation loss.

[0144] Step 7, collect the screw current data from the industrial site and perform fast Fourier transform, then normalize the data, and finally input the data into the feature extractor to obtain the fault features of the target domain.

[0145] Step 8, use the MMD distance measurement function to calculate the distance loss between the fault features of steps 4 and 6.

[0146] Step 9, add the losses calculated in steps 4, 5 and 7 and use them to optimize the feature extractor.

[0147] Step 10, collect data from the industrial site again and input the feature extractor to obtain a fault prediction result.

[0148] Step 11, end.

[0149] In this embodiment, the motor current signal is first selected as the diagnostic data to save economic cost, then a lightweight diagnostic network is designed to facilitate direct deployment of the model on a mobile terminal, and finally a method combining multi-source domain grouping adversarial and maximum mean discrepancy (MMD) feature distance is designed to improve the feature extraction capability of the model and further improve the generalization performance of the model.

[0150] The application also provides a motor current signal fault diagnosis device, referring to Figure 3 , the motor current signal fault diagnosis device comprises:

[0151] The acquisition module 10 is configured to acquire a motor current signal.

[0152] The identification module 20 is configured to input the motor current signal into a preset signal classification model, perform fault identification processing on the motor current signal based on the signal classification model, and obtain fault information; wherein the signal classification model comprises a feature extractor and a fault discriminator, and the feature extractor and the fault discriminator are trained based on a semantic alignment grouping adversarial method.

[0153] Optionally, the motor current signal fault diagnosis device further comprises:

[0154] The sample acquisition module is configured to acquire source domain current signal samples and fault result labels of the source domain current signal samples.

[0155] The first pre-training module is configured to pre-train an initial feature extractor based on the source domain current signal samples.

[0156] The grouping module is configured to group the source domain current signal samples into a preset first number of semantic alignment groups to obtain grouped source domain current signal samples.

[0157] The second pre-training module is configured to pre-train an initial fault discriminator based on the grouped source domain current signal samples.

[0158] The training module is configured to iteratively train the initial feature extractor and the initial fault discriminator based on the source domain current signal samples and the fault result labels of the source domain current signal samples to obtain a signal classification model that meets the accuracy condition.

[0159] Optionally, the training module comprises:

[0160] a feature extraction module, configured to input source domain current signal samples in the semantic alignment group to the initial feature extractor, extract first fault features of the source domain current signal samples based on the initial feature extractor, and determine predicted fault information based on the first fault features;

[0161] a classification loss determination module, configured to determine a fault classification loss based on the predicted fault information and the fault result label;

[0162] a group adversarial loss determination module, configured to input the first fault features to the initial fault discriminator to obtain a group adversarial loss;

[0163] a target loss determination module, configured to determine a target loss based on the fault classification loss and the group adversarial loss;

[0164] a judgment module, configured to judge whether the target loss meets a loss criterion indicated by a preset loss threshold range;

[0165] an iterative training module, configured to return to the step of inputting the source domain current signal samples in the semantic alignment group to the initial feature extractor, extracting first fault features of the source domain current signal samples based on the initial feature extractor, and obtaining predicted fault information, until the target loss meets the loss criterion indicated by the preset loss threshold range, and stop training to obtain a signal classification model meeting an accuracy condition, if the target loss does not meet the loss criterion indicated by the preset loss threshold range.

[0166] Optionally, the training module further includes:

[0167] a target domain signal sample acquisition module, configured to acquire target domain current signal samples;

[0168] a training module combined with a distance measurement method, configured to perform iterative training on the initial feature extractor and the initial fault discriminator based on the source domain current signal samples, fault result labels of the source domain current signal samples, and the target domain current signal samples, in combination with a maximum mean difference (MMD) distance measurement method, to obtain a signal classification model meeting an accuracy condition.

[0169] Optionally, the training module combined with a distance measurement method includes:

[0170] a feature extraction module, configured to input source domain current signal samples in the semantic alignment group to the initial feature extractor, extract first fault features of the source domain current signal samples based on the initial feature extractor, and determine predicted fault information based on the first fault features;

[0171] a classification loss determination module, configured to determine a fault classification loss based on the predicted fault information and the fault result label;

[0172] a group adversarial loss determination module, configured to input the first fault feature into the initial fault discriminator to obtain a group adversarial loss;

[0173] a fault feature extraction module, configured to input the target domain current signal sample into the initial feature extractor to obtain a second fault feature of the target domain;

[0174] a distance metric calculation module, configured to calculate a distance loss of the first fault feature and the second fault feature by using a maximum mean discrepancy (MMD) distance metric function;

[0175] a loss calculation module, configured to calculate a sum of the fault classification loss, the group adversarial loss and the distance loss to obtain a target loss;

[0176] a judgment module, configured to judge whether the target loss meets a loss criterion indicated by a preset loss threshold range;

[0177] an iterative training module, configured to return to the step of inputting the source domain current signal sample in the semantic alignment group into the initial feature extractor to extract a first fault feature of the source domain current signal sample based on the initial feature extractor to obtain predicted fault information until the target loss meets the loss criterion indicated by the preset loss threshold range to stop training and obtain a signal classification model meeting an accuracy condition, if the target loss does not meet the loss criterion indicated by the preset loss threshold range.

[0178] Optionally, the fault feature extraction module comprises:

[0179] a Fourier transform module, configured to perform fast Fourier transform on the target domain current signal sample to obtain a transformed target domain current signal sample;

[0180] a normalization module, configured to normalize the transformed target domain current signal sample to obtain a normalized target domain current signal sample;

[0181] a target domain fault feature determination module, configured to input the normalized target domain current signal sample into the initial feature extractor to obtain a second fault feature of the target domain.

[0182] The motor current signal fault diagnosis device specific implementation manner of the present application is basically the same as the above-mentioned motor current signal fault diagnosis method each embodiment, and will not be repeated here.

[0183] Reference Figure 1 ,Figure 1 is a terminal structure diagram of a hardware running environment involved in an embodiment scheme of the present application.

[0184] As shown in Figure 1 , the terminal can include a processor 1001, for example, a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002. The communication bus 1002 is used to realize the connection communication between the components. The user interface 1003 can include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 can further include a standard wired interface, a wireless interface. The network interface 1004 can optionally include a standard wired interface, a wireless interface (such as a WI-FI interface). The memory 1005 can be a high-speed RAM memory, or a stable memory (non-volatile memory) such as a magnetic disk memory. The memory 1005 can also be an independent storage device from the aforementioned processor 1001.

[0185] Optionally, the motor current signal fault diagnosis device can further include a rectangular user interface, a network interface, a camera, an RF (Radio Frequency, Radio Frequency) circuit, a sensor, an audio circuit, a WiFi module, etc. The rectangular user interface can include a display screen (Display), an input sub-module such as a keyboard (Keyboard), and the optional rectangular user interface can further include a standard wired interface, a wireless interface. The network interface can optionally include a standard wired interface, a wireless interface (such as a WI-FI interface).

[0186] Those skilled in the art can understand that Figure 1 the structure of the motor current signal fault diagnosis device shown in the figure does not constitute a limitation on the motor current signal fault diagnosis device, and can include more or fewer components than the figure, or combine certain components, or different component arrangements.

[0187] As shown in Figure 1 , the memory 1005 as a storage medium can include an operating system, a network communication module, and a motor current signal fault diagnosis program. The operating system is a program that manages and controls the hardware and software resources of the motor current signal fault diagnosis device, supports the running of the motor current signal fault diagnosis program and other software and / or programs. The network communication module is used to realize the communication between the components in the memory 1005, and the communication between other hardware and software in the motor current signal fault diagnosis system.

[0188] In Figure 1The processor 1001 is configured to execute a motor current signal fault diagnosis program stored in the memory 1005 to implement the steps of the motor current signal fault diagnosis method.

[0189] The motor current signal fault diagnosis device embodiment of the present application is basically the same as the above-mentioned motor current signal fault diagnosis method embodiments, and will not be repeated here.

[0190] The present application also provides a storage medium, the storage medium stores a motor current signal fault diagnosis method program, the motor current signal fault diagnosis method program is executed by a processor to implement the motor current signal fault diagnosis method as follows:

[0191] Obtaining a motor current signal;

[0192] Inputting the motor current signal into a preset signal classification model, based on the signal classification model, performing fault recognition processing on the motor current signal to obtain fault information;

[0193] The signal classification model includes a feature extractor and a fault discriminator, and the feature extractor and the fault discriminator are trained based on a semantic alignment grouping confrontation method.

[0194] Optionally, before the step of obtaining a motor current signal, the method comprises:

[0195] Obtaining source domain current signal samples and fault result labels of the source domain current signal samples;

[0196] Based on the source domain current signal samples, an initial feature extractor is pre-trained;

[0197] Grouping the source domain current signal samples into a preset first number of semantic alignment groups to obtain grouped source domain current signal samples;

[0198] Based on the grouped source domain current signal samples, an initial fault discriminator is pre-trained;

[0199] Based on the source domain current signal samples and the fault result labels of the source domain current signal samples, the initial feature extractor and the initial fault discriminator are iteratively trained to obtain a signal classification model that meets the accuracy condition.

[0200] Optionally, the step of iteratively training the initial feature extractor and the initial fault discriminator based on the source domain current signal samples and the fault result labels of the source domain current signal samples to obtain a signal classification model that meets the accuracy condition comprises:

[0201] inputting the source domain current signal samples in the semantic alignment group into the initial feature extractor, extracting first fault features of the source domain current signal samples based on the initial feature extractor, and determining predicted fault information based on the first fault features;

[0202] determining a fault classification loss based on the predicted fault information and the fault result label;

[0203] inputting the first fault features into the initial fault discriminator to obtain a group adversarial loss;

[0204] determining a target loss based on the fault classification loss and the group adversarial loss;

[0205] determining whether the target loss meets a loss criterion indicated by a preset loss threshold range;

[0206] If the target loss does not meet the loss criterion indicated by the preset loss threshold range, returning to the step of inputting the source domain current signal samples in the semantic alignment group into the initial feature extractor, extracting first fault features of the source domain current signal samples based on the initial feature extractor to obtain predicted fault information, until the target loss meets the loss criterion indicated by the preset loss threshold range, and stopping training to obtain a signal classification model meeting an accuracy condition.

[0207] Optionally, the step of iteratively training the initial feature extractor and the initial fault discriminator based on the source domain current signal samples and the fault result labels of the source domain current signal samples to obtain a signal classification model meeting an accuracy condition comprises:

[0208] obtaining target domain current signal samples;

[0209] iteratively training the initial feature extractor and the initial fault discriminator based on the source domain current signal samples, the fault result labels of the source domain current signal samples, and the target domain current signal samples in combination with a maximum mean discrepancy (MMD) distance measurement method to obtain a signal classification model meeting an accuracy condition.

[0210] Optionally, the step of iteratively training the initial feature extractor and the initial fault discriminator based on the source domain current signal samples, the fault result labels of the source domain current signal samples, and the target domain current signal samples in combination with a maximum mean discrepancy (MMD) distance measurement method to obtain a signal classification model meeting an accuracy condition comprises:

[0211] inputting the source domain current signal samples in the semantic alignment group into the initial feature extractor, extracting first fault features of the source domain current signal samples based on the initial feature extractor, and determining predicted fault information based on the first fault features;

[0212] determining a fault classification loss based on the predicted fault information and the fault result label;

[0213] inputting the first fault features into the initial fault discriminator to obtain a group adversarial loss;

[0214] inputting the target domain current signal samples into the initial feature extractor to obtain second fault features of the target domain;

[0215] calculating a distance loss of the first fault features and the second fault features by using a maximum mean difference (MMD) distance measurement function;

[0216] calculating a sum of the fault classification loss, the group adversarial loss, and the distance loss to obtain a target loss;

[0217] determining whether the target loss meets a loss criterion indicated by a preset loss threshold range;

[0218] If the target loss does not meet the loss criterion indicated by the preset loss threshold range, returning to the step of inputting the source domain current signal samples in the semantic alignment group into the initial feature extractor, extracting first fault features of the source domain current signal samples based on the initial feature extractor to obtain predicted fault information, until the target loss meets the loss criterion indicated by the preset loss threshold range, and stopping training to obtain a signal classification model meeting an accuracy condition.

[0219] Optionally, the step of inputting the target domain current signal samples into the initial feature extractor to obtain second fault features of the target domain includes:

[0220] performing fast Fourier transform on the target domain current signal samples to obtain transformed target domain current signal samples;

[0221] normalizing the transformed target domain current signal samples to obtain normalized target domain current signal samples;

[0222] inputting the normalized target domain current signal samples into the initial feature extractor to obtain second fault features of the target domain.

[0223] Optionally, the signal classification model uses a lightweight convolutional network, wherein the lightweight convolutional network is constructed by a depth (Depthwise) convolution, a pointwise (Pointwise) convolution, and a group (Group) convolution.

[0224] The storage medium embodiment of the present application is basically the same as the above-mentioned motor current signal fault diagnosis method embodiments, and will not be described here.

[0225] The present application also provides a computer program product, including a computer program, which is executed by a processor to implement the steps of the above-mentioned motor current signal fault diagnosis method.

[0226] The computer program product embodiment of the present application is basically the same as the above-mentioned motor current signal fault diagnosis method embodiments, and will not be described here.

[0227] It should be noted that in this paper, the term "includes", "contains" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or includes elements inherent to such process, method, article or device. Without more limitations, the element defined by the sentence "includes a" does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0228] The above-mentioned embodiment number of the present application is only for description, not representing the pros and cons of the embodiments.

[0229] Through the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment method can be realized by software plus the necessary general hardware platform, of course, it can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, including a plurality of instructions to make a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) execute the method described in each embodiment of the present application.

[0230] The above is only the preferred embodiment of the present application, and does not limit the patent scope of the present application, and any equivalent structure or equivalent process transformation using the contents of the present application specification and drawings, or direct or indirect application in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for fault diagnosis of motor current signals, characterized in that, The fault diagnosis method for the motor current signal includes: Obtain the motor current signal; The motor current signal is input into a preset signal classification model, and fault identification processing is performed on the motor current signal based on the signal classification model to obtain fault information; The signal classification model includes a feature extractor and a fault discriminator, which are trained based on a semantically aligned grouping adversarial method. The training method for the signal classification model includes: Obtain source domain current signal samples and fault result labels for the source domain current signal samples; Based on the source domain current signal samples, an initial feature extractor is obtained through pre-training; The source domain current signal samples are grouped into a preset first number of semantic alignment groups to obtain grouped source domain current signal samples, wherein the semantic alignment groups include different types of fault result labels; Based on the grouped source domain current signal samples, the initial fault discriminator is obtained through grouped adversarial training. Based on the source domain current signal samples and the fault result labels of the source domain current signal samples, the initial feature extractor and the initial fault discriminator are iteratively trained to obtain a signal classification model that meets the accuracy requirements.

2. The fault diagnosis method for motor current signals as described in claim 1, characterized in that, The step of iteratively training the initial feature extractor and the initial fault discriminator based on the source domain current signal samples and the fault result labels of the source domain current signal samples to obtain a signal classification model that meets the accuracy requirements includes: The source domain current signal samples in the semantic alignment group are input to the initial feature extractor. Based on the initial feature extractor, the first fault features of the source domain current signal samples are extracted, and based on the first fault features, the predicted fault information is determined. Based on the predicted fault information and the fault result label, determine the fault classification loss; The first fault feature is input into the initial fault discriminator to obtain the group adversarial loss; Based on the fault classification loss and the group adversarial loss, the target loss is determined; Determine whether the target loss meets the loss criteria indicated by the preset loss threshold range; If the target loss does not meet the loss criterion indicated by the preset loss threshold range, the process returns to the step of inputting the source domain current signal samples in the semantic alignment group into the initial feature extractor, extracting the first fault features of the source domain current signal samples based on the initial feature extractor, and obtaining the predicted fault information. Training stops when the target loss meets the loss criterion indicated by the preset loss threshold range, and a signal classification model that meets the accuracy conditions is obtained.

3. The fault diagnosis method for motor current signals as described in claim 1, characterized in that, The step of iteratively training the initial feature extractor and the initial fault discriminator based on the source domain current signal samples and the fault result labels of the source domain current signal samples to obtain a signal classification model that meets the accuracy requirements includes: Acquire current signal samples in the target domain; Based on the source domain current signal samples, the fault result labels of the source domain current signal samples, and the target domain current signal samples, the initial feature extractor and the initial fault discriminator are iteratively trained using the maximum mean difference (MMD) distance metric to obtain a signal classification model that meets the accuracy requirements.

4. The fault diagnosis method for motor current signals as described in claim 3, characterized in that, The step of iteratively training the initial feature extractor and the initial fault discriminator based on the source domain current signal samples, the fault result labels of the source domain current signal samples, and the target domain current signal samples, combined with the maximum mean difference (MMD) distance metric, to obtain a signal classification model that meets the accuracy requirements, includes: The source domain current signal samples in the semantic alignment group are input to the initial feature extractor. Based on the initial feature extractor, the first fault features of the source domain current signal samples are extracted, and based on the first fault features, the predicted fault information is determined. Based on the predicted fault information and the fault result label, determine the fault classification loss; The first fault feature is input into the initial fault discriminator to obtain the group adversarial loss; The target domain current signal sample is input into the initial feature extractor to obtain the second fault feature of the target domain; The distance loss of the first fault feature and the second fault feature is calculated using the maximum mean difference (MMD) distance metric function. The target loss is obtained by summing the fault classification loss, group adversarial loss, and distance loss. Determine whether the target loss meets the loss criteria indicated by the preset loss threshold range; If the target loss does not meet the loss criterion indicated by the preset loss threshold range, the process returns to the step of inputting the source domain current signal samples in the semantic alignment group into the initial feature extractor, extracting the first fault features of the source domain current signal samples based on the initial feature extractor, and obtaining the predicted fault information. Training stops when the target loss meets the loss criterion indicated by the preset loss threshold range, and a signal classification model that meets the accuracy conditions is obtained.

5. The fault diagnosis method for motor current signals as described in claim 4, characterized in that, The step of inputting the target domain current signal sample into the initial feature extractor to obtain the second fault feature of the target domain includes: Perform a Fast Fourier Transform on the target domain current signal sample to obtain the transformed target domain current signal sample; The transformed target domain current signal sample is normalized to obtain the normalized target domain current signal sample. The normalized target domain current signal sample is input into the initial feature extractor to obtain the second fault feature of the target domain.

6. The fault diagnosis method for motor current signals as described in claim 1, characterized in that, The signal classification model employs a lightweight convolutional network, which is constructed from depthwise convolutions, pointwise convolutions, and group convolutions.

7. A fault diagnosis device for motor current signals, characterized in that, The fault diagnosis device for the motor current signal includes: The acquisition module is used to acquire the motor current signal; The identification module is used to input the motor current signal into a preset signal classification model, and perform fault identification processing on the motor current signal based on the signal classification model to obtain fault information; wherein, the signal classification model includes a feature extractor and a fault discriminator, and the feature extractor and fault discriminator are trained based on a semantic alignment grouping adversarial method, and the training method of the signal classification model includes: Obtain source domain current signal samples and fault result labels for the source domain current signal samples; Based on the source domain current signal samples, an initial feature extractor is obtained through pre-training; The source domain current signal samples are grouped into a preset first number of semantic alignment groups to obtain grouped source domain current signal samples, wherein the semantic alignment groups include different types of fault result labels; Based on the grouped source domain current signal samples, the initial fault discriminator is obtained through grouped adversarial training. Based on the source domain current signal samples and the fault result labels of the source domain current signal samples, the initial feature extractor and the initial fault discriminator are iteratively trained to obtain a signal classification model that meets the accuracy requirements.

8. A fault diagnosis device for motor current signals, characterized in that, The fault diagnosis device for the motor current signal includes: a memory, a processor, and a program stored in the memory for implementing the fault diagnosis method for the motor current signal. The memory is used to store a program for a fault diagnosis method that implements motor current signals; The processor is configured to execute a program that implements a fault diagnosis method for the motor current signal, thereby implementing the steps of the fault diagnosis method for the motor current signal as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium stores a program for implementing a fault diagnosis method for motor current signals, and the program for implementing the fault diagnosis method for motor current signals is executed by a processor to implement the steps of the fault diagnosis method for motor current signals as described in any one of claims 1 to 6.

Images