Training method of bearing fault diagnosis model, bearing fault diagnosis method and device

Abstract

The present application is applicable to the technical field of bearing fault diagnosis and provides a bearing fault diagnosis model training method, a bearing fault diagnosis method, and an apparatus. The bearing fault diagnosis model training method includes: obtaining a vibration signal, a temperature signal, a speed signal of a first bearing, and an operating condition signal of a first motor, where the first motor is the drive motor of the first bearing; extracting the initial frequency domain features of the vibration signal and fusing the initial frequency domain features with the vibration signal to obtain a first fused feature; fusing the first fused feature, the temperature signal, the speed signal, and the operating condition signal of the first motor to obtain a second fused feature; and training a pre-established bearing fault diagnosis model based on the second fused feature to obtain a trained bearing fault diagnosis model. The input data of the pre-established bearing fault diagnosis model is the second fused feature, and the training label includes the fault type of the first bearing. The present application can effectively improve the accuracy of bearing fault diagnosis.

Description

Technical Field

[0001] The present application belongs to the technical field of bearing fault diagnosis, and in particular relates to a training method for a bearing fault diagnosis model, a bearing fault diagnosis method, and a bearing fault diagnosis device. Background Art

[0002] With the rapid development of modern industry, various mechanical equipment are widely used in the industrial field. The proper operation of equipment is crucial for its safe and efficient long-term operation and is also an important guarantee for the normal production of the entire enterprise. Therefore, timely equipment testing and fault diagnosis are crucial.

[0003] Traditional rotating machinery fault diagnosis methods based on vibration signals rely on diagnostic experts to observe the vibration signal spectrum to locate and diagnose the fault. However, this diagnostic approach is subject to significant subjective influence and has low diagnostic efficiency. Therefore, the industry currently often uses deep learning methods for fault diagnosis, which require large amounts of labeled data. However, given the variability of operating conditions, obtaining labeled datasets for all scenarios is impractical, and the operating conditions of actual industrial equipment are complex and highly variable. Consequently, in practice, the data obtained by deep learning-based fault diagnosis methods fails to meet the requirements of deep learning, resulting in inaccurate fault diagnosis results. Summary of the Invention

[0004] The embodiments of the present application provide a bearing fault diagnosis model training method, a bearing fault diagnosis method and a bearing fault diagnosis device to solve the problem of low accuracy of current bearing fault diagnosis methods.

[0005] This application is achieved through the following technical solutions:

[0006] In a first aspect, an embodiment of the present application provides a bearing fault diagnosis method, comprising: obtaining a vibration signal, a temperature signal, a speed signal of a first bearing, and an operating condition signal of a first motor, wherein the first motor is a drive motor of the first bearing; extracting an initial frequency domain feature of the vibration signal, and fusing the initial frequency domain feature and the vibration signal to obtain a first fusion feature; fusing the first fusion feature, the temperature signal, the speed signal, and the operating condition signal of the first motor to obtain a second fusion feature; based on the second fusion feature, training a pre-established bearing fault diagnosis model to obtain a trained bearing fault diagnosis model; wherein the input data of the pre-established bearing fault diagnosis model is the second fusion feature, and the training label includes the fault type of the first bearing; the trained bearing fault diagnosis model is used to perform fault diagnosis on the target bearing based on the vibration signal, temperature signal, speed signal of the target bearing, and the operating condition signal of the second motor; wherein the second motor is the drive motor of the target bearing.

[0007] In combination with the first aspect, in some embodiments, extracting the initial frequency domain features of the vibration signal and fusing the initial frequency domain features with the vibration signal to obtain a first fusion feature includes: performing a fast Fourier transform on the vibration signal to obtain the initial frequency domain features; inputting the initial frequency domain features and the vibration signal into a first convolution module to obtain a first frequency domain feature and a first time domain feature; inputting the first frequency domain feature and the first time domain feature into a second convolution module to obtain a second frequency domain feature and a second time domain feature; adding the second frequency domain feature and the second time domain feature to obtain The first intermediate feature is obtained by inputting the first intermediate feature into the compression and excitation network to obtain the second intermediate feature; the second intermediate feature is multiplied by the second frequency domain feature and the second time domain feature respectively to obtain the third frequency domain feature and the third time domain feature respectively; the third frequency domain feature and the third time domain feature are input into the third convolution module to obtain the fourth frequency domain feature and the fourth time domain feature; the fourth frequency domain feature and the fourth time domain feature are input into the fourth convolution module to obtain the fifth frequency domain feature and the fifth time domain feature; the fifth frequency domain feature is added to the fifth time domain feature to obtain the first fusion feature.

[0008] In combination with the first aspect, in some embodiments, the second fusion feature includes first data and second data, the first data is a labeled second fusion feature, and the second data is an unlabeled second fusion feature; based on the second fusion feature, a pre-established bearing fault diagnosis model is trained, including: inputting the first data and the second data into a first encoder, and after being processed by multiple encoders in sequence, obtaining third data and fourth data output by the last encoder, the third data is the feature vector corresponding to the first data, and the fourth data is the feature vector corresponding to the second data; performing cosine similarity calculation on the third data and the fourth data, and based on the calculation result, using a label propagation algorithm to assign a pseudo label to the fourth data to obtain fifth data; inputting the third data and the fifth data into the first fully connected layer to obtain the diagnosis result output by the first fully connected layer; using a label smoothing loss function to calculate the error of the diagnosis result output by the first fully connected layer, and updating the parameters of the pre-established bearing fault diagnosis model through the Adam optimizer.

[0009] In the second aspect, an embodiment of the present application provides a bearing fault diagnosis method, including: obtaining a vibration signal, a temperature signal, a speed signal of a target bearing, and an operating condition signal of a second motor, wherein the second motor is a drive motor of the target bearing; extracting the initial frequency domain features of the vibration signal of the target bearing, and fusing the initial frequency domain features with the vibration signal of the target bearing to obtain a third fusion feature; fusing the third fusion feature, the temperature signal of the target bearing, the speed signal of the target bearing, and the operating condition signal of the second motor to obtain a fourth fusion feature; inputting the fourth fusion feature into a trained bearing fault diagnosis model to obtain a diagnosis result of the target bearing output by the trained bearing fault diagnosis model; wherein the trained bearing fault diagnosis model is trained according to the above-mentioned fault diagnosis model training method.

[0010] In a third aspect, an embodiment of the present application provides a training device for a bearing fault diagnosis model, comprising: a first acquisition module for acquiring a vibration signal, a temperature signal, a speed signal of a first bearing, and an operating condition signal of a first motor, wherein the first motor is a drive motor of the first bearing; a first processing module for extracting an initial frequency domain feature and a first time domain feature of the vibration signal, and fusing the initial frequency domain feature and the first time domain feature to obtain a first fusion feature; fusing the first fusion feature, the temperature signal, the speed signal, and the operating condition signal of the first motor to obtain a second fusion feature; a training module for training a pre-established bearing fault diagnosis model based on the second fusion feature to obtain a trained bearing fault diagnosis model; wherein the input data of the pre-established bearing fault diagnosis model is the second fusion feature, and the training label includes the fault type of the first bearing; the trained bearing fault diagnosis model is used to perform fault diagnosis on the target bearing based on the vibration signal, temperature signal, speed signal of the target bearing, and the operating condition signal of the second motor; wherein the second motor is the drive motor of the target bearing.

[0011] In a fourth aspect, an embodiment of the present application provides a bearing fault diagnosis device, comprising: a second acquisition module, used to acquire the vibration signal, temperature signal, speed signal of the target bearing, and the operating condition signal of the second motor, the second motor being the drive motor of the target bearing; a second processing module, used to extract the initial frequency domain features of the vibration signal of the target bearing, and fuse the initial frequency domain features with the vibration signal of the target bearing to obtain a third fusion feature; fuse the third fusion feature, the temperature signal of the target bearing, the speed signal of the target bearing, and the operating condition signal of the second motor to obtain a fourth fusion feature; a diagnosis module, used to input the fourth fusion feature into a trained bearing fault diagnosis model to obtain a diagnosis result of the target bearing output by the trained bearing fault diagnosis model; wherein, the trained bearing fault diagnosis model is trained according to the training method of the fault diagnosis model described in any one of the first aspects above.

[0012] In a fifth aspect, an embodiment of the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the training method for the bearing fault diagnosis model described in any one of the first aspects above or implements the bearing fault diagnosis method described in any one of the second aspects above.

[0013] In a sixth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the training method of the bearing fault diagnosis model described in any one of the first aspects above or implements the bearing fault diagnosis method described in any one of the second aspects above.

[0014] In the seventh aspect, an embodiment of the present application provides a computer program product. When the computer program product is run on an electronic device, the electronic device executes the training method of the bearing fault diagnosis model described in any one of the first aspects above or implements the bearing fault diagnosis method described in any one of the second aspects above.

[0015] Compared with the related art, the embodiments of the present application have the following beneficial effects:

[0016] The training method of the bearing fault diagnosis model provided in the embodiment of the present application first extracts the initial frequency domain features of the vibration signal of the first bearing, and fuses the initial frequency domain features with the vibration signal to obtain a first fusion feature, and then fuses the first fusion feature, the temperature signal, the speed signal, and the operating condition signal of the drive motor of the first bearing to obtain a second fusion feature; then, based on the second fusion feature, the pre-established bearing fault diagnosis model is trained to obtain a trained bearing fault diagnosis model. In this way, before training the model, the data to be input is first deeply fused, and the vibration signal, temperature signal, speed signal and the operating condition signal of the drive motor of the first bearing are considered at the same time, thereby achieving a more comprehensive feature fusion. The fused features are then input into the pre-established fault diagnosis model for training, so that the model can achieve better deep learning effects under the conditions of a smaller amount of training data, thereby improving the accuracy of the bearing fault diagnosis model.

[0017] The bearing fault diagnosis method provided in the embodiment of the present application first obtains the vibration signal, temperature signal, speed signal of the target bearing, and the operating condition signal of the second motor; then extracts the initial frequency domain feature of the vibration signal of the target bearing, and fuses the initial frequency domain feature with the vibration signal of the target bearing to obtain a third fusion feature; fuses the third fusion feature, the temperature signal of the target bearing, the speed signal of the target bearing, and the operating condition signal of the second motor to obtain a fourth fusion feature; finally, inputs the fourth fusion feature into the above-mentioned trained bearing fault diagnosis model to obtain the diagnostic result of the target bearing output by the trained bearing fault diagnosis model. In this way, before performing fault diagnosis through the model, the acquired vibration signal is first subjected to feature extraction and deep fusion of features, so that the vibration signal has a stronger discrimination ability, and then other signals are fused with the fused vibration signal to obtain a more comprehensive feature for diagnosing bearing faults, so that the trained bearing fault diagnosis model can input more accurate results, thereby improving the accuracy of bearing fault diagnosis.

[0018] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0020] Figure 1 1 is a flow chart of a method for training a bearing fault diagnosis model provided in one embodiment of the present application;

[0021] Figure 2 This is a schematic diagram of the structure of a vibration signal feature fusion network provided in one embodiment of the present application;

[0022] Figure 3 Schematic diagram of the structure of the compression and excitation network provided in one embodiment of the present application;

[0023] Figure 4 This is a structural diagram of a bearing fault diagnosis model provided by an embodiment of the present application;

[0024] Figure 5 is a schematic diagram of a single encoder provided in one embodiment of the present application;

[0025] Figure 6 This is a loss curve diagram during the training process of the bearing fault diagnosis model provided in one embodiment of the present application;

[0026] Figure 7 This is a graph showing the accuracy of the bearing fault diagnosis model training process provided by an embodiment of the present application;

[0027] Figure 8 1 is a flow chart of a bearing fault diagnosis method provided in an embodiment of the present application;

[0028] Figure 9 Schematic diagram of the structure of the training device for the bearing fault diagnosis model provided in an embodiment of the present application;

[0029] Figure 10 Schematic diagram of the structure of the bearing fault diagnosis device provided in an embodiment of the present application;

[0030] Figure 11 It is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0031] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0032] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0033] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0034] References to "one embodiment" or "some embodiments" in this specification mean that a particular feature, structure, or characteristic described in conjunction with that embodiment is included in one or more embodiments of the present application. Thus, phrases such as "in one embodiment," "in some embodiments," "in other embodiments," and "in other embodiments" appearing in various places in this specification do not necessarily refer to the same embodiment, but rather mean "one or more but not all embodiments," unless otherwise specifically emphasized. The terms "including," "comprising," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0035] Figure 1 This is a flow chart of a training method for a bearing fault diagnosis model provided by an embodiment of the present application, with reference to Figure 1 , the training method of the bearing fault diagnosis model is detailed as follows:

[0036] S101, obtaining a vibration signal, a temperature signal, a speed signal of a first bearing, and an operating condition signal of a first motor, where the first motor is a driving motor of the first bearing.

[0037] In this embodiment of the present application, the first bearing refers to the bearing used for sample data collection. The sample data is used to train and validate the bearing fault diagnosis model. It is understood that the vibration signal, temperature signal, speed signal, and operating condition signal of the first motor are all sample data. It is also understood that there may be multiple first bearings. Optionally, the first bearings include normal bearings, bearings with inner ring faults, bearings with outer ring faults, and bearings with poor lubrication.

[0038] In some embodiments, the vibration signal of the first bearing can be collected by a sensor provided on a mechanical component in contact with the first bearing. Specifically, the vibration signal within a preset time length can be collected at a preset collection frequency. For example, the preset time length can be 3 seconds, and the collection frequency can be 4kHz. The temperature signal of the first bearing refers to the surface temperature of the first bearing, which can be collected by a non-contact temperature collection device. The speed signal of the first bearing refers to the speed of the first bearing itself. The operating condition signal of the first motor may include but is not limited to motor parameter signals such as current signal, power signal and load parameter signal.

[0039] Optionally, the vibration signal is vibration acceleration.

[0040] In one possible implementation, a set of sample data may include the vibration signal, temperature signal, speed signal, and operating condition signal of a first bearing within a preset time period. Since the vibration signal is collected according to a preset collection frequency, the vibration signal can be in the form of a time series within the preset time period, that is, the vibration signal in a set of sample data is actually a one-dimensional array. The temperature signal, speed signal, and operating condition signal of the first motor do not fluctuate significantly within the preset time period. Therefore, these signals can be collected once within each preset time period, or multiple data of each signal can be collected within each preset time period, and the average value can be calculated to determine the value of the temperature signal, speed signal, and operating condition data of the first motor within the preset time period.

[0041] In some embodiments, after step S101, the bearing fault diagnosis model training method may further include labeling a portion of the sample data as normal or faulty, and dividing the sample data into a training set and a validation set. The fault type labels include inner race fault labels, outer race fault labels, and poor lubrication labels. The remaining sample data is unlabeled and used to train the bearing fault diagnosis model.

[0042] Optionally, 70% of the labeled sample data is divided into a training set, and the remaining 30% of the labeled sample data is divided into a validation set.

[0043] S102, extracting the initial frequency domain features of the vibration signal of the first bearing, and fusing the initial frequency domain features with the vibration signal to obtain a first fusion feature; fusing the first fusion feature, the temperature signal, the speed signal, and the operating condition signal of the first motor to obtain a second fusion feature.

[0044] Optionally, the extraction of the initial frequency domain features of the vibration signal of the first bearing in step S102 may be achieved by performing a fast Fourier transform on the vibration signal of the first bearing.

[0045] In some embodiments, the fusion of the initial frequency domain features and the vibration signal in step S102 to obtain the first fused features can be achieved through a time-frequency domain feature fusion network. Figure 2 The time-frequency domain feature fusion network 200 includes a first convolution module, a second convolution module, a feature fusion module, a third convolution module, and a fourth convolution module. The feature fusion module includes an addition node, a Squeeze-and-Excitation Network (SENet), and a multiplication node. The process of processing a vibration signal using the time-frequency domain feature fusion network 200 includes steps S201 to S204:

[0046] S201, input the initial frequency domain features and the vibration signal into a first convolution module to obtain a first frequency domain feature and a first time domain feature; input the first frequency domain feature and the first time domain feature into a second convolution module to obtain a second frequency domain feature and a second time domain feature.

[0047] S202, through the addition node, add the second frequency domain feature and the second time domain feature to obtain the first intermediate feature, and input the first intermediate feature into the compression and excitation network to obtain the second intermediate feature; through the multiplication node, multiply the second intermediate feature with the second frequency domain feature and the second time domain feature respectively, to obtain the third frequency domain feature and the third time domain feature respectively.

[0048] S203, input the third frequency domain feature and the third time domain feature into the third convolution module to obtain a fourth frequency domain feature and a fourth time domain feature; input the fourth frequency domain feature and the fourth time domain feature into the fourth convolution module to obtain a fifth frequency domain feature and a fifth time domain feature.

[0049] S204: Add the fifth frequency domain feature and the fifth time domain feature to obtain a first fusion feature.

[0050] See also Figure 2 In the above embodiment, the first convolution module, the second convolution module, the third convolution module and the fourth convolution module each include two parallel convolution layers, and the output end of each convolution layer is connected to a batch normalization layer. The second convolution module and the fourth convolution module each also include two pooling layers, and the input end of each pooling layer is connected to the output end of a batch normalization layer. Among them, the first convolution module includes a first convolution layer 201 and a second convolution layer 202, the first convolution layer 201 is used to convolve the initial frequency domain features, and the second convolution layer 202 is used to convolve the above-mentioned vibration signal; the second convolution module includes a third convolution layer 203 and a fourth convolution layer 204, the third convolution layer 203 is used to convolve the first frequency domain features, and the fourth convolution layer 204 is used to convolve the first time domain features; the third convolution module includes a fifth convolution layer 205 and a sixth convolution layer 206, the fifth convolution layer 205 is used to convolve the third frequency domain features, and the sixth convolution layer 206 is used to convolve the third time domain features; the fourth convolution module includes a seventh convolution layer 207 and an eighth convolution layer 208, the seventh convolution layer 207 is used to convolve the fourth frequency domain features, and the eighth convolution layer 208 is used to convolve the fourth time domain features.

[0051] In this embodiment, the input data for the time-frequency domain feature fusion network 200 is the vibration signal and the initial frequency domain features obtained by fast Fourier transforming the vibration signal. Since both the vibration signal and the initial frequency domain features are one-dimensional arrays, each of the convolutional layers performs a one-dimensional convolution calculation.

[0052] In a possible implementation, each convolution layer in the time-frequency domain feature fusion network 200 includes multiple convolution kernels. Therefore, the first fusion feature finally output by the time-frequency domain feature fusion network 200 is a two-dimensional array.

[0053] See also Figure 3 The compression and excitation network 300 includes a global pooling layer, a fully connected layer 1, a ReLU activation function, a fully connected layer 2, and a Sigmoid activation function. The compression and excitation network 300 can increase the weight of useful features in the first intermediate features, improve the effect of feature fusion, and thus obtain more effective second intermediate features.

[0054] In some embodiments, the step S102 of fusing the first fusion feature, the temperature signal, the speed signal, and the operating condition signal of the first motor to obtain the second fusion feature may include: combining the temperature signal, the speed signal, and the operating condition signal of the first bearing and the first motor in the form of a matrix to obtain a first matrix; and splicing the first fusion feature with the first matrix to obtain the second fusion feature.

[0055] In other embodiments, the step S102 of fusing the first fusion feature, the temperature signal, the speed signal, and the operating condition signal of the first motor to obtain the second fusion feature may include: converting the temperature signal, the speed signal, and the operating condition signal of the first bearing and the first motor into matrices of a first size, respectively, wherein the width in the first size is the same as the width of the size of the first fusion feature; and splicing the matrices of each first size and the first fusion feature into one matrix to obtain the second fusion feature.

[0056] It can be understood that since the first fusion feature is a two-dimensional array, or it can be said that the first fusion feature is a matrix, and in a set of sample data, the temperature signal, the speed signal, and the operating condition signal of the first motor are all just a numerical value, the temperature signal, the speed signal, and the operating condition signal of the first motor need to be converted into a matrix of a size that can be spliced ​​with the first fusion feature.

[0057] For example, if the size of the first fusion feature is 64*16, the temperature signal can be first converted into a vector of size 1*16. The value of the temperature signal can be the first element of the vector, and the other elements are set to 0. Similarly, the same processing is performed on the speed signal, the current signal of the first motor, the power signal of the first motor, and the load signal of the first motor. In this way, 5 1*16 vectors can be obtained. In the first method, these 5 vectors are spliced ​​into a 5*16 feature matrix, that is, the feature matrix contains information about the temperature signal, the speed signal, and the operating condition signal of the first motor. At this time, the feature matrix and the first fusion feature can be spliced ​​into a matrix of size 69*16, that is, the second fusion feature. In the second method, the 5 vectors and the first fusion feature are directly spliced ​​to obtain the second fusion feature.

[0058] S103, based on the second fusion feature, train the pre-established bearing fault diagnosis model to obtain a trained bearing fault diagnosis model; wherein, the input data of the pre-established bearing fault diagnosis model is the second fusion feature, and the training label includes the fault type of the first bearing; the trained bearing fault diagnosis model is used to perform fault diagnosis on the target bearing based on the vibration signal, temperature signal, speed signal of the target bearing, and the operating condition signal of the second motor; wherein, the second motor is the drive motor of the target bearing.

[0059] In the embodiment of the present application, the training labels also include normal labels, which refer to the normal labels or fault labels assigned to some sample data.

[0060] It can be understood that the first bearing and the target bearing are bearings of the same model and the same application scenario, or the first bearing and the target bearing are bearings of the same type and similar model and the same application scenario.

[0061] In one possible implementation, the pre-established bearing fault diagnosis model may include an encoder and a first fully connected layer. It should be noted that the bearing fault diagnosis model in the implementation example of this application is constructed based on the structure of the transformer model. Figure 4 This is a structural diagram of a bearing fault diagnosis model provided in one embodiment of the present application. Figure 5 The structure of a single encoder provided by an embodiment of the present application is shown. Figure 4 , there are multiple encoders in the above bearing fault diagnosis model, and the multiple encoders are connected in sequence. Except for the last encoder, the output of each encoder is the input of the next encoder, and the output of the last encoder is the input of the first fully connected layer. Figure 5, the encoder 500 includes a multi-head self-attention network 510, a first residual network, a feedforward network 520 and a second residual network. Since the residual network is usually illustrated by the residual connection mechanism, Figure 5 In the figure, the first residual connection 511 is used to represent the first residual network, and the second residual connection 521 is used to represent the second residual network. The input data of the encoder 500 is the input data of the multi-head self-attention network 510, and the input data of the first residual network includes the input data and output data of the multi-head self-attention network 510, that is, through the first residual connection, the input data and output data of the multi-head self-attention network 510 are added, and then the added data are layer-normalized. The input data of the feedforward network 520 is the data after the output data of the first residual network is layer-normalized (Layer Norm), and the input data of the second residual network is the input data and output data of the feedforward network 520, that is, through the second residual connection, the input data and output data of the feedforward network 520 are added, and then the added data are layer-normalized. The output data of the encoder 500 is the data after the output data of the second residual network is layer-normalized.

[0062] Optionally, the output dimension of the first fully connected layer is 4, that is, the first fully connected layer can output 4 classification results, including inner race fault, outer race fault, poor lubrication and normal.

[0063] Furthermore, the multi-head self-attention network 510 includes multiple self-attention modules, and the feedforward network 520 includes two fully connected layers and two layers of ReLU activation functions.

[0064] In each self-attention module of the multi-head self-attention network 510, three matrix vectors are defined: W Q 、W K 、W V After these three matrix vectors make linear changes to the second fusion feature, we get three matrix vectors again: query matrix Q, key matrix K, value matrix V. Then, we calculate the output Z of the self-attention module according to the following formula:

[0065]

[0066] Among them, d k is the dimension of matrix Q, and the softmax function is:

[0067]

[0068] Among them, e i and e j Indicates that the exponential operation is performed on the element.

[0069] It can be understood that {Q, K, V} of multiple self-attention modules can generate multiple Z. The first residual connection can add multiple Z to the second fusion feature.

[0070] The formula for the above layer normalization is:

[0071] D=LayerNorm(c+MultiHeadAttention(c))

[0072] Among them, c is the input of the multi-head self-attention network 510, and MultiHeadAttention represents multiple groups performing attention network calculations.

[0073] In some embodiments, the second fused feature includes first data and second data. The first data refers to the labeled second fused feature, and the second data refers to the unlabeled second fused feature. It is understood that the first data is obtained through step S102 above based on the labeled data in the training set, and the second data is obtained through step S102 above based on the unlabeled data in the training set.

[0074] In some embodiments, the step S103 of training a pre-established bearing fault diagnosis model based on the second fusion feature includes:

[0075] S301, input first data and second data into a first encoder, and after being processed by multiple encoders in sequence, obtain third data and fourth data output by the last encoder, where the third data is a feature vector corresponding to the first data, and the fourth data is a feature vector corresponding to the second data.

[0076] S302 , performing cosine similarity calculation on the third data and the fourth data, and based on the calculation result, using a label propagation algorithm to assign a pseudo label to the fourth data to obtain fifth data.

[0077] S303: Input the third data and the fifth data into the first fully connected layer to obtain a diagnosis result output by the first fully connected layer.

[0078] S304: Calculate the error of the diagnosis result output by the first fully connected layer using a label smoothing loss function, and update the parameters of the bearing fault diagnosis model using an Adam optimizer.

[0079] The calculation formula of the above cosine similarity is:

[0080]

[0081] Among them, A i is the i-th element of vector A, B iis the i-th element of vector B, n is the length of vector A or vector B, and vector A and vector B have the same length.

[0082] The calculation formula of the label smoothing loss function is:

[0083]

[0084] Among them, y i represents the validation set label, p i represents the probability of the predicted fault label, and ε represents a small constant.

[0085] In this embodiment, the training process of the above-mentioned bearing fault diagnosis model also includes: setting the number of training iterations, learning rate and batch size; selecting a preset batch size of sample data from the shuffled training set, and inputting it into the above-mentioned bearing fault diagnosis model, and stopping training when the accuracy of the verification set is greater than a preset threshold, or stopping training when the number of training iterations is reached.

[0086] Optionally, the number of training iterations is 50, the learning rate is 0.001, and the batch size is 32.

[0087] The training method of the bearing fault diagnosis model provided in the embodiment of the present application is to extract the initial frequency domain features of the vibration signal of the first bearing, fuse the initial frequency domain features with the vibration signal, obtain a first fusion feature, and then fuse the first fusion feature, the temperature signal, the speed signal, and the operating condition signal of the drive motor of the first bearing to obtain a second fusion feature; then, based on the second fusion feature, the pre-established bearing fault diagnosis model is trained to obtain a trained bearing fault diagnosis model. In this way, before training the model, the data to be input is first deeply fused, and the vibration signal, temperature signal, speed signal and the operating condition signal of the drive motor of the first bearing are considered at the same time, thereby achieving a more comprehensive feature fusion. The fused features are then input into the pre-established fault diagnosis model for training, so that the model can achieve better deep learning effects under the conditions of a smaller amount of training data, thereby improving the accuracy of the bearing fault diagnosis model.

[0088] The present application also provides a bearing fault diagnosis method. Figure 8 This is a flow chart of a bearing fault diagnosis method provided by an embodiment of the present application. Figure 8 , the bearing fault diagnosis method includes:

[0089] S801, obtaining a vibration signal, a temperature signal, a speed signal of a target bearing, and an operating condition signal of a second motor, where the second motor is a drive motor of the target bearing.

[0090] S802, extract the initial frequency domain features of the vibration signal of the target bearing, and fuse the initial frequency domain features with the vibration signal of the target bearing to obtain a third fusion feature; fuse the third fusion feature, the temperature signal of the target bearing, the speed signal of the target bearing, and the operating condition signal of the second motor to obtain a fourth fusion feature.

[0091] S803: Input the fourth fusion feature into the trained bearing fault diagnosis model to obtain a diagnosis result of the target bearing output by the trained bearing fault diagnosis model.

[0092] In some embodiments, the above-mentioned step S802 may include: performing fast Fourier transform on the vibration signal of the target bearing to obtain the initial frequency domain characteristics of the vibration signal of the target bearing, and then processing the vibration signal of the target bearing and its initial frequency domain characteristics through the time-frequency domain feature fusion network in the above-mentioned embodiment to obtain a third fusion feature; and then according to the method of fusing the first fusion feature, the temperature signal, the speed signal, and the operating condition signal of the first motor in the above-mentioned step S102 to obtain the second fusion feature, the third fusion feature, the temperature signal of the target bearing, the speed signal of the target bearing, and the operating condition signal of the second motor are fused to obtain a fourth fusion feature.

[0093] The above-mentioned bearing fault diagnosis method obtains the vibration signal, temperature signal, speed signal of the target bearing, and the operating condition signal of the second motor; then extracts the initial frequency domain features of the vibration signal of the target bearing, and fuses the initial frequency domain features with the vibration signal of the target bearing to obtain a third fusion feature; fuses the third fusion feature, the temperature signal of the target bearing, the speed signal of the target bearing, and the operating condition signal of the second motor to obtain a fourth fusion feature; finally, inputs the fourth fusion feature into the above-mentioned trained bearing fault diagnosis model to obtain the diagnostic result of the target bearing output by the trained bearing fault diagnosis model. In this way, before fault diagnosis is performed through the model, the acquired vibration signal is first feature extracted and deeply fused, so that the vibration signal has stronger discrimination ability. Then, other signals are fused with the fused vibration signal to obtain more comprehensive features for diagnosing bearing faults, so that the trained bearing fault diagnosis model can input more accurate results, thereby improving the accuracy of bearing fault diagnosis.

[0094] The following describes the bearing fault diagnosis model training method and the bearing fault diagnosis method of the present application in conjunction with a specific embodiment.

[0095] First, sensors were used to collect time-domain vibration acceleration, temperature, and speed signals from multiple sample bearings in an experimental setup. Simultaneously, motor current, motor power, and motor load were collected from the mechanical equipment housing the sample bearings. The vibration acceleration signals were sampled at a 4kHz frequency, and for each sample bearing, four measurement points were set up on the mechanical equipment. Multiple sets of experiments were conducted, collecting both normal and abnormal data. A portion of the normal and abnormal data were labeled as normal and as fault type, respectively. The collected data was shuffled (each sample data set was not shuffled), and the labeled data was divided into a training set x_train and a validation set x_test, with a ratio of 70% for the training set x_train and 30% for the validation set x_test. The labeled data was stored in one file, and the remaining unlabeled data was stored in another file. The sample bearings included normal bearings, bearings with inner race faults, bearings with outer race faults, and bearings with poor lubrication.

[0096] In this embodiment, Table 1 shows a set of vibration acceleration data of a bearing with faults.

[0097] Table 1 Vibration acceleration data

[0098]

[0099]

[0100] It can be understood that the data in Table 1 are only partial data of vibration acceleration collected within a preset time period. Due to the limited space of the description, they are not listed here one by one.

[0101] Next, the vibration signal of the training set x_train is subjected to fast Fourier transform to obtain array a (initial frequency domain features). The processing results are shown in Table 2.

[0102] Table 2 Initial frequency domain characteristics of vibration signal

[0103] Frequency (Hz) Amplitude Frequency (Hz) Amplitude 0.1 0.023 0.7 0.026 0.2 0.06 0.8 0.057 0.3 0.24 0.9 0.002 0.4 0.11 1.0 0.003 0.5 0.05 …… …… 0.6 0.022 …… ……

[0104] Afterwards, the vibration signal and the array a are input into the time-frequency domain feature fusion network. In this embodiment, the size of the vibration signal is 1*4096, and the size of the array a is 1*2048.

[0105] In this embodiment, the first convolution layer of the time-frequency domain feature fusion network contains 32 convolution kernels, the convolution kernel size is 3, and the stride is 32. After the array a is input into the first convolution layer of the first convolution module for processing, feature a1 (first frequency domain feature) with a size of 32*64 is obtained. The second convolution layer contains 32 convolution kernels, the convolution kernel size is 3, and the stride is 64. After the vibration signal is input into the second convolution layer of the first convolution module for processing, feature b1 (first time domain feature) with a size of 32*64 is obtained.

[0106] The third convolutional layer includes 64 convolution kernels of size 3 and stride 1. Feature a1 is fed into the third convolutional layer of the second convolutional module and processed to produce feature a2 (the second frequency domain feature) with a size of 64*32. The fourth convolutional layer includes 64 convolution kernels of size 3 and stride 1. Feature b1 is fed into the fourth convolutional layer of the second convolutional module and processed to produce feature b2 (the second time domain feature) with a size of 64*32.

[0107] In the time-frequency domain feature fusion network, features a2 and b2 are first added together to obtain feature c1 (the first intermediate feature), which is 64*32 in size. Feature c1 is then input into SENet, resulting in a 64*32 matrix c2 (the second intermediate feature). Specifically, in SENet, feature c1 is first processed through global pooling to obtain a 1*64 feature vector. Two fully connected layers are then used to model inter-channel correlations. Fully connected layer 1 (the first fully connected layer) transforms the 64 channels into 16 channels. The ReLU activation function is then used to obtain a 1*16 vector. The number of channels is then restored from 16 to 64 through fully connected layer 2 (the second fully connected layer), resulting in a 1*64 matrix with weight parameters for the attention mechanism. Finally, the Sigmoid activation function is used to apply each weight to the features of each channel, resulting in a 1*64 weight matrix c2. Afterwards, the matrix c2 is multiplied by the feature a2 and the feature b2 respectively to obtain the feature a3 (the third frequency domain feature) and the feature b3 (the third time domain feature).

[0108] The fifth convolutional layer includes 64 convolution kernels of size 3 and stride 1. Feature a3 is fed into the fifth convolutional layer of the third convolutional module and processed to produce feature a4 (the fourth frequency domain feature) with a size of 64*32. The sixth convolutional layer includes 64 convolution kernels of size 3 and stride 1. Feature b3 is fed into the sixth convolutional layer of the third convolutional module and processed to produce feature b4 (the fourth time domain feature) with a size of 64*32.

[0109] The seventh convolutional layer includes 64 convolution kernels of size 3 and stride 1. Feature a4 is fed into the seventh convolutional layer of the fourth convolutional module and processed to produce feature a5 (the fifth frequency domain feature) with a size of 64*16. The eighth convolutional layer includes 64 convolution kernels of size 3 and stride 1. Feature b4 is fed into the eighth convolutional layer of the fourth convolutional module and processed to produce feature b5 (the fifth time domain feature) with a size of 64*16.

[0110] Next, feature a5 and feature b5 are added together to obtain feature d1 (the first fused feature).

[0111] In this embodiment, the temperature collected during the vibration signal collection period is 25.6°C, the speed is 3213.2 r / min, the motor current is 12.1A, the motor power is 500W, and the motor load is 36. After obtaining feature d1, the temperature, speed, motor current, motor power, and motor load parameters are each converted into a 1*16 matrix.

[0112] Temperature converted to [25.6, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0];

[0113] The speed is converted to [3213.2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0];

[0114] The motor current is converted to [12.1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0];

[0115] The motor power is converted to [500,0,0,0,0,0,0,0,0,0,0,0,0,0,0];

[0116] The motor load is converted to [36,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0].

[0117] Merge feature d1 with the above five matrices to obtain feature d2 (the second fused feature) of size 69*16.

[0118] A bearing fault diagnosis model based on the Transformer model was constructed and feature d2 was input into the model. In this model, multiple self-attention modules within a multi-head self-attention network extract features from different dimensions. This self-attention mechanism extracts global features of the vibration signal and allows for parallel computation, overcoming the sequential execution limitations of traditional time series networks. Multiple encoders are connected to continuously extract features from the bearing data. The first fully connected layer outputs the bearing fault diagnosis result. A label smoothing loss function is then used to calculate the error of the diagnosis result to further improve the model's classification performance.

[0119] In this example, a bearing fault diagnosis model was trained with 50 training iterations, a learning rate of 0.001, and a batch size of 32. During training, the Adam optimization algorithm was used to continuously update the network parameters based on the original gradients. Furthermore, pseudo-labels were used for training unlabeled sample data, and the pseudo-labels were updated with each training round. Using pseudo-labels for training can help the neural network learn features that are useful for fault diagnosis, improving the model's generalization capabilities. Figure 6 is a loss curve diagram during the training process of the bearing fault diagnosis model provided in this embodiment; Figure 7 This is a graph showing the accuracy of the bearing fault diagnosis model training process provided by this embodiment. Figure 6 , when the number of iterations reaches 50, the training loss has dropped to 0.2. Figure 7 , when the number of iterations reaches 50, the training accuracy reaches 0.95.

[0120] Finally, a bearing to be diagnosed with the same model and application scenario as the sample bearing is selected. The vibration signal, temperature signal, speed signal, and corresponding motor current, motor power and motor load parameters of the bearing to be diagnosed are collected in the same collection method as the sample bearing. The above-mentioned feature fusion is performed on these signals to obtain the second fused feature of the bearing to be diagnosed. The second fused feature is input into the trained bearing fault diagnosis model to obtain the fault diagnosis result of the bearing to be diagnosed.

[0121] The above embodiment can more comprehensively extract the characteristics of the vibration signal through the deep fusion of cross-adaptive time domain feature channels; then combine the vibration, speed, temperature, motor current, motor power, and motor load data, use the bearing fault diagnosis model based on the transformer model for fault diagnosis, and adopt pseudo-label training in the training process, which can improve the generalization ability of the model and the accuracy of fault diagnosis. Compared with the traditional fault diagnosis method of calculating feature state indicators and constructing thresholds, the method in this embodiment is simple and takes less time, and still has a high accuracy rate on a data set with a small amount of training data.

[0122] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0123] Corresponding to the training method of the bearing fault diagnosis model described in the above embodiment, Figure 9 A structural schematic diagram of a training device for a bearing fault diagnosis model provided in an embodiment of the present application is shown. For ease of explanation, only the parts related to the embodiment of the present application are shown.

[0124] See also Figure 9 The training device for the bearing fault diagnosis model in the embodiment of the present application may include a first acquisition module 910 , a first processing module 920 and a training module 930 .

[0125] The first acquisition module 910 is configured to acquire a vibration signal, a temperature signal, a speed signal of the first bearing, and an operating condition signal of the first motor, where the first motor is a drive motor of the first bearing.

[0126] The first processing module 920 is configured to extract an initial frequency domain feature and a first time domain feature of the vibration signal, fuse the initial frequency domain feature and the first time domain feature to obtain a first fused feature, and fuse the first fused feature, the temperature signal, the speed signal, and the operating condition signal of the first motor to obtain a second fused feature.

[0127] Training module 930 is used to train a pre-established bearing fault diagnosis model based on the second fusion feature to obtain a trained bearing fault diagnosis model; wherein, the input data of the pre-established bearing fault diagnosis model is the second fusion feature, and the training label includes the fault type of the first bearing; the trained bearing fault diagnosis model is used to perform fault diagnosis on the target bearing based on the vibration signal, temperature signal, speed signal of the target bearing, and the operating condition signal of the second motor; wherein, the second motor is the drive motor of the target bearing.

[0128] Optionally, the first processing module 920 is specifically used to: perform fast Fourier transform on the vibration signal to obtain initial frequency domain features; input the initial frequency domain features and the vibration signal into the first convolution module to obtain first frequency domain features and first time domain features; input the first frequency domain features and the first time domain features into the second convolution module to obtain second frequency domain features and second time domain features; add the second frequency domain features to the second time domain features to obtain a first intermediate feature, and input the first intermediate feature into the compression and excitation network to obtain a second intermediate feature; multiply the second intermediate feature with the second frequency domain features and the second time domain features respectively to obtain a third frequency domain feature and a third time domain feature respectively; input the third frequency domain feature and the third time domain feature into the third convolution module to obtain a fourth frequency domain feature and a fourth time domain feature; input the fourth frequency domain feature and the fourth time domain feature into the fourth convolution module to obtain a fifth frequency domain feature and a fifth time domain feature; add the fifth frequency domain feature and the fifth time domain feature to obtain a first fusion feature.

[0129] Optionally, the first processing module 920 is specifically used to: combine the temperature signal, the speed signal and the operating condition signal of the first motor in the form of a matrix to obtain a first matrix; and splice the first fusion feature with the first matrix to obtain a second fusion feature.

[0130] Optionally, the training module 930 is specifically used to: input the first data and the second data into the first encoder, and after being processed by multiple encoders in sequence, obtain the third data and the fourth data output by the last encoder, where the third data is the feature vector corresponding to the first data, and the fourth data is the feature vector corresponding to the second data; perform cosine similarity calculation on the third data and the fourth data, and based on the calculation result, use the label propagation algorithm to assign a pseudo label to the fourth data to obtain the fifth data; input the third data and the fifth data into the first fully connected layer to obtain the diagnosis result output by the first fully connected layer; use the label smoothing loss function to calculate the error of the diagnosis result output by the first fully connected layer, and update the parameters of the pre-established bearing fault diagnosis model through the Adam optimizer.

[0131] Corresponding to the bearing fault diagnosis method described in the above embodiment, Figure 10 A structural schematic diagram of a bearing fault diagnosis device provided in an embodiment of the present application is shown. For ease of explanation, only the parts related to the embodiment of the present application are shown.

[0132] See also Figure 10 The bearing fault diagnosis device 1000 in the embodiment of the present application may include a second acquisition module 1010 , a second processing module 1020 and a diagnosis module 1030 .

[0133] The second acquisition module 1010 is used to acquire the vibration signal, temperature signal, speed signal of the target bearing, and the operating condition signal of the second motor, where the second motor is the drive motor of the target bearing.

[0134] The second processing module 1020 is used to extract the initial frequency domain features of the vibration signal of the target bearing, and fuse the initial frequency domain features with the vibration signal of the target bearing to obtain a third fusion feature; and fuse the third fusion feature, the temperature signal of the target bearing, the speed signal of the target bearing, and the operating condition signal of the second motor to obtain a fourth fusion feature.

[0135] The diagnosis module 1030 is used to input the fourth fusion feature into the above-mentioned trained bearing fault diagnosis model to obtain the diagnosis result of the target bearing output by the trained bearing fault diagnosis model.

[0136] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of this application. Their specific functions and technical effects can be found in the method embodiment section and will not be repeated here.

[0137] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0138] The present application also provides an electronic device. Figure 11 The electronic device 1100 may include: at least one processor 1110, a memory 1120, and a computer program stored in the memory 1120 and executable on the at least one processor 1110. When the processor 1110 executes the computer program, the steps in any of the above-mentioned method embodiments are implemented, for example Figure 1 Steps S101 to S103 in the embodiment shown, or as Figure 8 Alternatively, when the processor 1110 executes the computer program, the functions of the modules / units in the above-mentioned device embodiments are realized, for example, Figure 9 The functions of modules 910 to 930 shown, or as Figure 10The functions of modules 1010 to 1030 are shown.

[0139] For example, the computer program may be divided into one or more modules / units, one or more modules / units being stored in the memory 1120 and executed by the processor 1110 to complete the present application. The one or more modules / units may be a series of computer program segments capable of completing specific functions, and the program segments are used to describe the execution process of the computer program in the electronic device 1100.

[0140] Those skilled in the art will understand that Figure 11 These are merely examples of electronic devices and do not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown in the figure, or a combination of certain components, or different components, such as input and output devices, network access devices, buses, etc.

[0141] The processor 1110 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0142] The memory 1120 can be an internal storage unit of the electronic device or an external storage device of the electronic device, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. The memory 1120 is used to store the computer program and other programs and data required by the electronic device. The memory 1120 can also be used to temporarily store data that has been output or is about to be output.

[0143] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. Buses can be classified into address buses, data buses, and control buses. For ease of illustration, the buses in the drawings of this application are not limited to just one bus or just one type of bus.

[0144] The bearing fault diagnosis method provided in the embodiments of the present application can be applied to electronic devices such as computers, wearable devices, vehicle-mounted devices, tablet computers, laptop computers, netbooks, personal digital assistants (PDAs), augmented reality (AR) / virtual reality (VR) devices, and mobile phones. The embodiments of the present application do not impose any restrictions on the specific types of electronic devices.

[0145] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in each embodiment of the above-mentioned bearing fault diagnosis method can be implemented.

[0146] An embodiment of the present application provides a computer program product. When the computer program product is run on a mobile terminal, the mobile terminal can implement the steps in each embodiment of the above-mentioned bearing fault diagnosis method when executing the computer program product.

[0147] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. Among them, the computer program includes computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device that can carry the computer program code to the camera / electronic device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal and a software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk.

[0148] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0149] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0150] In the embodiments provided in this application, it should be understood that the disclosed devices / network equipment and methods can be implemented in other ways. For example, the device / network equipment embodiments described above are merely illustrative. For example, the division of the modules or units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0151] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0152] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. A training method for a bearing fault diagnosis model, characterized in that: include: Acquire a vibration signal, a temperature signal, a speed signal of a first bearing, and an operating condition signal of a first motor, where the first motor is a drive motor of the first bearing; Extracting an initial frequency domain feature of the vibration signal, and fusing the initial frequency domain feature with the vibration signal to obtain a first fused feature; fusing the first fused feature, the temperature signal, the speed signal, and the operating condition signal of the first motor to obtain a second fused feature; Based on the second fused feature, a pre-established bearing fault diagnosis model is trained to obtain a trained bearing fault diagnosis model; wherein the input data of the pre-established bearing fault diagnosis model is the second fused feature, and the training label includes the fault type of the first bearing; the trained bearing fault diagnosis model is used to perform fault diagnosis on the target bearing based on the vibration signal, temperature signal, speed signal of the target bearing, and the operating condition signal of the second motor; wherein the second motor is the drive motor of the target bearing; The pre-established bearing fault diagnosis model includes an encoder and a first fully connected layer; wherein the encoder is multiple and the multiple encoders are connected in sequence, and the last encoder is connected to the first fully connected layer; The second fused feature includes first data and second data, the first data is a second fused feature with a label, and the second data is a second fused feature without a label; The step of training a pre-established bearing fault diagnosis model based on the second fusion feature includes: Inputting the first data and the second data into a first encoder, and sequentially processing them through multiple encoders to obtain third data and fourth data output by a final encoder, wherein the third data is a feature vector corresponding to the first data, and the fourth data is a feature vector corresponding to the second data; Calculating cosine similarity between the third data and the fourth data, and assigning a pseudo label to the fourth data using a label propagation algorithm based on the calculation result to obtain fifth data; Inputting the third data and the fifth data into the first fully connected layer to obtain a diagnosis result output by the first fully connected layer; The label smoothing loss function is used to calculate the error of the diagnosis result output by the first fully connected layer, and the parameters of the pre-established bearing fault diagnosis model are updated through the Adam optimizer.

2. The training method for a bearing fault diagnosis model according to claim 1, wherein: The extracting the initial frequency domain feature of the vibration signal and fusing the initial frequency domain feature with the vibration signal to obtain a first fused feature includes: Performing a fast Fourier transform on the vibration signal to obtain the initial frequency domain feature; Inputting the initial frequency domain feature and the vibration signal into a first convolution module to obtain a first frequency domain feature and a first time domain feature; inputting the first frequency domain feature and the first time domain feature into a second convolution module to obtain a second frequency domain feature and a second time domain feature; Adding the second frequency domain feature and the second time domain feature to obtain a first intermediate feature, and inputting the first intermediate feature into a compression and excitation network to obtain a second intermediate feature; multiplying the second intermediate feature by the second frequency domain feature and the second time domain feature respectively to obtain a third frequency domain feature and a third time domain feature respectively; Inputting the third frequency domain feature and the third time domain feature into a third convolution module to obtain a fourth frequency domain feature and a fourth time domain feature; inputting the fourth frequency domain feature and the fourth time domain feature into a fourth convolution module to obtain a fifth frequency domain feature and a fifth time domain feature; The fifth frequency domain feature and the fifth time domain feature are added to obtain the first fusion feature.

3. The training method for a bearing fault diagnosis model according to claim 1, wherein: The fusing the first fusion feature, the temperature signal, the speed signal, and the operating condition signal of the first motor to obtain a second fusion feature includes: Combining the temperature signal, the speed signal, and the operating condition signal of the first motor in a matrix form to obtain a first matrix; The first fused feature is concatenated with the first matrix to obtain the second fused feature.

4. The training method for a bearing fault diagnosis model according to claim 1, wherein: The encoder includes a multi-head self-attention network, a first residual network, a feedforward network and a second residual network; wherein, the input data of the encoder is the input data of the multi-head self-attention network, the input data of the first residual network includes the input data and output data of the multi-head self-attention network, the input data of the feedforward network is the data after layer normalization processing of the output data of the first residual network, the input data of the second residual network is the input data and output data of the feedforward network, and the output data of the encoder is the data after layer normalization processing of the output data of the second residual network.

5. The training method for a bearing fault diagnosis model according to claim 1, wherein: The operating condition signal includes a current signal, a power signal and a load parameter signal; the fault type includes an inner ring fault, an outer ring fault and poor lubrication.

6. A bearing fault diagnosis method, characterized in that: include: Acquire a vibration signal, a temperature signal, a speed signal of a target bearing, and an operating condition signal of a second motor, where the second motor is a drive motor of the target bearing; Extracting an initial frequency domain feature of the vibration signal of the target bearing, and fusing the initial frequency domain feature with the vibration signal of the target bearing to obtain a third fused feature; fusing the third fused feature, the temperature signal of the target bearing, the speed signal of the target bearing, and the operating condition signal of the second motor to obtain a fourth fused feature; The fourth fusion feature is input into a trained bearing fault diagnosis model to obtain a diagnosis result of the target bearing output by the trained bearing fault diagnosis model; wherein the trained bearing fault diagnosis model is trained according to the method described in any one of claims 1 to 5.

7. A training device for a bearing fault diagnosis model, characterized in that: include: a first acquisition module, configured to acquire a vibration signal, a temperature signal, and a speed signal of a first bearing, and an operating condition signal of a first motor, where the first motor is a drive motor of the first bearing; a first processing module, configured to extract an initial frequency domain feature and a first time domain feature of the vibration signal, fuse the initial frequency domain feature and the first time domain feature to obtain a first fused feature; and fuse the first fused feature, the temperature signal, the speed signal, and the operating condition signal of the first motor to obtain a second fused feature; a training module for training a pre-established bearing fault diagnosis model based on the second fused feature to obtain a trained bearing fault diagnosis model; wherein the input data of the pre-established bearing fault diagnosis model is the second fused feature, and the training label includes the fault type of the first bearing; the trained bearing fault diagnosis model is used to perform fault diagnosis on the target bearing based on the vibration signal, temperature signal, speed signal of the target bearing, and the operating condition signal of the second motor; wherein the second motor is the drive motor of the target bearing; The pre-established bearing fault diagnosis model includes an encoder and a first fully connected layer; wherein the encoder is multiple and the multiple encoders are connected in sequence, and the last encoder is connected to the first fully connected layer; The second fused feature includes first data and second data, the first data is a second fused feature with a label, and the second data is a second fused feature without a label; The training module is specifically used for: Inputting the first data and the second data into a first encoder, and sequentially processing them through multiple encoders to obtain third data and fourth data output by a final encoder, wherein the third data is a feature vector corresponding to the first data, and the fourth data is a feature vector corresponding to the second data; Calculating cosine similarity between the third data and the fourth data, and assigning a pseudo label to the fourth data using a label propagation algorithm based on the calculation result to obtain fifth data; Inputting the third data and the fifth data into the first fully connected layer to obtain a diagnosis result output by the first fully connected layer; The label smoothing loss function is used to calculate the error of the diagnosis result output by the first fully connected layer, and the parameters of the pre-established bearing fault diagnosis model are updated through the Adam optimizer.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images