Classification method, device and equipment for bearing faults based on artificial intelligence, and storage medium

Abstract

The invention provides a bearing fault classification method, device and equipment based on artificial intelligence, and a storage medium, and the method comprises the steps: obtaining a bearing rotation signal collected by a sensor, carrying out the preprocessing of the bearing rotation signal, and carrying out the continuous wavelet transformation, so as to generate a two-dimensional time-frequency graph; next, a pre-trained discriminator is called to carry out sliding feature extraction on the two-dimensional time-frequency graph to generate height features and width features, and key features are identified according to the height features and the width features through an attention mechanism; then, the key features are converted into one-dimensional feature vectors, the one-dimensional feature vectors are used for classifying the key features through a plurality of convolution layers and full connection layers, corresponding fault categories are output, and the fault categories comprise an inner ring fault, an outer ring fault and a rolling body fault. The problem that complex signal features are difficult to capture in existing time domain or frequency domain analysis is solved.

Description

Technical Field

[0001] The present invention relates to the field of bearing fault processing, and in particular to a method, device, equipment, and storage medium for classifying bearing faults based on artificial intelligence. Background Art

[0002] Rotating machinery is an important part of industrial production, and its health status is directly related to the operation stability of the entire production line. Rolling bearings are important components of rotating machinery, and their fault diagnosis is particularly concerned.

[0003] Traditional rolling bearing fault diagnosis relies on a combination of signal processing and machine learning techniques, which usually requires processing the bearing's vibration signal and extracting features, and then using a machine learning model to classify the fault. However, due to data noise, class imbalance, and inconsistent data distribution under actual working conditions, it is difficult to capture complex signal features in traditional time domain or frequency domain analysis.

[0004] In view of this, this application is filed. Summary of the invention

[0005] The present invention discloses a method, device, equipment, and storage medium for classifying bearing faults based on artificial intelligence, aiming to solve the problem that it is difficult to capture complex signal features in existing time domain or frequency domain analysis.

[0006] The first embodiment of the present invention provides a method for classifying bearing faults based on artificial intelligence, comprising: Acquire the bearing rotation signal collected by the sensor, and perform continuous wavelet transform on the bearing rotation signal after preprocessing to generate a two-dimensional time-frequency diagram, wherein the two-dimensional time-frequency diagram can provide the frequency change of the signal at different time points; Calling a pre-trained discriminator to perform sliding feature extraction on the two-dimensional time-frequency graph to generate height features and width features, and identifying key features according to the height features and the width features through an attention mechanism, wherein the discriminator measures real data and false data based on a focal frequency loss function during training until the discriminator and a generator generating the false data reach a Nash equilibrium; The key features are converted into one-dimensional feature vectors, and the one-dimensional feature vectors are used to classify the key features through multiple convolutional layers and fully connected layers, and corresponding fault categories are output, wherein the fault categories include inner race faults, outer race faults, and rolling element faults.

[0007] Preferably, the calling of the pre-trained discriminator to perform sliding feature extraction on the two-dimensional time-frequency graph to generate height features and width features, and identifying key features according to the height features and the width features through an attention mechanism, specifically: Through two global pooling operations, the input two-dimensional time-frequency graph is processed in the spatial dimension to extract the height feature and width feature respectively; Performing a shared convolution operation on the height feature and the width feature to generate a direction-aware feature map, and generating attention weights in the height and width directions; The attention weight is applied to the two-dimensional time-frequency graph, and key features are generated through a Hard-Swish activation function.

[0008] Preferably, the expression of the continuous wavelet transform is: ; in, is a two-dimensional time-frequency diagram, To control the scale of the wavelet function, For time shift, is the preprocessed one-dimensional bearing data, is the scale transform and translation transform of the wavelet function.

[0009] Preferably, the training process of the discriminator includes: The one-dimensional bearing data in the data set is transformed into a two-dimensional time-frequency diagram as the real data by continuous wavelet transformation; Generate random noise, and send the random noise to a generator to generate false data; The real data and the false data are input into a discriminator, and the difference between the real data and the false data in the frequency domain is discriminated, wherein the difference in the frequency domain is measured by a focal frequency loss function, and the weight of the discriminator is updated by back propagation and an optimizer.

[0010] Based on the identification result of the discriminator, the weight of the generator is adjusted until the false data generated by the generator cannot be distinguished by the discriminator.

[0011] Preferably, the expression of the global pooling operation is: , ; Among them, the is the globally pooled feature in the height dimension for each channel c, is the feature of global pooling in the width dimension for each channel c, is the width of the feature map, is the height of the feature map, For the input feature map in channel ,Location The value at The expression of the direction-aware feature map generated by the shared convolution operation on the height feature and the width feature is: ; in, To perform convolution operation and The processed feature map, is the activation function; The expression for generating the attention weights in the height and width directions is: , ; in, The attention weights generated for the height direction, The attention weights generated for the width direction.

[0012] Preferably, the expression of the focal frequency loss function is: ; in, is the value of the frequency loss function, is the total number of frequency points, is the index of the frequency point, To indicate the frequency point The weight of To represent the real time-frequency diagram and the generated time-frequency diagram at the frequency point The distance between ; in, Used to control the rate at which weights are adjusted. Used to determine the weight adjustment direction of the frequency point; ; in, The frequency point is The frequency amplitude of the real time-frequency diagram at The frequency point is The frequency amplitude of the false time-frequency diagram.

[0013] The second embodiment of the present invention provides a device for classifying bearing faults based on artificial intelligence, comprising: A two-dimensional time-frequency diagram generating unit is used to obtain a bearing rotation signal collected by a sensor, and perform a continuous wavelet transform on the bearing rotation signal after preprocessing to generate a two-dimensional time-frequency diagram, wherein the two-dimensional time-frequency diagram can provide a frequency change of the signal at different time points; A key feature identification unit is used to call a pre-trained discriminator to perform sliding feature extraction on the two-dimensional time-frequency graph to generate height features and width features, and identify key features according to the height features and the width features through an attention mechanism, wherein the discriminator measures real data and false data based on a focal frequency loss function during training until the discriminator and the generator generating the false data reach a Nash equilibrium; A fault classification unit is used to convert the key features into a one-dimensional feature vector, and classify the key features through multiple convolutional layers and a fully connected layer, and output corresponding fault categories, wherein the fault categories include inner race fault, outer race fault, and rolling element fault.

[0014] The third embodiment of the present invention provides a downhill assist device for a pure electric vehicle, including a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement a method for classifying bearing faults based on artificial intelligence as described in any one of the above.

[0015] The fourth embodiment of the present invention provides a computer-readable storage medium, characterized in that a computer program is stored therein, and the computer program can be executed by a processor of a device where the computer-readable storage medium is located to implement a method for classifying bearing faults based on artificial intelligence as described in any one of the above items.

[0016] Based on the artificial intelligence-based classification method, device, equipment, and storage medium for bearing faults provided by the present invention, a bearing rotation signal collected by a sensor is obtained, and the bearing rotation signal is pre-processed and then subjected to continuous wavelet transform to generate a two-dimensional time-frequency diagram, wherein the two-dimensional time-frequency diagram can provide the frequency change of the signal at different time points; then, a pre-trained discriminator is called to extract sliding features from the two-dimensional time-frequency diagram to generate height features and width features, and key features are identified according to the height features and the width features through an attention mechanism; then, the loss of the key features in frequency is calculated through a frequency loss function, and the loss in frequency is used to classify the key features through multiple convolutional layers and fully connected layers, and the corresponding fault categories are output, wherein the fault categories include inner ring faults, outer ring faults, and rolling element faults. The problem of difficulty in capturing complex signal features in existing time domain or frequency domain analysis is solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of a method for classifying bearing faults based on artificial intelligence provided by the first embodiment of the present invention; Figure 2It is a module schematic diagram of a device for classifying bearing faults based on artificial intelligence provided by the second embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] In order to better understand the technical solution of the present invention, the embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0020] It should be clear that the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0022] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0023] The word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0024] The "first\second" mentioned in the embodiments is only to distinguish similar objects, and does not represent a specific order for the objects. It is understandable that the "first\second" can be interchanged with the specific order or sequence where permitted. It should be understood that the objects distinguished by "first\second" can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than those illustrated or described herein.

[0025] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0026] The present invention discloses a method, device, equipment, and storage medium for classifying bearing faults based on artificial intelligence, aiming to solve the problem that it is difficult to capture complex signal features in existing time domain or frequency domain analysis.

[0027] The first embodiment of the present invention provides a method for classifying bearing faults based on artificial intelligence, which can be performed by a bearing fault classification device (hereinafter referred to as the classification device), and in particular, by one or more processors in the classification device to implement at least the following steps: S101, obtaining a bearing rotation signal collected by a sensor, and performing a continuous wavelet transform on the bearing rotation signal after preprocessing to generate a two-dimensional time-frequency diagram, wherein the two-dimensional time-frequency diagram can provide frequency changes of the signal at different time points; In this embodiment, the classification device can be a terminal with data processing capabilities such as a server, workstation, desktop computer, laptop computer, desktop computer, etc., which can establish a communication connection with the sensor. The classification device can be installed with a corresponding operating system and application software, and the functions required by this embodiment can be realized through the combination of the operating system and application software.

[0028] It should be noted that, in this embodiment, the operating state of the bearing is monitored in real time by a sensor installed on the bearing to obtain its vibration signal under different working conditions. The collected signal is in the form of a one-dimensional time series, representing the instantaneous acceleration or speed change of the bearing during operation. However, the original signal usually contains noise or other interference information, so preprocessing is required before further analysis. The preprocessing includes denoising the signal (such as using a low-pass filter or other denoising algorithm) to remove high-frequency noise or interference signals from the external environment. In addition, the signal can also be normalized to ensure that the input data amplitude is within a certain range and reduce the impact of outliers on subsequent analysis. The preprocessed signal can more accurately reflect the actual operating state of the bearing.

[0029] Next, the preprocessed signal is subjected to a continuous wavelet transform, which can convert the one-dimensional time series data into a two-dimensional time-frequency diagram, which can simultaneously capture the frequency changes of the vibration signal at different time points. Compared with the traditional Fourier transform, the wavelet transform can provide better localization characteristics in the time domain and frequency domain, and is particularly suitable for processing non-stationary signals, such as instantaneous change signals in bearing faults; in this embodiment, the expression of the continuous wavelet transform is: ; in, is a two-dimensional time-frequency diagram, To control the scale of the wavelet function, For time shift, is the preprocessed one-dimensional bearing data, is the scale transform and translation transform of the wavelet function.

[0030] It should be noted that the scale parameter The function of is to adjust the frequency resolution and time resolution of the wavelet function. When is small, the wavelet function has a higher time resolution but a lower frequency resolution, which is suitable for capturing short-term high-frequency features in the signal; when When it is larger, the wavelet function has higher frequency resolution but lower time resolution, which is suitable for capturing long-term low-frequency features in the signal. Value, wavelet transform can extract the changing characteristics of the signal in different frequency bands.

[0031] After wavelet transform, the generated two-dimensional time-frequency diagram can intuitively show the changes in the bearing vibration signal in time and frequency. Under normal circumstances, the bearing vibration signal will fluctuate within a certain frequency range; when a fault occurs, abnormal high-frequency or low-frequency signals will appear in certain specific frequency bands. By analyzing these changes, early detection and accurate classification of bearing faults can be achieved.

[0032] S102, calling a pre-trained discriminator to perform sliding feature extraction on the two-dimensional time-frequency graph to generate height features and width features, and identifying key features according to the height features and the width features through an attention mechanism, wherein the discriminator measures real data and false data based on a focal frequency loss function during training until the discriminator and a generator generating the false data reach a Nash equilibrium; It should be noted that before using the discriminator, the discriminator must also be trained. The training process is as follows: First, the one-dimensional bearing vibration signal in the data set is preprocessed to ensure the quality of the input signal. Next, the preprocessed signal is converted into a two-dimensional time-frequency diagram through continuous wavelet transform (CWT). CWT can capture the time and frequency characteristics of the signal at the same time, so that the generated time-frequency diagram has information in both the time domain and the frequency domain.

[0033] During the training process of the generator, random noise is first generated as the input of the generator. The random noise is a set of random vectors, which the generator converts into a fake two-dimensional time-frequency diagram with the purpose of mimicking the real bearing fault signal. The goal of the generator is to deceive the discriminator by constantly adjusting its weights to make the generated fake data as close to the real data as possible. In each training iteration, the time-frequency diagram generated by the generator will be input into the discriminator together with the real time-frequency diagram (converted from the one-dimensional bearing data).

[0034] The discriminator extracts features from the input real data and false data through a convolutional neural network (CNN), and determines whether the input data is a real bearing signal. The discriminator is not only responsible for classifying real data and false data, but also needs to detect the feature differences of the input signal in the frequency domain. To achieve this, a focal frequency loss function is introduced to quantify the differences between the real time-frequency graph and the generated time-frequency graph at different frequency points. The focal frequency loss function dynamically adjusts the weight of each frequency point so that those frequency features that are difficult to generate are given greater weights, allowing the generator to focus on these features and improve the authenticity of the generated data.

[0035] Specifically, when calculating the frequency loss, the frequency features extracted from the real time-frequency graph are first compared with the frequency features extracted from the fake time-frequency graph generated by the generator. By comparing these differences at different frequency points, the gap between the generated data and the real data can be captured. The frequency loss function is used to measure this difference, and its expression is: The expression of the focal frequency loss function is: ; in, is the value of the frequency loss function, is the total number of frequency points, is the index of the frequency point, To indicate the frequency point The weight of To represent the real time-frequency diagram and the generated time-frequency diagram at the frequency point The model dynamically assigns weights based on the differences at different frequency points, focusing on those frequency points with large differences, thereby increasing the model's sensitivity to difficult-to-capture features. The adjustment rate of this weight is determined by the parameter Control; the formula is: ; in, Used to control the rate at which weights are adjusted. Used to determine the weight adjustment direction of the frequency point; In specific calculations, the frequency loss function measures the difference in frequency amplitude between the real time-frequency graph and the false time-frequency graph at each frequency point. Their difference is expressed by the following formula: ; in, The frequency point is The frequency amplitude of the real time-frequency diagram at The frequency point is The frequency amplitude of the false time-frequency diagram.

[0036] During the entire training process, the weights of the discriminator are updated through the back-propagation algorithm. In each iteration, the loss function is calculated based on the classification accuracy of the discriminator for real data and fake data, and the weights are updated through an optimizer (such as the Adam optimizer) to ensure that the discriminator can better identify the difference between real data and fake data. At the same time, the generator will also update its own weights based on the feedback results of the discriminator. The goal of the generator is to generate a more realistic time-frequency graph in each iteration, so that it gradually approaches the characteristics of the real data, and ultimately achieves the effect that the generated data cannot be distinguished by the discriminator.

[0037] In this embodiment, the input two-dimensional time-frequency graph is feature extracted by calling a pre-trained discriminator to generate height features and width features. The discriminator gradually captures local information in the time-frequency graph through the sliding feature extraction method in the convolutional neural network. In order to more effectively capture the global characteristics of the two-dimensional time-frequency graph in feature extraction, the discriminator uses a global pooling operation to process the spatial dimensions of the input feature graph and extract the height features and width features respectively. Through global pooling, the dimension of the feature graph is effectively reduced, key global information is retained, and the amount of data is reduced to avoid model overfitting.

[0038] In the global pooling process of the height dimension, the feature map is compressed in the column dimension to generate the height feature of each channel, and the formula is: , ; Among them, the is the globally pooled feature in the height dimension for each channel c, is the feature of global pooling in the width dimension for each channel c, is the width of the feature map, is the height of the feature map, For the input feature map in channel ,Location The value at After the height and width features are extracted, these features are then shared by the convolution operation. The shared convolution operation can capture the interdependence of the height and width directions and generate a feature map based on direction perception. By combining the height feature and the width feature, a direction-aware feature map is generated, which is expressed as: ; in, To perform convolution operation and The processed feature map, is a nonlinear activation function (such as ReLU); In order to strengthen the key features more effectively, the attention mechanism is adopted. It first generates the attention weights in the height and width directions. The attention weight in the height direction is generated by the convolution operation, and the expression is: , ; in, The attention weights generated for the height direction, The attention weights generated for the width direction.

[0039] The generated attention weights are applied to the original two-dimensional time-frequency graph. By multiplying the attention weights in the height and width directions with the original time-frequency graph element by element, important spatial positions can be effectively strengthened and irrelevant information can be suppressed. The expression after applying the attention weights is:

[0040] in, Feature map after applying attention weights; the features are further processed by the Hard-Swish activation function to generate the final key features, which are expressed as:

[0041] S103, converting the key features into one-dimensional feature vectors, and classifying the key features through multiple convolutional layers and fully connected layers, and outputting corresponding fault categories, wherein the fault categories include inner race faults, outer race faults, and rolling element faults.

[0042] It should be noted that when the time-frequency graph extracts key spatial and frequency features through the convolution layer and the pooling layer, these features will be converted into a one-dimensional vector. This step is to "flatten" the two-dimensional feature map and convert it into a one-dimensional feature vector. The flattened feature vector will be passed as input to the fully connected layer. The fully connected layer integrates the local features extracted previously and establishes a global connection between different features. Each node is connected to all nodes in the previous layer, which can convert the local representation of the feature into a global information aggregation to understand the global characteristics of the time-frequency graph.

[0043] The output of the fully connected layer is processed by an activation function (such as ReLU or Softmax). The role of the activation function is to introduce nonlinearity so that the model can capture more complex feature patterns. Preferably, the Softmax function can be used to map the output of the network to a probability distribution. The output of the fully connected layer is passed to the final classification layer of the model. Here, the model maps the input features to different fault categories. Each category (such as inner race fault, outer race fault, rolling element fault, or normal operation status) corresponds to an output node, and the model predicts the fault category of the input time-frequency graph through these output nodes.

[0044] The second embodiment of the present invention provides a device for classifying bearing faults based on artificial intelligence, comprising: A two-dimensional time-frequency diagram generating unit is used to obtain a bearing rotation signal collected by a sensor, and perform a continuous wavelet transform on the bearing rotation signal after preprocessing to generate a two-dimensional time-frequency diagram, wherein the two-dimensional time-frequency diagram can provide a frequency change of the signal at different time points; A key feature identification unit is used to call a pre-trained discriminator to perform sliding feature extraction on the two-dimensional time-frequency graph to generate height features and width features, and identify key features according to the height features and the width features through an attention mechanism, wherein the discriminator measures real data and false data based on a focal frequency loss function during training until the discriminator and the generator generating the false data reach a Nash equilibrium; A fault classification unit is used to convert the key features into a one-dimensional feature vector, and classify the key features through multiple convolutional layers and a fully connected layer, and output corresponding fault categories, wherein the fault categories include inner race fault, outer race fault, and rolling element fault.

[0045] The third embodiment of the present invention provides a classification device for bearing faults based on artificial intelligence, including a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement a downhill assist method for a pure electric vehicle as described in any one of the above.

[0046] The fourth embodiment of the present invention provides a computer-readable storage medium, characterized in that a computer program is stored therein, and the computer program can be executed by a processor of a device where the computer-readable storage medium is located to implement a method for classifying bearing faults based on artificial intelligence as described in any one of the above items.

[0047] Based on the artificial intelligence-based classification method, device, equipment, and storage medium for bearing faults provided by the present invention, a bearing rotation signal collected by a sensor is obtained, and the bearing rotation signal is pre-processed and then subjected to continuous wavelet transform to generate a two-dimensional time-frequency diagram, wherein the two-dimensional time-frequency diagram can provide the frequency change of the signal at different time points; then, a pre-trained discriminator is called to extract sliding features from the two-dimensional time-frequency diagram to generate height features and width features, and key features are identified according to the height features and the width features through an attention mechanism; then, the loss of the key features in frequency is calculated through a frequency loss function, and the loss in frequency is used to classify the key features through multiple convolutional layers and fully connected layers, and the corresponding fault categories are output, wherein the fault categories include inner ring faults, outer ring faults, and rolling element faults. The problem of difficulty in capturing complex signal features in existing time domain or frequency domain analysis is solved.

[0048] Exemplarily, the computer program described in the third and fourth embodiments of the present invention may be divided into one or more modules, and the one or more modules are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program in the device for implementing a classification of bearing faults based on artificial intelligence. For example, the device described in the second embodiment of the present invention.

[0049] The processor 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 gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the method for classifying bearing faults based on artificial intelligence, and uses various interfaces and lines to connect the various parts of the method for classifying bearing faults based on artificial intelligence.

[0050] The memory can be used to store the computer program and / or module, and the processor realizes various functions of a method for classifying bearing faults based on artificial intelligence by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, a text conversion function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, text message data, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0051] Wherein, if the implemented module 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 such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0052] It should be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without paying any creative effort.

[0053] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for classifying bearing faults based on artificial intelligence, characterized in that: include: Acquire the bearing rotation signal collected by the sensor, and perform continuous wavelet transform on the bearing rotation signal after preprocessing to generate a two-dimensional time-frequency diagram, wherein the two-dimensional time-frequency diagram can provide the frequency change of the signal at different time points; Calling a pre-trained discriminator to perform sliding feature extraction on the two-dimensional time-frequency graph to generate height features and width features, and identifying key features according to the height features and the width features through an attention mechanism, wherein the discriminator measures real data and false data based on a focal frequency loss function during training until the discriminator and a generator generating the false data reach a Nash equilibrium; The key features are converted into one-dimensional feature vectors, and the one-dimensional feature vectors are classified into the key features through multiple convolutional layers and fully connected layers, and the corresponding fault categories are output, wherein the fault categories include inner race faults, outer race faults, and rolling element faults.

2. The method for classifying bearing faults based on artificial intelligence according to claim 1 is characterized in that: The calling of the pre-trained discriminator performs sliding feature extraction on the two-dimensional time-frequency graph to generate height features and width features, and identifies key features according to the height features and the width features through an attention mechanism, specifically: Through two global pooling operations, the input two-dimensional time-frequency graph is processed in the spatial dimension to extract the height feature and width feature respectively; Performing a shared convolution operation on the height feature and the width feature to generate a direction-aware feature map, and generating attention weights in the height and width directions; The attention weight is applied to the two-dimensional time-frequency graph, and key features are generated through a Hard-Swish activation function.

3. The method for classifying bearing faults based on artificial intelligence according to claim 1 is characterized in that: The expression of the continuous wavelet transform is: ； in, is a two-dimensional time-frequency diagram, To control the scale of the wavelet function, For time shift, is the preprocessed one-dimensional bearing data, is the scale transform and translation transform of the wavelet function.

4. The method for classifying bearing faults based on artificial intelligence according to claim 1 is characterized in that: The training process of the discriminator includes: The one-dimensional bearing data in the data set is transformed into a two-dimensional time-frequency diagram as the real data by continuous wavelet transformation; Generate random noise, and send the random noise to a generator to generate false data; Inputting the real data and the false data into a discriminator, and identifying the difference between the real data and the false data in the frequency domain, wherein the difference in the frequency domain is measured by a focal frequency loss function, and updating the weight of the discriminator by back propagation and an optimizer; Based on the identification result of the discriminator, the weight of the generator is adjusted until the false data generated by the generator cannot be distinguished by the discriminator.

5. The method for classifying bearing faults based on artificial intelligence according to claim 2 is characterized in that: The expression of the global pooling operation is: , ; Among them, the is the globally pooled feature in the height dimension for each channel c, is the feature of global pooling in the width dimension for each channel c, is the width of the feature map, is the height of the feature map, For the input feature map in channel ,Location The value at The expression of the direction-aware feature map generated by the shared convolution operation on the height feature and the width feature is: ； in, To perform convolution operation and The processed feature map, is the activation function; The expression for generating the attention weights in the height and width directions is: , ; in, The attention weights generated for the height direction, The attention weights generated for the width direction.

6. The method for classifying bearing faults based on artificial intelligence according to claim 1 is characterized in that: The expression of the focal frequency loss function is: ； in, is the value of the frequency loss function, is the total number of frequency points, is the index of the frequency point, To indicate the frequency point The weight of To represent the real time-frequency diagram and the generated time-frequency diagram at the frequency point The distance between ； in, Used to control the rate at which weights are adjusted. Used to determine the weight adjustment direction of the frequency point; ; in, The frequency point is The frequency amplitude of the real time-frequency diagram at The frequency point is The frequency amplitude of the false time-frequency diagram.

7. A device for classifying bearing faults based on artificial intelligence, characterized in that: include: A two-dimensional time-frequency diagram generating unit is used to obtain a bearing rotation signal collected by a sensor, and perform a continuous wavelet transform on the bearing rotation signal after preprocessing to generate a two-dimensional time-frequency diagram, wherein the two-dimensional time-frequency diagram can provide a frequency change of the signal at different time points; A key feature identification unit is used to call a pre-trained discriminator to perform sliding feature extraction on the two-dimensional time-frequency graph to generate height features and width features, and identify key features according to the height features and the width features through an attention mechanism, wherein the discriminator measures real data and false data based on a focal frequency loss function during training until the discriminator and the generator generating the false data reach a Nash equilibrium; A fault classification unit is used to convert the key features into a one-dimensional feature vector, and classify the key features through multiple convolutional layers and a fully connected layer, and output corresponding fault categories, wherein the fault categories include inner race fault, outer race fault, and rolling element fault.

8. A device for classifying bearing faults based on artificial intelligence, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement a method for classifying bearing faults based on artificial intelligence as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: A computer program is stored, and the computer program can be executed by a processor of the device where the computer-readable storage medium is located to implement a method for classifying bearing faults based on artificial intelligence as described in any one of claims 1 to 6.

Images