Bearing fault detection method, training method, device, equipment, medium and product

Abstract

The application relates to a bearing fault detection method, a bearing fault detection model training method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: in response to a bearing fault detection instruction of a target bearing, acquiring bearing data and a target bearing fault detection model; then inputting the bearing data into the target bearing fault detection model; fusing bearing time domain features and bearing frequency domain features in the bearing data through the target bearing fault detection model; performing fault identification on the features obtained through the fusion processing; and obtaining a bearing fault category of the target bearing. The method can improve the accuracy of bearing fault detection results.

Description

Technical Field

[0001] The present application relates to the field of fault detection technology, and in particular to a bearing fault detection method, a training method, an apparatus, a computer device, a storage medium, and a computer program product for a bearing fault detection model. Background Art

[0002] Bearings are crucial components in mechanical equipment, supporting rotating bodies, reducing friction during movement, and ensuring rotational accuracy. Therefore, bearing fault detection is crucial to the safe operation of mechanical equipment.

[0003] In related technologies, when performing bearing fault detection, it is usually necessary to label the bearing data, then obtain a deep learning classification model based on the labeled data set, and finally detect the bearing through the deep learning classification model to obtain the bearing fault type.

[0004] However, the related art has the problem of low accuracy of fault detection results. Summary of the Invention

[0005] Based on this, it is necessary to provide a bearing fault detection method, a training method, device, computer equipment, storage medium and computer program product for a bearing fault detection model that can improve the accuracy of bearing fault detection results in response to the above technical problems.

[0006] In a first aspect, the present application provides a bearing fault detection method, the method comprising:

[0007] Responding to a fault detection instruction of a target bearing, acquiring bearing data and a target bearing fault detection model;

[0008] The bearing data is input into the target bearing fault detection model. The bearing time domain features and bearing frequency domain features in the bearing data are fused through the target bearing fault detection model. Fault identification is performed on the fused features to obtain the bearing fault category of the target bearing.

[0009] In one embodiment, the target bearing fault detection model includes a time domain feature extraction network and a frequency domain feature extraction network. The target bearing fault detection model is used to fuse the bearing time domain features and the bearing frequency domain features in the bearing data, including:

[0010] Inputting the bearing data into a time domain feature extraction network to obtain the bearing time domain features; and inputting the bearing data into a frequency domain feature extraction network to obtain the bearing frequency domain features;

[0011] The bearing time domain features and bearing frequency domain features are superimposed to obtain the fused features.

[0012] In one embodiment, bearing data is input into a frequency domain feature extraction network to obtain bearing frequency domain features, including:

[0013] Perform Fourier transform and radiometric transform on the bearing data to obtain the initial frequency domain characteristics of the bearing;

[0014] The initial frequency domain features of the bearing are filtered based on the spectrum intensity distribution in the initial frequency domain features of the bearing to obtain the bearing frequency domain features.

[0015] In one embodiment, the target bearing fault detection model further includes a data enhancement network. Before fusing the bearing time domain features and the bearing frequency domain features in the bearing data, the method further includes:

[0016] The bearing data is input into the data enhancement network, and the bearing data is standardized through the data enhancement network to obtain the bearing time series array. The bearing time series array is then subjected to feature fusion processing to obtain the processed bearing data.

[0017] In a second aspect, the present application provides a method for training a bearing fault detection model, the method comprising:

[0018] Obtaining an initial bearing fault detection model and a bearing fault data set; the initial bearing fault detection model includes an initial time domain feature extraction network and an initial frequency domain feature extraction network;

[0019] Based on the bearing fault dataset, the initial time domain feature extraction network and the initial frequency domain feature extraction network are trained to obtain the target bearing fault detection model; the target bearing fault detection model is used to fuse the bearing time domain features and bearing frequency domain features of the bearing data in the target bearing, and perform fault identification on the fused features to obtain the bearing fault category of the target bearing.

[0020] In one embodiment, the initial bearing fault detection model further includes an initial data augmentation network; an initial time domain feature extraction network and an initial frequency domain feature extraction network in the initial bearing fault detection model are trained based on the bearing fault dataset to obtain a target bearing fault detection model, including:

[0021] According to the bearing fault data set, the initial data enhancement network, the initial time domain feature extraction network and the initial frequency domain feature extraction network are trained simultaneously to obtain the data enhancement network, the time domain feature extraction network and the frequency domain feature extraction network;

[0022] Based on the data enhancement network, time domain feature extraction network and frequency domain feature extraction network, the target bearing fault detection model is obtained.

[0023] In one embodiment, a bearing fault data set includes bearing data samples; based on the bearing data samples, an initial data enhancement network, a time domain feature extraction network, and an initial frequency domain feature extraction network are simultaneously trained to obtain a data enhancement network, a time domain feature extraction network, and a frequency domain feature extraction network, including:

[0024] Input the bearing data samples into the initial data enhancement network to obtain the bearing standardized data;

[0025] Inputting the bearing normalized data into the initial time domain feature extraction network to obtain time domain loss, and inputting the bearing normalized data into the initial frequency domain feature extraction network to obtain frequency domain loss;

[0026] Based on the time domain loss and the frequency domain loss, the initial data enhancement network, the time domain feature extraction network and the initial frequency domain feature extraction network are trained to obtain the data enhancement network, the time domain feature extraction network and the frequency domain feature extraction network.

[0027] In one embodiment, the bearing data sample is input into the initial data augmentation network, including:

[0028] Store bearing data samples in a queue with a preset capacity and continuously update the bearing data samples in the queue;

[0029] The bearing data samples currently stored in the queue are input into the initial data augmentation network.

[0030] In one embodiment, the bearing normalized data is input into the initial time domain feature extraction network to obtain the time domain loss, including:

[0031] Different parameter updating methods are used to update the network parameters in the initial time domain feature extraction network, and two sets of parameter updating results are obtained. By comparing the two sets of parameter updating results, the time domain loss is obtained.

[0032] In one embodiment, the bearing normalized data is input into an initial frequency domain feature extraction network to obtain a frequency domain loss, including:

[0033] Different parameter updating methods are used to update the network parameters in the initial frequency domain feature extraction network, and two sets of parameter updating results are obtained. By comparing the two sets of parameter updating results, the frequency domain loss is obtained.

[0034] In one embodiment, a bearing fault dataset includes bearing test samples and bearing test labels; an initial bearing fault detection model also includes an initial fault classification network; and a target bearing fault detection model is obtained based on a data augmentation network, a time domain feature extraction network, and a frequency domain feature extraction network, including:

[0035] Input the bearing test sample into the initial bearing fault detection model to obtain the bearing test result;

[0036] Based on the loss value between the bearing test results and the bearing test labels, the parameters in the initial fault classification network are trained while keeping the parameters of the data augmentation network, the time domain feature extraction network, and the frequency domain feature extraction network unchanged to obtain the target bearing fault detection model.

[0037] In a third aspect, the present application further provides a bearing fault detection device, comprising:

[0038] An instruction response module, configured to respond to a fault detection instruction of a target bearing and obtain bearing data and a target bearing fault detection model;

[0039] The fault identification module is used to input the bearing data into the target bearing fault detection model, fuse the bearing time domain features and bearing frequency domain features in the bearing data through the target bearing fault detection model, and perform fault identification on the features obtained by the fusion processing to obtain the bearing fault category of the target bearing.

[0040] In a fourth aspect, the present application further provides a training device for a bearing fault detection model, comprising:

[0041] A data acquisition module is used to obtain an initial bearing fault detection model and a bearing fault data set; the initial bearing fault detection model includes an initial time domain feature extraction network and an initial frequency domain feature extraction network;

[0042] The model training module is used to train the initial time domain feature extraction network and the initial frequency domain feature extraction network according to the bearing fault data set to obtain the target bearing fault detection model; the target bearing fault detection model is used to fuse the bearing time domain features and bearing frequency domain features of the bearing data in the target bearing, and perform fault identification on the features obtained by the fusion processing to obtain the bearing fault category of the target bearing.

[0043] In a fifth aspect, the present application also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method in any one of the embodiments of the first or second aspect are implemented.

[0044] In a sixth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method in any one of the embodiments of the first or second aspect above.

[0045] In a seventh aspect, the present application also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the method in any one of the embodiments of the first or second aspect above.

[0046] The above-mentioned bearing fault detection method, apparatus, computer equipment, storage medium, and computer program product, in response to a fault detection instruction for a target bearing, obtain bearing data and a target bearing fault detection model, and input the bearing data into the target bearing fault detection model. The target bearing fault detection model is used to fuse the bearing time domain features and the bearing frequency domain features in the bearing data, and fault identification is performed on the features obtained by the fusion process to obtain the bearing fault category of the target bearing. In this method, the target bearing fault detection model is used to fuse the bearing time domain features and the bearing frequency domain features in the bearing data, and fault identification is performed on the features obtained by the fusion process to obtain the bearing fault category of the target bearing. This is equivalent to comprehensively considering the characteristics of the bearing data in the time domain and frequency domain when performing fault identification on the bearing data to characterize more comprehensive and accurate bearing status information. Furthermore, the bearing fault category determined based on the complete and comprehensive bearing status information is also more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, 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 paying any creative work.

[0048] Figure 1 is a diagram of the internal structure of a computer device in one embodiment;

[0049] Figure 2 1 is a flow chart of a bearing fault detection method according to an embodiment;

[0050] Figure 3 Schematic diagram of a feature fusion process in one embodiment;

[0051] Figure 4 Schematic diagram of a flow chart of a frequency domain feature extraction step in one embodiment;

[0052] Figure 5 is a flow chart of a bearing fault detection method according to another embodiment;

[0053] Figure 6 1 is a flow chart of a method for training a bearing fault detection model in one embodiment;

[0054] Figure 7Schematic diagram of the architecture of a target bearing fault detection model in one embodiment;

[0055] Figure 8 Schematic diagram of a flow chart of steps for obtaining a target bearing fault detection model in one embodiment;

[0056] Figure 9 A schematic flow chart of steps for acquiring a target bearing fault detection model in another embodiment;

[0057] Figure 10 A schematic flow chart of a feature extraction network training step in another embodiment;

[0058] Figure 11 Schematic diagram of a flow chart of comparative learning training steps in one embodiment;

[0059] Figure 12 Schematic diagram of a process of updating queue samples in one embodiment;

[0060] Figure 13 Schematic diagram of a flow chart of a method for training a bearing fault detection model in another embodiment;

[0061] Figure 14 Schematic diagram of the architecture of a feature extraction network in one embodiment;

[0062] Figure 15 is a structural block diagram of a bearing fault detection device in one embodiment;

[0063] Figure 16 4 is a structural block diagram of a training device for a bearing fault detection model in one embodiment. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0065] The bearing fault detection method or bearing fault detection model training method provided in the embodiment of the present application can be applied to a computer device. The computer device can be a server, and its internal structure diagram can be as follows: Figure 1As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store fault detection data or model training data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a bearing fault detection method or a bearing fault detection model training method is implemented.

[0066] Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0067] Bearings are a crucial component of mechanical equipment, supporting rotating parts and reducing friction during movement. Therefore, bearing fault detection is crucial for the safe operation of mechanical equipment.

[0068] In the related art, when performing bearing fault detection, there are two methods, including supervised learning and unsupervised learning.

[0069] Supervised learning requires labeling bearing data, then training a deep learning classification model based on the labeled dataset. Finally, the deep learning classification model is used to detect the bearings and determine the bearing fault type. However, in the specific field of bearing fault detection, generating bearing fault or abnormal data is expensive, standardization is difficult, and the dataset size is small, making it difficult to fully train the deep learning classification model, resulting in low classification accuracy.

[0070] Unsupervised learning, typically used to learn representations of unlabeled time series data, often focuses too much on the relationships between data samples within the time domain while ignoring the characteristics of time series data in other domains. Furthermore, unsupervised learning methods are suitable for detecting stationary data. For non-stationary data like bearing fault data, feature extraction often leads to overfitting, resulting in poor model robustness and affecting the accuracy of model fault identification.

[0071] In summary, considering the low accuracy of fault detection results based on bearing data in related technologies, this application proposes inputting the bearing data into a target bearing fault detection model, fusing the bearing time domain features and bearing frequency domain features in the bearing data through the target bearing fault detection model, and performing fault identification on the fused features to obtain the bearing fault category of the target bearing. This is equivalent to comprehensively considering the characteristics of the bearing data in the time domain and frequency domain when performing fault identification on the bearing data to characterize more comprehensive and accurate bearing data characteristics. Furthermore, the bearing fault category determined based on the bearing data characteristics is also more accurate.

[0072] In an exemplary embodiment, Figure 2 As shown, a bearing fault detection method is provided, the method comprising:

[0073] S201 , in response to a fault detection instruction of a target bearing, acquiring bearing data and a target bearing fault detection model.

[0074] The target bearing can be any working bearing currently running on the mechanical equipment, or a historical bearing that has run on the mechanical equipment before.

[0075] The target bearing fault detection instruction is used to instruct the computer device to execute the bearing fault detection step, specifically to determine whether the target bearing has a fault, and if the target bearing has a fault, further determine the bearing fault category of the target bearing.

[0076] The fault detection instruction can be triggered automatically or manually when a bearing fault is detected, that is, when a target bearing is determined; it can also be triggered during the operation of the mechanical equipment after a designated target bearing on the mechanical equipment is determined as the target bearing.

[0077] In an embodiment of the present application, in response to a fault detection instruction of a target bearing, operating data of the target bearing is collected as bearing data, and a target bearing fault detection model is determined from a database.

[0078] For example, if the target bearing is a confirmed faulty bearing, the fault detection instruction carries bearing data. In this case, the computer device can directly extract the bearing data from the fault detection instruction in response to the fault detection instruction.

[0079] For example, if the target bearing is currently operating and the fault detection instruction includes a data collection time, the computer device, in response to the fault detection instruction, collects operating data for a preset time period starting at the current moment as bearing data. The computer device can be directly connected to the device where the target bearing is located via a cable to collect bearing data, or it can be connected to the mechanical equipment to which the target bearing belongs via an intelligent collection device to obtain bearing data collected by the intelligent collection device.

[0080] The target bearing fault detection model is a trained deep learning classification model that is suitable for fault detection of different types of target bearings and is reusable. In the embodiment of the present application, the input of the target bearing fault detection model is bearing data, and the output is the bearing fault type.

[0081] S202: Input the bearing data into the target bearing fault detection model, fuse the bearing time domain features and bearing frequency domain features in the bearing data through the target bearing fault detection model, and perform fault identification on the features obtained by the fusion process to obtain the bearing fault category of the target bearing.

[0082] After determining the bearing data and the target bearing fault detection model, the bearing data is used as the input of the target bearing fault detection model. The target bearing fault detection model is used to extract the bearing time domain features and the bearing frequency domain features of the bearing data, respectively obtaining the bearing time domain features and the bearing frequency domain features. The bearing time domain features and the bearing frequency domain features are then fused. This fused feature covers the characteristics of the bearing data in both the time domain and the frequency domain, enabling a more comprehensive and complete characterization of the bearing data. On this basis, fault identification is performed on the fused features, that is, identification is performed by integrating the characteristics of the bearing data in multiple dimensions to obtain a more accurate bearing fault classification result.

[0083] In an embodiment of the present application, in response to a fault detection instruction for a target bearing, bearing data and a target bearing fault detection model are acquired, and the bearing data is input into the target bearing fault detection model. The target bearing fault detection model is used to fuse the bearing time domain features and the bearing frequency domain features in the bearing data, and fault identification is performed on the features obtained by the fusion process to obtain the bearing fault category of the target bearing. In this method, the target bearing fault detection model is used to fuse the bearing time domain features and the bearing frequency domain features in the bearing data, and fault identification is performed on the features obtained by the fusion process to obtain the bearing fault category of the target bearing. This is equivalent to comprehensively considering the characteristics of the bearing data in the time domain and frequency domain when performing fault identification on the bearing data to characterize more comprehensive and accurate bearing status information. Furthermore, the bearing fault category result determined based on the complete and comprehensive bearing status information is also more accurate.

[0084] As can be seen from the preceding examples, bearing fault classification is based on features derived from fusing bearing time series features with bearing frequency domain features. Based on this, the following example further illustrates the steps for obtaining and fusing bearing time series and frequency domain features.

[0085] In an exemplary embodiment, Figure 3 As shown in FIG, the target bearing fault detection model includes a time domain feature extraction network and a frequency domain feature extraction network. The target bearing fault detection model is used to fuse the bearing time domain features and the bearing frequency domain features in the bearing data, including:

[0086] S301, inputting bearing data into a time domain feature extraction network to obtain bearing time domain features; and inputting bearing data into a frequency domain feature extraction network to obtain bearing frequency domain features.

[0087] The target bearing fault detection model includes two feature extraction networks in parallel: a time domain feature extraction network and a frequency domain feature extraction network.

[0088] Among them, the time domain feature extraction network includes multiple one-dimensional convolution layers and pooling layers, which are used to extract the time series features of bearing data in the time domain, namely the bearing time domain features.

[0089] It should be noted that, considering that time-domain feature extraction is very sensitive to the size of the one-dimensional convolution kernel, in order to avoid overfitting or underfitting of the time-domain feature extraction network, the convolution kernels in the time-domain feature extraction network are of different sizes, that is, different sizes of convolution kernels are set in the time-domain feature extraction network. The pooling layer can be an average pooling layer, which is used to pool the convolution features along the convolution kernel dimension and the time dimension. While retaining important feature information, some unimportant details are discarded, and the spatial dimension of the features output by the convolution layer is reduced, thereby achieving feature dimensionality reduction. This is shown in the following formula:

[0090]

[0091] In the above formula, M is the number of convolution kernels, L is the sequence length of the sample, and X i is the i-th feature output by the convolution module, and T is the sequential time dimension of the bearing data.

[0092] Among them, the frequency domain feature extraction network includes frequency domain features and frequency domain filtering features, which are used to extract the time-invariant components of bearing data in the frequency domain, namely, bearing frequency domain features.

[0093] The bearing data is input into the frequency domain feature extraction network. The frequency domain features of the bearing data are first extracted, and then the frequency domain features are differentiated to obtain time-varying components and time-invariant components. The time-invariant components are then determined as the bearing frequency domain features.

[0094] S302: Superimpose the bearing time domain features and the bearing frequency domain features to obtain fused features.

[0095] The bearing time domain features and the bearing frequency domain features have the same size. Based on this, the bearing time domain features can be superimposed with the features at the corresponding positions in the bearing frequency domain features according to the feature positions to obtain the fused features.

[0096] In this embodiment of the present application, bearing data is input into a time-domain feature extraction network to obtain bearing time-domain features; and the bearing data is input into a frequency-domain feature extraction network to obtain bearing frequency-domain features. The time-domain and frequency-domain features are then superimposed to produce fused features, fully and accurately characterizing the bearing data. Furthermore, by processing the bearing data in parallel using two networks (the time-domain feature extraction network and the bearing frequency-domain feature extraction network) to obtain both time-domain and frequency-domain features, feature extraction time can be shortened to a certain extent, improving fault detection efficiency.

[0097] The following describes the steps for obtaining the frequency domain characteristics of the bearing. In an exemplary embodiment, Figure 4 As shown in Figure 1, the bearing data is input into the frequency domain feature extraction network to obtain the bearing frequency domain features, including:

[0098] S401, performing Fourier transform processing and radiometric transform processing on the bearing data in sequence to obtain initial frequency domain characteristics of the bearing.

[0099] First, a discrete Fourier transform (DFT) is performed on the bearing data to transform the time domain signal samples into discrete time Fourier transform frequency domain samples to separate the frequency domain features in the bearing data. The DFT calculation formula is as follows:

[0100]

[0101] In the above formula, i is the i-th sample in the sample set, k is the k-th frequency domain component, c is the number of channels, L is the sample length, and j is the complex operator.

[0102] In order to obtain frequency domain features that can discriminate different fault stages and categories, the amplitude of the bearing data in the frequency domain is further obtained. The calculation method is as follows:

[0103]

[0104] In the above formula, A i is the amplitude of the frequency domain feature, R e is the real part of the frequency domain feature, I mis the imaginary part of the frequency domain feature, the i-th sample in the i-sample set, k is the k-th frequency domain component, and C is the number of channels.

[0105] Furthermore, in order to enable the frequency domain features between different channels to interact, a self-attention mechanism is used to design a learnable affine transformation layer. The implementation is as follows:

[0106]

[0107]

[0108] In the above formula, T1 and T2 are two parameters of the affine transformation layer, F is the frequency domain length, i is the i-th sample in the sample set, k is the k-th frequency domain component, c is the number of channels, C is the total number of channels, and einsum() is the affine transformation layer operation.

[0109] S402 : Filtering the initial frequency domain features of the bearing based on the spectrum intensity distribution in the initial frequency domain features of the bearing to obtain the frequency domain features of the bearing.

[0110] It should be noted that the initial frequency domain characteristics of the bearing obtained after discrete Fourier transform and radioactive transform processing are generated by taking into account the decomposable characteristics of non-stationary data in the frequency domain. This feature can already characterize the bearing characteristics under complex fault scenarios to a certain extent.

[0111] In actual scenarios, based on the decomposable characteristics of frequency domain features, the time-invariant features after decomposition, the time-varying features after decomposition, and the frequency domain features before decomposition are identified respectively to obtain corresponding identification results. Through a large number of experimental verifications, it is concluded that the time-invariant features can best reflect the characteristics of bearing faults.

[0112] Based on this, for the bearing fault diagnosis task, the embodiment of the present application also needs to extract the time-invariant components of the data of each specific stage and specific fault category, and filter them based on the spectral intensity distribution to select the frequency set with high intensity in each stage as the main frequency feature of that stage.

[0113] The specific process is as follows:

[0114] Based on the data set of each fault category, the spectral distribution intensity of its amplitude is calculated, and then it is arranged according to the intensity to determine the globally shared spectrum. Average pooling is performed along the frequency domain dimension, and the frequency domain features are filtered to obtain the bearing frequency domain features.

[0115] The features of the data set whose spectral distribution intensity is greater than the spectrum are averaged to obtain the filtered bearing frequency domain features.

[0116] In an embodiment of the present application, the bearing data is sequentially subjected to Fourier transform processing and radiometric transform processing to obtain the initial frequency domain characteristics of the bearing, and the initial frequency domain characteristics of the bearing are filtered based on the spectral intensity distribution in the initial frequency domain characteristics of the bearing to obtain the frequency domain characteristics of the bearing, which is equivalent to extracting the time-invariant components in the bearing data and filtering the time-varying components in the bearing data to improve the effectiveness of the bearing frequency domain characteristics.

[0117] Before extracting features from the bearing data, the bearing data can also be preprocessed, such as by normalizing the scale and increasing the receptive field, to capture the long-term variability of the target bearing data and improve the accuracy of feature extraction. In an exemplary embodiment, the target bearing fault detection model also includes a data enhancement network. Before fusing the bearing time domain features and bearing frequency domain features in the bearing data, the method further includes:

[0118] The bearing data is input into the data enhancement network, and the bearing data is standardized through the data enhancement network to obtain the bearing time series array. The bearing time series array is then subjected to feature fusion processing to obtain the processed bearing data.

[0119] For the bearing data generated by the system operation, the scale differences are eliminated through the standardization module in the data enhancement network. Then, the dilated convolution module is used to perform feature fusion extraction on the bearing data to effectively capture the long-term change characteristics of the data and expand the receptive field without increasing the number of network parameters.

[0120] Each dilated convolution module contains two dilated convolution layers and a GeLU activation function. At the same time, the input signal of the convolution module is fed forward to the output end while passing through the convolution layer operation, and is added to the convolution layer operation result as the final output of the module. This residual structure can effectively alleviate the gradient vanishing problem of deep neural networks and improve the convergence effect of network training.

[0121] In an embodiment of the present application, the bearing data is input into a data enhancement network, the bearing data is standardized by the data enhancement network to obtain a bearing time series array, and the bearing time series array is subjected to feature fusion processing to obtain processed bearing data, so as to eliminate scale differences through standardization, increase the receptive field without increasing the number of network parameters, and improve the effectiveness of subsequent feature extraction.

[0122] In an exemplary embodiment, Figure 5 As shown, a bearing fault detection method is provided, the method comprising:

[0123] S501 , in response to a fault detection instruction of a target bearing, acquiring bearing data and a target bearing fault detection model.

[0124] S502: Input the bearing data into the data enhancement network, perform standardization processing on the bearing data through the data enhancement network to obtain a bearing time series array, and perform feature fusion processing on the bearing time series array to obtain processed bearing data.

[0125] S503: Input the bearing data into the time domain feature extraction network to obtain the bearing time domain features.

[0126] S504: Perform Fourier transform and radiometric transform processing on the bearing data in sequence to obtain initial frequency domain characteristics of the bearing.

[0127] S505 , filtering the initial frequency domain features of the bearing based on the spectrum intensity distribution in the initial frequency domain features of the bearing to obtain the frequency domain features of the bearing.

[0128] S506: Superimpose the bearing time domain features and the bearing frequency domain features to obtain features obtained by fusion processing.

[0129] S507: Perform fault identification on the features obtained through the fusion process to obtain a bearing fault category of the target bearing.

[0130] In an embodiment of the present application, the bearing time domain features and the bearing frequency domain features in the bearing data are fused through a target bearing fault detection model, and fault identification is performed on the features obtained by the fusion processing to obtain the bearing fault category of the target bearing. This is equivalent to comprehensively considering the characteristics of the bearing data in the time domain and frequency domain when performing fault identification on the bearing data to characterize more comprehensive and accurate bearing status information. Furthermore, the bearing fault category result determined based on the complete and comprehensive bearing status information is also more accurate.

[0131] In an exemplary embodiment, Figure 6 As shown, a training method for a bearing fault detection model is also provided, the method comprising:

[0132] S601, obtaining an initial bearing fault detection model and a bearing fault data set; the initial bearing fault detection model includes an initial time domain feature extraction network and an initial frequency domain feature extraction network.

[0133] The initial bearing fault detection model includes an initial time domain feature extraction network and an initial frequency domain feature extraction network, and the network parameters of the initial bearing fault detection model are all initialization parameters or empirical parameters.

[0134] It should be known that, in addition to the above-mentioned initial time domain feature extraction network and initial frequency domain feature extraction network, the initial bearing fault detection model may also include other sub-networks, such as an initial feature enhancement network, an initial classification network, etc.

[0135] In an embodiment of the present application, the bearing fault data set may include a large number of unlabeled bearing data samples and a small number of labeled bearing data sets. Of course, if the bearing data generation cost and labeling cost are controllable, the bearing fault data sets may all be labeled bearing data sets.

[0136] S602: Based on the bearing fault data set, the initial time domain feature extraction network and the initial frequency domain feature extraction network are trained to obtain a target bearing fault detection model; the target bearing fault detection model is used to fuse the bearing time domain features and the bearing frequency domain features of the bearing data in the target bearing, and perform fault identification on the features obtained by the fusion processing to obtain the bearing fault category of the target bearing.

[0137] In the embodiment of the present application, the initial time domain feature extraction network and the initial frequency domain features can be obtained by training in an unsupervised learning manner.

[0138] In one scenario, unlabeled bearing data samples can be used to train some subnetworks in the initial bearing fault detection model. Labeled bearing datasets can then be used to train the remaining subnetworks in the initial bearing fault detection model to obtain the target bearing fault detection model. This phased model training approach can ensure model detection accuracy while reducing the cost of bearing data annotation.

[0139] In another scenario, a labeled bearing dataset can also be used to train all subnetworks in the initial bearing fault detection model until the training is completed to obtain the target bearing fault detection model.

[0140] See Figure 7 , Figure 7 The schematic diagram of the target bearing fault detection model is shown in Figure 2. Figure 7 It can be seen that the target bearing fault detection model includes a data augmentation network, a time domain feature extraction network, a frequency domain feature extraction network, and a classification layer network. The bearing data is input to the data augmentation network, and the output of the data augmentation network is simultaneously input to the time domain feature extraction network and the frequency domain feature extraction network. The output of the time domain feature extraction network and the output of the frequency domain feature extraction network are fused and input to the classification layer network. The classification layer network outputs the fault classification result and obtains the bearing fault category of the bearing data.

[0141] After obtaining the target bearing fault detection model, it can be stored in the computer's memory, allowing the computer to subsequently call the model and perform the bearing fault detection steps at any time. The specific calling process and beneficial effects are consistent with the specific embodiment of the bearing fault detection method described above and will not be further described here. For details, please refer to the aforementioned embodiment of the bearing fault detection method.

[0142] In an embodiment of the present application, an initial bearing fault detection model and a bearing fault data set are obtained, and based on the bearing fault data set, the initial time domain feature extraction network and the initial frequency domain feature extraction network in the initial bearing fault detection model are trained to obtain a target bearing fault detection model. The target bearing fault detection model is used to fuse the bearing time domain features and the bearing frequency domain features of the bearing data in the target bearing, and perform fault identification on the features obtained by the fusion process to obtain the bearing fault category of the target bearing. In this method, the initial time domain feature extraction network and the initial frequency domain feature extraction network in the initial bearing fault detection model are trained, so that the target bearing detection model obtained includes the trained time domain feature extraction network and the frequency domain feature extraction network, so that when the target bearing is subjected to fault identification, the bearing time domain features and the bearing frequency domain features of the bearing data can be fused, so that the processed state information can be more comprehensive and accurate, thereby improving the accuracy of the final determined bearing fault category result.

[0143] Next, the training process of the target bearing fault monitoring model is further explained.

[0144] In an exemplary embodiment, Figure 8 As shown in FIG, the initial bearing fault detection model also includes an initial data enhancement network. The initial time domain feature extraction network and the initial frequency domain feature extraction network in the initial bearing fault detection model are trained according to the bearing fault dataset to obtain a target bearing fault detection model, including:

[0145] S801, based on the bearing fault data set, the initial data enhancement network, the initial time domain feature extraction network and the initial frequency domain feature extraction network are trained simultaneously to obtain the data enhancement network, the time domain feature extraction network and the frequency domain feature extraction network.

[0146] Among them, the initial data enhancement network is connected to the initial time domain feature extraction network and the initial frequency domain feature extraction network respectively, that is, the output of the initial data enhancement network is also the input of the initial time domain feature extraction network and the input of the initial frequency domain feature extraction network. Furthermore, the output of the initial time domain feature extraction network and the output of the initial frequency domain feature extraction network are fused to form the initial feature extraction network.

[0147] The bearing fault dataset is input into the initial feature extraction network in batches to update the parameters of the initial feature extraction network, and a data enhancement network, a time domain feature extraction network and a frequency domain feature extraction network are obtained.

[0148] S802: Obtain a target bearing fault detection model based on the data enhancement network, the time domain feature extraction network, and the frequency domain feature extraction network.

[0149] It should be emphasized again that in the initial bearing fault detection model, in addition to the initial time domain feature extraction network and the initial frequency domain feature extraction network, other sub-networks can also be included, such as the initial feature enhancement network, the initial fault classification network, etc.

[0150] Taking the initial bearing fault detection model that also includes the initial fault classification network as an example, it can be trained in stages. For example, the unlabeled bearing data samples in the bearing fault data set are used to train the initial data enhancement network, the initial time domain feature extraction network and the initial frequency domain feature extraction network to obtain the data enhancement network, the time domain feature extraction network and the frequency domain feature extraction network; the labeled bearing data in the bearing fault data set are used to train the initial classification network to obtain the classification network, and then the data enhancement network, the time domain feature extraction network, the frequency domain feature extraction network and the classification network are spliced ​​to obtain the target bearing fault detection model.

[0151] In one scenario, after the data augmentation network, time series feature extraction network, and frequency domain feature extraction network have been trained, the trained network and the initial classification network can be spliced ​​together. The trained network parameters can then be frozen, and the spliced ​​network can be trained using a bearing fault dataset to adjust the network parameters in the initial classification network until the training is completed, thereby obtaining the target bearing fault detection model.

[0152] In an exemplary embodiment, Figure 9 As shown in the figure, the bearing fault dataset includes bearing test samples and bearing test labels; the initial bearing fault detection model also includes an initial fault classification network; based on the data enhancement network, the time domain feature extraction network, and the frequency domain feature extraction network, the target bearing fault detection model is obtained, including:

[0153] S901: Input the bearing test sample into the initial bearing fault detection model to obtain the bearing test result.

[0154] The bearing test samples are input into the initial bearing fault detection model. The data enhancement network performs data enhancement processing on the bearing test samples, the time domain feature extraction network extracts the time domain features of the bearing test samples after data enhancement, the frequency domain feature extraction network extracts the frequency domain features of the bearing test samples after data enhancement, and the initial fault classification network classifies the features of the fused time domain features and frequency domain features to obtain the bearing test results.

[0155] S902, based on the loss value between the bearing test results and the bearing test labels, while keeping the parameters of the data enhancement network, the time domain feature extraction network and the frequency domain feature extraction network unchanged, the parameters in the initial fault classification network are trained to obtain the target bearing fault detection model.

[0156] Among them, the function expression of the loss value is as follows:

[0157]

[0158] Where, L cls is the classification loss, N label is the number of bearing test labels, yi is the bearing test label of the i-th data, that is, the true label, p i is the bearing test result of the bearing test sample, that is, the predicted label.

[0159] Based on the bearing test results and bearing test labels, the loss value of the initial bearing fault monitoring model is calculated. While keeping the parameters of the data enhancement network, time domain feature extraction network, and frequency domain feature extraction network unchanged, a multi-round iterative training method is used to adjust the parameters in the initial fault classification network until the training is completed to obtain the target bearing fault detection model.

[0160] In this embodiment, a phased training approach is used to train the bearing fault detection model. In the first phase, unlabeled bearing data samples are used to train the initial data augmentation network, initial time-domain feature extraction network, and initial frequency-domain feature extraction network in the initial bearing fault detection model. In the second phase, labeled bearing test samples and bearing test labels are used to train the classification model in the initial bearing fault detection model. This training approach requires only a small number of sample labels, significantly reducing model training costs.

[0161] In an embodiment of the present application, the initial data enhancement network, the initial time domain feature extraction network and the initial frequency domain feature extraction network are trained simultaneously, and the data enhancement network, the time domain feature extraction network and the frequency domain feature extraction network are obtained at one time, which can improve the training time of the network and improve the training efficiency.

[0162] The following describes the training process of the data enhancement network, the time domain feature extraction network, and the frequency domain feature extraction network in the aforementioned embodiment through an embodiment.

[0163] In an exemplary embodiment, Figure 10 As shown, the bearing fault data set includes bearing data samples; based on the bearing data samples, the initial data enhancement network, the initial time domain feature extraction network, and the initial frequency domain feature extraction network are trained simultaneously to obtain the data enhancement network, the time domain feature extraction network, and the frequency domain feature extraction network, including:

[0164] S1001: Input the bearing data sample into the initial data enhancement network to obtain the bearing standardized data.

[0165] Among them, the initial data enhancement network includes an expanded convolution module, which is used to standardize and fuse the bearing data samples to enhance the effectiveness of the bearing data samples and obtain standardized bearing data.

[0166] S1002, inputting the bearing standardized data into the initial time domain feature extraction network to obtain time domain loss, and inputting the bearing standardized data into the initial frequency domain feature extraction network to obtain frequency domain loss.

[0167] The bearing standardized data is input into the initial time domain feature extraction network, the time domain features of the bearing standardized data are extracted through the initial time domain feature extraction network, and the loss between the time domain features and the bearing standardized data is calculated to obtain the time domain loss.

[0168] The bearing standardized data is input into the initial frequency domain feature extraction network, the frequency domain features of the bearing standardized data are extracted through the initial frequency domain feature extraction network, and the loss between the frequency domain features and the bearing standardized data is calculated to obtain the frequency domain loss.

[0169] In another scenario, a contrastive learning training strategy can also be used to drive the initial time domain feature extraction network / initial frequency domain feature extraction network to learn good features or representations by comparing the similarities or differences between data samples. Figure 11 As shown, specifically, different parameter updating methods are used to update the parameters in the network to obtain the network loss. Figure 11 In the

[15] , the feature extraction module is used to extract the features of the bearing data; the projection head module is used to map the features obtained by the feature extraction module to the contrast loss space.

[0170] Exemplarily, different parameter updating methods are used to update the network parameters in the initial time domain feature extraction network to obtain two sets of parameter updating results. By comparing the two sets of parameter updating results, the time domain loss is obtained.

[0171] For example, the gradient update method is used to update the parameters of the initial time domain feature extraction network to obtain the first loss, and the momentum update method is used to update the parameters of the initial time domain feature extraction network to obtain the second loss. The formula for momentum update is expressed as follows:

[0172]

[0173] In the above formula, is the parameter updated at time (t+1), m is the number of convolution kernels, q is the network parameter of the first branch, and k is the network parameter of the second branch.

[0174] Then the first loss and the second loss are mapped to the contrast loss space to obtain the contrast cross entropy loss, that is, the time domain loss. Temp The expression is as follows:

[0175]

[0176] In the above formula, N is the total number of samples, q i 、k i represents a positive sample pair, q j 、k j represents the negative sample pair.

[0177] Exemplarily, different parameter updating methods are used to update the network parameters in the initial frequency domain feature extraction network to obtain two sets of parameter updating results. The frequency domain loss is obtained by comparing the two sets of parameter updating results.

[0178] For example, the parameters of the initial frequency domain feature extraction network are updated by gradient update to obtain the third loss, and the parameters of the initial frequency domain feature extraction network are updated by momentum update to obtain the fourth loss. Then, the third loss and the fourth loss are mapped to the contrast loss space to obtain the contrast cross entropy loss, that is, the frequency domain loss.

[0179] S1003, based on the time domain loss and the frequency domain loss, the initial data enhancement network, the time domain feature extraction network and the initial frequency domain feature extraction network are trained to obtain the data enhancement network, the time domain feature extraction network and the frequency domain feature extraction network.

[0180] The time domain loss and frequency domain loss are superimposed as the comparative cross entropy loss, and the bearing sample data are used to train the initial data enhancement network, time domain feature extraction network and initial frequency domain feature extraction network in batches until the overall loss meets the loss threshold. The training is completed and the data enhancement network, time domain feature extraction network and frequency domain feature extraction network are obtained.

[0181] In an embodiment of the present application, the bearing data sample is input into the initial data enhancement network and subjected to standardization processing by the initial data enhancement network to facilitate the subsequent extraction of time domain features and frequency domain features. Then, the standardized bearing data sample is simultaneously input into the initial time domain feature extraction network and the initial frequency domain feature extraction network to obtain the time domain features and frequency domain features of the bearing data sample, and then the time domain loss and frequency domain loss are obtained to update the parameters of the above-mentioned initial data enhancement network, time domain feature extraction network and initial frequency domain feature extraction network to obtain the data enhancement network, time domain feature extraction network and frequency domain feature extraction network.

[0182] As can be seen from the above embodiments, the initial data enhancement network, the initial time domain feature extraction network, and the initial frequency domain feature extraction network are trained using unlabeled bearing data samples. Based on this, multiple bearing data samples in the bearing fault data set can be input into the above initial networks in batches for training. In an exemplary embodiment, Figure 12 As shown, the bearing data samples are input into the initial data augmentation network, including:

[0183] S1201, storing bearing data samples in a queue of preset capacity, and updating the bearing data samples in the queue.

[0184] The queue stores a preset capacity of bearing data samples. When the queue reaches the preset capacity, new bearing data samples are added to the queue. The oldest bearing data samples used for training are removed from the queue, and the newest bearing data samples are added to the queue for network training. The earlier the bearing data samples enter the queue, the earlier the training begins.

[0185] S1202: Input the bearing data samples currently stored in the queue into the initial data enhancement network.

[0186] The bearing data samples in the queue are used to update the initial data enhancement network, the initial time domain feature enhancement network and the initial frequency domain feature enhancement network.

[0187] In an embodiment of the present application, each time the bearing data samples in the queue are updated, the bearing data samples currently stored in the updated queue are input into the initial data enhancement network, and feature extraction is performed through subsequent initial time domain feature extraction and initial frequency domain feature extraction networks to obtain time domain features and frequency domain features, as well as corresponding time domain losses and frequency domain losses, thereby realizing the update of the initial data enhancement network, the initial time domain feature extraction network, and the initial frequency domain feature extraction network.

[0188] In this embodiment of the present application, bearing data samples are stored in a queue with a preset capacity, the bearing data samples in the queue are updated, and the bearing input samples currently stored in the queue are fed into the initial data augmentation network to update it. This training method updates the network parameters each time the samples in the queue are updated, allowing the network to be trained multiple times until training is complete. This ensures that all bearing data samples participate in network training, resulting in more accurate trained model parameters.

[0189] In an exemplary embodiment, Figure 13 As shown, a training method for a bearing fault detection model is provided, the method comprising:

[0190] S1301, obtaining an initial bearing fault detection model and a bearing fault dataset.

[0191] Among them, the initial bearing fault detection model includes an initial data enhancement network, an initial time domain feature extraction network, an initial frequency domain feature extraction network and an initial fault classification network.

[0192] The bearing fault dataset includes bearing data samples, bearing test samples, and bearing test labels.

[0193] The bearing data samples are used for the first stage of training, specifically for training the initial data enhancement network, the initial time domain feature extraction network, and the initial frequency domain feature extraction network;

[0194] The bearing test samples and bearing test labels are used for the second stage of training, specifically for training the initial fault classification network.

[0195] S1302, storing the bearing data samples in a queue of preset capacity, and continuously updating the bearing data samples in the queue.

[0196] S1303: Input the bearing data samples currently stored in the queue into the initial data enhancement network to obtain standardized bearing data.

[0197] S1304: Using different parameter updating methods, the network parameters in the initial time domain feature extraction network are updated to obtain two sets of parameter updating results. The time domain loss is obtained by comparing the two sets of parameter updating results.

[0198] S1305 , using different parameter updating methods to update the network parameters in the initial frequency domain feature extraction network, obtaining two sets of parameter updating results, and obtaining the frequency domain loss by comparing the two sets of parameter updating results.

[0199] S1306 , based on the time domain loss and the frequency domain loss, the initial data enhancement network, the time domain feature extraction network, and the initial frequency domain feature extraction network are trained to obtain the data enhancement network, the time domain feature extraction network, and the frequency domain feature extraction network.

[0200] The above steps S1302 to S1306 are the first stage of training, please refer to Figure 14 , Figure 14 Schematic diagram of the network architecture for the first stage of training.

[0201] S1307: Input the bearing test sample into the initial bearing fault detection model to obtain the bearing test result.

[0202] S1308, based on the loss value between the bearing test results and the bearing test labels, while keeping the parameters of the data enhancement network, the time domain feature extraction network and the frequency domain feature extraction network unchanged, the parameters in the initial fault classification network are trained to obtain the target bearing fault detection model.

[0203] Among them, the target bearing fault detection model is used to fuse the bearing time domain features and bearing frequency domain features of the bearing data in the target bearing, and perform fault identification on the features obtained by the fusion processing to obtain the bearing fault category of the target bearing.

[0204] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0205] Based on the same inventive concept, embodiments of the present application further provide a bearing fault detection device for implementing the aforementioned bearing fault detection method. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more bearing fault detection device embodiments provided below can be found in the limitations of the bearing fault detection method described above and will not be further elaborated here.

[0206] In an exemplary embodiment, Figure 15 As shown, a bearing fault detection device is provided, including: an instruction response module 1501 and a fault identification module 1502, wherein:

[0207] The instruction response module 1501 is used to respond to the fault detection instruction of the target bearing and obtain bearing data and the target bearing fault detection model;

[0208] The fault identification module 1502 is used to input the bearing data into the target bearing fault detection model, fuse the bearing time domain features and bearing frequency domain features in the bearing data through the target bearing fault detection model, and perform fault identification on the features obtained by the fusion processing to obtain the bearing fault category of the target bearing.

[0209] In an exemplary embodiment, the fault identification module 1502 includes: a feature extraction unit and a feature superposition unit, wherein:

[0210] A feature extraction unit is used to input the bearing data into a time domain feature extraction network to obtain the bearing time domain features; and input the bearing data into a frequency domain feature extraction network to obtain the bearing frequency domain features;

[0211] The feature superposition unit is used to superimpose the bearing time domain features and the bearing frequency domain features to obtain fused features.

[0212] In an exemplary embodiment, the feature extraction unit is also used to perform Fourier transform processing and radiometric transform processing on the bearing data in sequence to obtain the initial frequency domain characteristics of the bearing, and filter the initial frequency domain characteristics of the bearing based on the spectral intensity distribution in the initial frequency domain characteristics of the bearing to obtain the frequency domain characteristics of the bearing.

[0213] In an exemplary embodiment, the bearing fault detection device further includes a data enhancement module for inputting bearing data into a data enhancement network, performing standardization processing on the bearing data through the data enhancement network to obtain a bearing time series array, and performing feature fusion processing on the bearing time series array to obtain processed bearing data.

[0214] Each module in the bearing fault detection device described above can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in the computer device in the form of software, so that the processor can call and execute the corresponding operations of each module.

[0215] The present application also provides a bearing fault detection model training device for implementing the aforementioned bearing fault detection model training method. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of the one or more bearing fault detection model training device embodiments provided below can be found in the limitations of the bearing fault detection model training method described above and will not be further elaborated here.

[0216] In an exemplary embodiment, Figure 16 As shown, a training device for a bearing fault detection model is provided, comprising: a data acquisition module 1601 and a model training module 1602, wherein:

[0217] Data acquisition module 1601 is used to obtain an initial bearing fault detection model and a bearing fault data set; the initial bearing fault detection model includes an initial time domain feature extraction network and an initial frequency domain feature extraction network;

[0218] The model training module 1602 is used to train the initial time domain feature extraction network and the initial frequency domain feature extraction network based on the bearing fault data set to obtain a target bearing fault detection model; the target bearing fault detection model is used to fuse the bearing time domain features and the bearing frequency domain features of the bearing data in the target bearing, and perform fault identification on the features obtained by the fusion processing to obtain the bearing fault category of the target bearing.

[0219] In an exemplary embodiment, the model training module 1602 includes a network training unit and a model acquisition unit, wherein:

[0220] A network training unit is used to simultaneously train an initial data enhancement network, an initial time domain feature extraction network, and an initial frequency domain feature extraction network based on a bearing fault data set to obtain a data enhancement network, a time domain feature extraction network, and a frequency domain feature extraction network;

[0221] The model acquisition unit is used to obtain a target bearing fault detection model based on a data enhancement network, a time domain feature extraction network and a frequency domain feature extraction network.

[0222] In an exemplary embodiment, the network training unit is configured to include a normalization subunit, a loss acquisition subunit, and a joint training subunit, wherein:

[0223] A normalization subunit, used to input bearing data samples into the initial data augmentation network to obtain bearing normalized data;

[0224] A loss acquisition subunit, configured to input the bearing normalized data into an initial time domain feature extraction network to obtain a time domain loss, and input the bearing normalized data into an initial frequency domain feature extraction network to obtain a frequency domain loss;

[0225] The joint training subunit is used to train the initial data enhancement network, the time domain feature extraction network and the initial frequency domain feature extraction network based on the time domain loss and the frequency domain loss to obtain the data enhancement network, the time domain feature extraction network and the frequency domain feature extraction network.

[0226] In an exemplary embodiment, the normalization subunit is specifically configured to store the bearing data samples in a queue of preset capacity, update the bearing data samples in the queue, and input the bearing data samples currently stored in the queue into the initial data enhancement network.

[0227] In an exemplary embodiment, the loss acquisition subunit is specifically used to update the network parameters in the initial time domain feature extraction network using different parameter update methods, obtain two sets of parameter update results, and obtain the time domain loss by comparing the two sets of parameter update results.

[0228] In an exemplary embodiment, the loss acquisition subunit is specifically used to update the network parameters in the initial frequency domain feature extraction network using different parameter update methods, obtain two sets of parameter update results, and obtain the frequency domain loss by comparing the two sets of parameter update results.

[0229] In an exemplary embodiment, the model acquisition unit includes a result acquisition unit and a model determination unit, wherein:

[0230] A result acquisition unit, used for inputting the bearing test sample into the initial bearing fault detection model to obtain the bearing test result;

[0231] The model determination unit is used to train the parameters in the initial fault classification network based on the loss value between the bearing test results and the bearing test labels while keeping the parameters of the data enhancement network, the time domain feature extraction network and the frequency domain feature extraction network unchanged, so as to obtain the target bearing fault detection model.

[0232] Each module in the aforementioned bearing fault detection model training device can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor within a computer device in hardware form, or can be stored in a computer device memory in software form, allowing the processor to call and execute the corresponding operations of each module.

[0233] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0234] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0235] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0236] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), data processing logic devices based on quantum computing, and the like.

[0237] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0238] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A bearing fault detection method, characterized in that: The method comprises: In response to a fault detection instruction of a target bearing, a target bearing fault detection model is obtained, and according to a data collection time carried in the fault detection instruction, operating data of the target bearing within a preset time period from the current moment is collected as bearing data; Inputting the bearing data into the target bearing fault detection model, and eliminating scale differences of the bearing data through a normalization module in the target bearing fault detection model; A dilated convolution module is used to perform feature fusion extraction on the bearing data, and the bearing data is sequentially subjected to Fourier transform and affine transform processing to obtain the initial frequency domain features of the bearing; the dilated convolution module includes a residual structure and a GeLU activation function, which is used to expand the receptive field without increasing the number of parameters; The initial frequency domain features of the bearing are filtered based on the spectral intensity distribution in the initial frequency domain features of the bearing to obtain the bearing frequency domain features, and the bearing time domain features and the bearing frequency domain features in the bearing data are superimposed to obtain the features after fusion processing; fault identification is performed on the features obtained by fusion processing to obtain the bearing fault category of the target bearing; the bearing frequency domain features are time-invariant frequency domain features extracted by filtering the initial frequency domain features of the bearing data based on the spectral intensity distribution.

2. The method according to claim 1, characterized in that The target bearing fault detection model includes a time domain feature extraction network and a frequency domain feature extraction network. The bearing time domain features and bearing frequency domain features in the bearing data are superimposed to obtain fused features, including: Inputting the bearing data into the time domain feature extraction network to obtain the bearing time domain feature; and inputting the bearing data into the frequency domain feature extraction network to obtain the bearing frequency domain feature; The bearing time domain features and the bearing frequency domain features are superimposed to obtain fused features.

3. A training method for a bearing fault detection model, characterized in that: The method comprises: Obtaining an initial bearing fault detection model and a bearing fault data set; the initial bearing fault detection model includes an initial time domain feature extraction network and an initial frequency domain feature extraction network; According to the bearing fault data set, the initial time domain feature extraction network and the initial frequency domain feature extraction network are trained to obtain a target bearing fault detection model; the target bearing fault detection model is used to eliminate the scale differences of bearing data in the target bearing through a normalization module, and use a dilated convolution module to perform feature fusion extraction on the bearing data, and then perform Fourier transform processing and affine transform processing on the bearing data in sequence to obtain the initial frequency domain features of the bearing, and filter the initial frequency domain features of the bearing based on the spectral intensity distribution in the initial frequency domain features of the bearing to obtain the bearing frequency domain features, and superimpose the bearing time domain features and bearing frequency domain features in the bearing data to obtain the fused features; fault identification is performed on the features obtained by the fusion process to obtain the bearing fault category of the target bearing; the dilated convolution module includes a residual structure and a GeLU activation function, which is used to expand the receptive field without increasing the number of parameters; the bearing data includes the operating data of the target bearing within a preset time period collected from the current moment according to the data collection time carried by the fault detection instruction.

4. The method according to claim 3, characterized in that The initial bearing fault detection model further includes an initial data enhancement network; the initial time domain feature extraction network and the initial frequency domain feature extraction network in the initial bearing fault detection model are trained according to the bearing fault data set to obtain a target bearing fault detection model, including: According to the bearing fault data set, the initial data enhancement network, the initial time domain feature extraction network and the initial frequency domain feature extraction network are trained simultaneously to obtain a data enhancement network, a time domain feature extraction network and a frequency domain feature extraction network; Based on the data enhancement network, the time domain feature extraction network and the frequency domain feature extraction network, the target bearing fault detection model is obtained.

5. The method according to claim 4, characterized in that The bearing fault data set includes bearing data samples; the initial data enhancement network, the initial time domain feature extraction network, and the initial frequency domain feature extraction network are simultaneously trained based on the bearing data data set to obtain a data enhancement network, a time domain feature extraction network, and a frequency domain feature extraction network, including: Inputting the bearing data sample into the initial data enhancement network to obtain bearing standardized data; Inputting the bearing normalized data into the initial time domain feature extraction network to obtain time domain loss, and inputting the bearing normalized data into the initial frequency domain feature extraction network to obtain frequency domain loss; Based on the time domain loss and the frequency domain loss, the initial data enhancement network, the time domain feature extraction network and the initial frequency domain feature extraction network are trained to obtain the data enhancement network, the time domain feature extraction network and the frequency domain feature extraction network.

6. The method according to claim 5, characterized in that Inputting the bearing data sample into the initial data enhancement network comprises: storing the bearing data samples in a queue of preset capacity, and updating the bearing data samples in the queue; The bearing data samples currently stored in the queue are input into the initial data enhancement network.

7. The method according to claim 5, characterized in that Inputting the bearing normalized data into the initial time domain feature extraction network to obtain the time domain loss includes: Different parameter updating methods are used to update the network parameters in the initial time domain feature extraction network to obtain two sets of parameter updating results. The time domain loss is obtained by comparing the two sets of parameter updating results.

8. The method according to claim 5, characterized in that Inputting the bearing normalized data into the initial frequency domain feature extraction network to obtain frequency domain loss includes: Different parameter updating methods are used to update the network parameters in the initial frequency domain feature extraction network to obtain two sets of parameter updating results. The frequency domain loss is obtained by comparing the two sets of parameter updating results.

9. The method according to any one of claims 4 to 8, characterized in that The bearing fault dataset includes bearing test samples and bearing test labels; the initial bearing fault detection model also includes an initial fault classification network; the target bearing fault detection model is obtained based on the data enhancement network, the time domain feature extraction network, and the frequency domain feature extraction network, including: Inputting the bearing test sample into the initial bearing fault detection model to obtain a bearing test result; Based on the loss value between the bearing test results and the bearing test labels, the parameters in the initial fault classification network are trained while keeping the parameters of the data enhancement network, the time domain feature extraction network and the frequency domain feature extraction network unchanged to obtain the target bearing fault detection model.

10. A bearing fault detection device, characterized in that: The device comprises: an instruction response module, configured to respond to a fault detection instruction of a target bearing, obtain a fault detection model for the target bearing, and collect operating data of the target bearing within a preset time period from the current moment according to a data collection time carried in the fault detection instruction, as bearing data; a fault identification module, configured to input the bearing data into the target bearing fault detection model and eliminate scale differences of the bearing data through a standardization module in the target bearing fault detection model; A dilated convolution module is used to perform feature fusion extraction on the bearing data, and the bearing data is sequentially subjected to Fourier transform and affine transform processing to obtain the initial frequency domain features of the bearing; the dilated convolution module includes a residual structure and a GeLU activation function, which is used to expand the receptive field without increasing the number of parameters; The initial frequency domain features of the bearing are filtered based on the spectral intensity distribution in the initial frequency domain features of the bearing to obtain the bearing frequency domain features, the bearing time domain features and the bearing frequency domain features in the bearing data are superimposed to obtain the fused features, fault identification is performed on the fused features to obtain the bearing fault category of the target bearing; the bearing frequency domain features are time-invariant frequency domain features extracted by filtering the initial frequency domain features of the bearing data based on the spectral intensity distribution.

11. A training device for a bearing fault detection model, characterized in that: The device comprises: A data acquisition module is used to obtain an initial bearing fault detection model and a bearing fault data set; the initial bearing fault detection model includes an initial time domain feature extraction network and an initial frequency domain feature extraction network; A model training module is used to train the initial time domain feature extraction network and the initial frequency domain feature extraction network according to the bearing fault data set to obtain a target bearing fault detection model; the target bearing fault detection model is used to eliminate the scale differences of the bearing data in the target bearing through a normalization module, and use an expansion convolution module to perform feature fusion extraction on the bearing data, and then perform Fourier transform processing and affine transform processing on the bearing data in sequence to obtain the initial frequency domain features of the bearing, filter the initial frequency domain features of the bearing based on the spectral intensity distribution in the initial frequency domain features of the bearing to obtain the bearing frequency domain features, superimpose the bearing time domain features and the bearing frequency domain features in the bearing data to obtain the fused features; fault identification is performed on the features obtained by the fusion process to obtain the bearing fault category of the target bearing; the expansion convolution module includes a residual structure and a GeLU activation function, which is used to expand the receptive field without increasing the number of parameters; the bearing data includes the operating data of the target bearing within a preset time period collected from the current moment according to the data collection time carried by the fault detection instruction.

12. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

14. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

Images