Fault detection method and device for variable-speed rotating part and terminal equipment

Abstract

The invention discloses a fault detection method and device for a variable-speed rotating part and terminal equipment, and relates to the technical field of fault diagnosis. The method comprises the following steps: firstly, acquiring a vibration signal of a variable-speed rotating part in a preset time period; inputting the vibration signal into a feature enhancement layer of a pre-trained fault detection model for fault feature extraction, obtaining fault features of the vibration signal at different time points, and determining a one-dimensional feature map corresponding to a preset time period; inputting the one-dimensional feature map into a wavelet convolution layer of a fault detection model, and performing multi-frequency scale feature extraction on the one-dimensional feature map through a plurality of channels to obtain a plurality of one-dimensional sub-feature maps of different frequency scales; and splicing the plurality of one-dimensional sub-feature maps into a two-dimensional time-frequency map according to the frequency scale, inputting the two-dimensional time-frequency map into a two-dimensional convolutional neural network of the fault detection model, and determining a fault detection result of the variable-speed rotating member based on the two-dimensional time-frequency map. According to the embodiment of the invention, the fault features of different frequency scales can be associated, and the fault detection accuracy of the variable-speed rotating part is improved.

Description

Technical Field

[0001] The present application belongs to the field of fault diagnosis technology, and in particular relates to a method, apparatus and terminal equipment for detecting faults of variable speed rotating parts. Background Art

[0002] As mechanical equipment becomes increasingly integrated and complex, the likelihood of failure increases. Rotating components such as roller bearings and gears are widely used in various industrial production lines and are prone to failure, making them a major research area in fault diagnosis technology. To provide interpretable fault diagnosis for rotating components, existing methods often replace the convolution kernel with a signal analysis operator.

[0003] However, existing explainable fault diagnosis methods usually perform fault detection based on signal analysis operators with fixed frequency characteristics. When performing fault detection on rotating parts with variable speeds, the variable speed will cause nonlinear changes in the frequency components of the vibration signal of the rotating parts. Therefore, the existing explainable fault diagnosis methods have poor adaptability to frequency changes, resulting in low fault detection accuracy for rotating parts with variable speeds. Summary of the Invention

[0004] The embodiments of the present application provide a method, apparatus, and terminal device for detecting faults of variable-speed rotating parts to solve the problem of low accuracy in fault detection of variable-speed rotating parts in the prior art.

[0005] In a first aspect, an embodiment of the present application provides a method for detecting a fault in a variable speed rotating component, the method comprising:

[0006] Acquiring a vibration signal of a variable speed rotating part within a preset time period;

[0007] The vibration signal is input into the feature enhancement layer of the pre-trained fault detection model. The feature enhancement layer extracts the fault features of the vibration signal to obtain the fault features of the vibration signal at different time points. The one-dimensional feature map corresponding to the preset time period is determined based on the fault features at different time points.

[0008] The one-dimensional feature map is input into the wavelet convolution layer of the fault detection model. The multi-frequency scale feature extraction is performed on the one-dimensional feature map through multiple channels of the wavelet convolution layer to obtain multiple one-dimensional sub-feature maps with different frequency scales. Each channel corresponds to a Morley wavelet.

[0009] Multiple one-dimensional sub-feature graphs are spliced ​​into a two-dimensional time-frequency graph according to the frequency scale, and the two-dimensional time-frequency graph is input into the two-dimensional convolutional neural network of the fault detection model. The fault detection results of variable speed rotating parts are determined based on the two-dimensional time-frequency graph through the two-dimensional convolutional neural network.

[0010] In a second aspect, an embodiment of the present application provides a device for detecting faults of a variable speed rotating component, the device comprising:

[0011] An acquisition module, configured to acquire a vibration signal of a variable speed rotating part within a preset time period;

[0012] A feature extraction module is used to input the vibration signal into the feature enhancement layer of the pre-trained fault detection model, extract the fault features of the vibration signal through the feature enhancement layer, obtain the fault features of the vibration signal at different time points, and determine the one-dimensional feature map corresponding to the preset time period based on the fault features at different time points;

[0013] The feature extraction module is also used to input the one-dimensional feature map into the wavelet convolution layer of the fault detection model, and perform multi-frequency scale feature extraction on the one-dimensional feature map through multiple channels of the wavelet convolution layer to obtain multiple one-dimensional sub-feature maps of different frequency scales, where each channel corresponds to a Morley wavelet;

[0014] The determination module is used to splice multiple one-dimensional sub-feature graphs into a two-dimensional time-frequency graph according to the frequency scale, and input the two-dimensional time-frequency graph into the two-dimensional convolutional neural network of the fault detection model. Based on the two-dimensional time-frequency graph, the two-dimensional convolutional neural network is used to determine the fault detection result of the variable speed rotating part.

[0015] In a third aspect, an embodiment of the present application provides a terminal device comprising: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the method for detecting a variable speed rotating part fault as in the first aspect is implemented.

[0016] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are executed by a processor, the method for detecting a variable speed rotating part fault according to the first aspect is implemented.

[0017] In a fifth aspect, an embodiment of the present application provides a computer program product. When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device executes the method for detecting a variable speed rotating part fault as described in the first aspect.

[0018] The embodiment of the present application provides a method for detecting faults in variable speed rotating parts. The fault detection model in this method is a structure including a feature enhancement layer, a wavelet convolution layer, and a two-dimensional convolutional neural network. The three layers are interrelated. The feature enhancement layer enhances the fault features in the vibration signal. The wavelet convolution layer simultaneously captures fault features at different frequency scales through multiple channels to achieve multi-frequency scale fault feature extraction. The two-dimensional convolutional neural network correlates the fault features between multiple frequency scales, thereby identifying the drift of the fault features of the variable speed rotating parts at different frequency scales due to the speed change, ensuring that the fault features are not missed. Therefore, the embodiment of the present application can improve the fault detection accuracy of variable speed rotating parts. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0020] Figure 1 Schematic diagram of the structure of the training initial fault detection model provided in the embodiment of the present application;

[0021] Figure 2 1 is a flow chart of a method for detecting a fault of a variable speed rotating component provided by an embodiment of the present application;

[0022] Figure 3 This is a schematic diagram of the structure of the wind turbine test bench provided in an embodiment of the present application;

[0023] Figure 4 Schematic diagram of the training results of the characteristic self-enhanced composite dimensional wavelet network and the control group network provided in the embodiment of the present application;

[0024] Figure 5 This is a schematic diagram of a fault signal provided by an embodiment of the present application;

[0025] Figure 6 It is a two-dimensional time-frequency diagram of two composite dimensional wavelet networks provided in the embodiment of the present application;

[0026] Figure 7 is an envelope spectrum diagram of the fault signal after the autocorrelation operation provided by the embodiment of the present application;

[0027] Figure 8 It is a two-dimensional time-frequency diagram of the control group network provided in the embodiment of the present application;

[0028] Figure 9 1 is a schematic structural diagram of a device for detecting faults of a variable speed rotating member provided in an embodiment of the present application;

[0029] Figure 10 It is a structural diagram of the terminal device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0030] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present application by illustrating the examples of the present application.

[0031] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0032] Variable speed rotating parts refer to mechanical rotating parts whose speed is not constant during operation. The speed of these parts will change according to external conditions, load changes or adjustments of the control system. This change in speed will cause nonlinear changes in the vibration frequency components of the rotating parts.

[0033] Existing explainable fault diagnosis methods mostly rely on signal analysis operators with fixed frequency characteristics for fault detection. These methods have significant limitations when detecting faults in variable-speed rotating components. However, the frequency components of the vibration signal of a variable-speed rotating component can vary nonlinearly with speed. For example, when mechanical equipment accelerates from low to high speed, the frequency of the vibration signal can exhibit complex nonlinear variations. Existing explainable fault diagnosis methods often struggle to adapt to these dynamic frequency variations. Their signal analysis operator design often assumes relatively stable signal frequency components, which does not hold true under variable-speed operating conditions. This results in low accuracy in fault detection for variable-speed rotating components, making it difficult to meet the demands of practical industrial applications. However, speed variations in rotating machinery are common in industrial environments. For example, fans, pumps, and other equipment adjust their speed accordingly when operating under different loads or operating conditions. In these scenarios, the poor adaptability of existing technologies to frequency variations limits the performance of existing fault diagnosis methods under these variable operating conditions, resulting in poor accuracy in fault detection for variable-speed rotating components.

[0034] In order to solve the existing technology, the embodiment of the present application provides a method for detecting faults of variable speed rotating parts. After obtaining the vibration signal of the variable speed rotating part within a preset time period, the method inputs it into a pre-trained fault detection model, and sequentially passes through the feature enhancement layer, wavelet convolution layer and two-dimensional convolutional neural network of the fault detection model, and sequentially performs fault feature extraction, multi-frequency scale feature extraction and fault detection to obtain a predicted fault classification result. Among them, the wavelet convolution layer includes multiple channels, each channel corresponds to a Morley wavelet. The embodiment of the present application first extracts the fault features after feature enhancement, and simultaneously captures the fault features at different frequency scales through multiple channels of the wavelet convolution layer, which can cover the complete fault feature frequency band, and recognizes the two-dimensional time-frequency diagram through the two-dimensional convolutional neural network to realize the extraction of correlated fault features across frequency scales, which can prevent the drift of the feature frequency band caused by the speed change, thereby improving the fault detection accuracy of the variable speed rotating part.

[0035] In order to more clearly understand the embodiments of the present application, the model used in the embodiments of the present application is first described in detail.

[0036] The architecture of the fault detection model in the embodiment of the present application includes three parts: a feature enhancement layer, a wavelet convolution layer, and a two-dimensional convolutional neural network.

[0037] The feature enhancement layer performs feature extraction on the input vibration signal, enhancing the fault features within the vibration signal and outputting a one-dimensional feature map containing the enhanced fault features. Because raw vibration signals typically contain significant noise and interference, which can obscure useful fault features, the present embodiment uses the feature enhancement layer to extract and highlight the fault features within the vibration signal.

[0038] The wavelet convolution layer consists of multiple channels, each containing a Morlet wavelet at a specific frequency scale. A one-dimensional feature map is fed into the wavelet convolution layer, where the multiple channels extract features at the corresponding frequency scales, resulting in multiple one-dimensional sub-feature maps at different frequency scales.

[0039] The wavelet convolution layer of the embodiment of the present application extracts features of different frequency scales through multiple channels, which can capture multiple frequency components of fault features at different frequency scales, covering a wide frequency range, and can effectively prevent missed detection or false detection caused by characteristic frequency band drift, thereby improving the accuracy of fault detection.

[0040] The wavelet convolution layer of the embodiment of the present application uses the Morlet wavelet, which is a complex wavelet composed of a complex exponential function modulated by a Gaussian window. It maintains good localization characteristics in both the time domain and the frequency domain, is relatively concentrated in both the time domain and the frequency domain, and can effectively capture the characteristics of the signal within a specific time domain range and frequency domain range, thereby effectively detecting the time point when the fault occurred and the corresponding frequency components. In addition, the vibration signal of a faulty variable speed rotating part usually contains a large number of frequency components and periodic impact components corresponding to the fault characteristics. The shape and frequency characteristics of the Morlet wavelet have a high similarity with these signals, thereby more effectively extracting the fault characteristics in the vibration signal.

[0041] The two-dimensional convolutional neural network further extracts and analyzes the input two-dimensional time-frequency diagram and finally outputs the fault detection result. By extracting the two-dimensional time-frequency diagram through the two-dimensional convolutional neural network, the fault features of adjacent frequency scales can be obtained simultaneously, thereby extracting more complex correlation features and realizing feature extraction across frequency scales, avoiding the information loss caused by focusing only on a single frequency scale, thereby improving the accuracy of the final fault detection.

[0042] The following describes the method for detecting faults of variable speed rotating parts provided by the embodiments of the present application in conjunction with the accompanying drawings.

[0043] like Figure 1 As shown, before performing fault detection on a variable speed rotating component, it is first necessary to train the constructed initial fault detection model, which may include the following steps: S101 to S105.

[0044] S101, obtaining vibration signal samples of a variable speed rotating part within a preset time period and corresponding fault labels;

[0045] S102, inputting the vibration signal samples into a feature enhancement layer of the initial fault detection model, extracting fault features from the vibration signal samples through the feature enhancement layer to obtain fault features of the vibration signal samples at different time points, and determining an intermediate one-dimensional feature graph corresponding to a preset time period based on the fault features of the vibration signal samples at different time points;

[0046] S103: Input the intermediate one-dimensional feature map into the wavelet convolution layer of the initial fault detection model, perform multi-frequency scale feature extraction on the intermediate one-dimensional feature map through multiple channels of the wavelet convolution layer, and obtain multiple intermediate one-dimensional sub-feature maps of different frequency scales;

[0047] S104: splicing the multiple intermediate one-dimensional sub-feature graphs into an intermediate two-dimensional time-frequency graph based on the frequency scale, inputting the intermediate two-dimensional time-frequency graph into a two-dimensional convolutional neural network of the initial fault detection model, and determining a predicted fault detection result of the variable speed rotating component through the two-dimensional convolutional neural network based on the intermediate two-dimensional time-frequency graph;

[0048] S105 , determining a loss function value of the initial fault detection model based on the predicted fault detection result and the fault label, and training model parameters of the initial fault detection model based on the loss function value until a training stop condition is met, thereby obtaining a pre-trained fault detection model.

[0049] The embodiment of the present application utilizes vibration signal samples with fault labels and adopts a supervised learning approach to train an initial fault detection model. The model can continuously adjust parameters during the training process, comprehensively learn the manifestations of different types of faults in vibration signals, and then learn the correct fault feature and classification mapping relationship, thereby improving the accuracy and reliability of the fault detection model.

[0050] In some embodiments, when the present application trains the model parameters of the initial fault detection model according to the loss function value, the cross entropy loss function can be used as the objective function and the Adam algorithm can be used as the optimizer to update the model parameters.

[0051] As an example, the initial learning rate of the Adam optimizer can be set to 0.0001, the batch size to 64, and the training iteration to 100,000 epochs, where epoch represents the number of updates in which all training sample data are used once.

[0052] In some embodiments, training model parameters of the initial fault detection model according to the loss function value until a training stop condition is satisfied to obtain a pre-trained fault detection model may include:

[0053] Determining the gradient of the model parameters of the initial fault detection model according to the relationship between the loss function value and the gradient of the model parameters;

[0054] The target model parameters are determined according to the relationship between the gradient of the model parameters and the model parameters, and the initial fault detection model is updated according to the target model parameters until the training stop condition is met to obtain a pre-trained fault detection model.

[0055] The embodiment of the present application determines the gradient of the model parameters through the relationship between the loss function value and the gradient, which can provide a clear direction and basis for updating the model parameters, and determines the target model parameters based on the relationship between the gradient of the model parameters and the parameters, thereby updating the initial fault detection model, ensuring that the optimization process is efficient and has clear goals, and quickly adjusting the model parameters in the direction of reducing the loss function value, thereby improving the efficiency of model training and the performance of the model.

[0056] In some embodiments, the training stop condition may include the number of iterations reaching a preset number of iterations, or the loss function value of the initial fault detection model being less than a set threshold.

[0057] In some embodiments, the model parameters include a first frequency scale of a first channel of a wavelet convolution layer and a frequency scale increment between adjacent channels. Determining target model parameters based on a gradient of the model parameters and a relationship between the model parameters, and updating the initial fault detection model based on the target model parameters may include:

[0058] Subtract the product of the gradient of the first frequency scale and the preset learning rate from the first frequency scale to obtain the target first frequency scale;

[0059] The target frequency scale increment is obtained by subtracting the product of the gradient of the frequency scale increment and the preset learning rate from the frequency scale increment, where the initial frequency scales between adjacent channels are equidistantly distributed;

[0060] The first frequency scale of the first channel and the frequency scale increments between adjacent channels of the wavelet convolution layer of the initial fault detection model are updated according to the target first frequency scale and the target frequency scale increment.

[0061] The setting of the arithmetic distribution of the initial frequency scales between adjacent channels in the embodiment of the present application provides an orderly basis for parameter updating, so that the updated frequency scale can still maintain a certain regularity, which helps the model to stably learn the characteristics of different frequency bands. By continuously updating the frequency scale parameters, the model can better fit the fault characteristics in the training data and improve the detection accuracy of the model.

[0062] In some embodiments, the gradient calculation formulas for the first frequency scale and frequency scale increment of the first channel of the wavelet convolution layer are as shown in formulas (1)-(3), specifically:

[0063]

[0064] Among them, δ a is the gradient of the first frequency scale, is the partial derivative of the loss function with respect to the first frequency scale, is the partial derivative of the loss function with respect to the wavelet convolution kernel, is the partial derivative of the wavelet convolution kernel with respect to the first frequency scale, δ i is the gradient of the frequency scale increment, is the partial derivative of the loss function with respect to the frequency scale increment, is the partial derivative of the wavelet convolution kernel with respect to the frequency scale energy, β is the Gaussian window width of the Morley wavelet, t is the signal time point, b is the time offset parameter, a is the first frequency scale, and i is the frequency scale increment.

[0065] In some embodiments, a calculation formula for the relationship between the gradient of the first frequency scale and the frequency scale increment of the first channel of the wavelet convolution layer and the first frequency scale and the frequency scale increment of the first channel of the wavelet convolution layer is shown in formula (4), specifically:

[0066]

[0067] Among them, a k is the first frequency scale of the first channel convolution kernel of the wavelet convolution layer at the kth iteration, a k-1 is the first frequency scale of the first channel convolution kernel of the wavelet convolution layer at the k-1th iteration, γ is the learning rate, and δ a is the gradient of the first frequency scale, i k is the increment of the frequency scale of the convolution kernel of adjacent channels at the kth iteration, i k-1 is the increment of the frequency scale of the convolution kernel of adjacent channels at the k-1th iteration, δ i is the gradient of the frequency scale increment.

[0068] Figure 2 FIG. 1 is a flow chart showing a method for detecting a fault of a variable speed rotating member provided by an embodiment of the present application. Figure 2 As shown, the method may include the following steps: S201 to S204.

[0069] S201, obtaining a vibration signal of a variable speed rotating part within a preset time period.

[0070] The vibration signal is an electrical signal converted from the mechanical vibration generated by the variable speed rotating parts when the equipment is running;

[0071] In some embodiments, a vibration sensor may be provided on the device housing near the variable speed rotating part, and the present application may obtain a vibration signal collected by the vibration sensor.

[0072] In some embodiments, the preset period is a signal acquisition window of fixed length. This application can set the preset period based on the speed change rate of the variable-speed rotating component. For devices with rapidly changing speeds, a shorter preset period is set to accommodate rapidly changing frequency components, while for devices with slowly changing speeds, a longer preset period is set. For example, if a rolling bearing in a wind turbine test bench operates in an environment with variable wind speeds, the preset period can be set to 0.5 seconds to accommodate rapid changes in wind speed.

[0073] The embodiment of the present application can directly collect the original vibration signal of the variable speed rotating part during operation. When the variable speed rotating part fails, the vibration signal will show a local impact with aggravated periodic vibration during the rotation process. The periodic impact component can be used as a fault feature, so the vibration signal can be used as the basic data for subsequent fault detection.

[0074] S202: Input the vibration signal into the feature enhancement layer of the pre-trained fault detection model, extract the fault features of the vibration signal through the feature enhancement layer, obtain the fault features of the vibration signal at different time points, and determine the one-dimensional feature map corresponding to the preset time period based on the fault features at different time points.

[0075] In some embodiments, the present application can set a preset fault signal as a one-dimensional convolution kernel of a feature enhancement layer, and perform a convolution operation on the vibration signal through the preset fault signal to obtain fault features of the vibration signal at different time points.

[0076] The embodiment of the present application uses a preset fault signal as a convolution kernel to specifically enhance features related to the preset fault. This method can be used to specifically detect the preset fault.

[0077] In some embodiments, the present application can set the input signal itself as a one-dimensional convolution kernel of the feature enhancement layer, use the input vibration signal as the convolution kernel to perform autocorrelation convolution operation on the vibration signal, and obtain the fault characteristics of the vibration signal at different time points.

[0078] The embodiment of the present application uses the input vibration signal as a convolution kernel to perform an autocorrelation convolution operation, which can automatically enhance the fault characteristics of the signal itself and can detect a variety of fault types.

[0079] In some embodiments, determining a one-dimensional feature map corresponding to a preset time period based on fault features at different time points may include: treating each fault feature as an element of the one-dimensional feature map, arranging the fault features according to the time sequence of the time points corresponding to the fault features, and concatenating the fault features into the one-dimensional feature map. For example, if a preset time period of 1 second is used and sampling is performed at a sampling frequency of 5 Hz, a vibration signal including signal sampling values ​​at time sequences of t1, t2, t3, t4, and t5 is obtained in sequence. After feature extraction is performed on the vibration signal, the fault features corresponding to time points t1, t2, t3, t4, and t5 are concatenated in sequence into the one-dimensional feature map.

[0080] Since the vibration signal of a variable speed rotating part has a periodic impact component when it fails, the embodiment of the present application can extract the fault features of the vibration signal through a feature enhancement layer, enhance the periodic impact components corresponding to the fault features in the vibration signal and use them as input data for subsequent detection, thereby improving the accuracy of subsequent fault detection.

[0081] S203: Input the one-dimensional feature map into the wavelet convolution layer of the fault detection model, perform multi-frequency scale feature extraction on the one-dimensional feature map through multiple channels of the wavelet convolution layer, and obtain multiple one-dimensional sub-feature maps of different frequency scales, where each channel corresponds to a Morley wavelet.

[0082] In some embodiments, the present application can replace the convolution sum of the wavelet convolution layer with Mrolet wavelet, and extract the features of the channel corresponding to the frequency scale of the one-dimensional feature map through the Mrole wavelet of each channel to obtain a single-band one-dimensional sub-feature map of the corresponding frequency scale.

[0083] In some embodiments, the present application can set the frequency scales of the Morlet wavelet between adjacent channels to be an equidistant relationship, and perform feature extraction of the equidistant frequency scale on the one-dimensional feature through multiple channels of the wavelet convolution layer to obtain multiple one-dimensional sub-feature maps of equidistant frequency scales.

[0084] Features at different frequency scales may correspond to different fault types or degrees. This embodiment of the present application sets the frequency scales of the Morlet wavelets corresponding to adjacent channels to an arithmetic progression. The frequency scale distribution of the arithmetic progression ensures uniform and complete coverage of all frequency bands within the target frequency range, avoiding feature loss due to frequency hopping. This helps capture features across different frequency bands in the signal, prevents oversampling or undersampling of certain frequency bands, and ensures the continuity and integrity of feature extraction.

[0085] In one example, the frequency scale of the Morlet wavelet of the first channel is a, the frequency scale of the Morlet wavelet of the second channel is a+i, the frequency scale of the Morlet wavelet of the nth channel is a+(n-1)*i, and so on, where the length of the Morlet wavelet is m and the number of channels is n.

[0086] The embodiment of the present application uses Morlet wavelet to extract features from a one-dimensional feature map using different frequency scales corresponding to multiple channels. Since different fault types manifest differently at different frequency scales, fault features at different frequency scales can be obtained, reducing the possibility of missing fault features. At the same time, since the Morlet wavelet has similar shape and frequency characteristics to the vibration signal, features in the vibration signal can be better extracted, thereby improving the accuracy of subsequent fault detection.

[0087] In step S204, the multiple one-dimensional sub-feature graphs are spliced ​​into a two-dimensional time-frequency graph according to the frequency scale, and the two-dimensional time-frequency graph is input into a two-dimensional convolutional neural network of the fault detection model. The fault detection result of the variable speed rotating part is determined based on the two-dimensional time-frequency graph through the two-dimensional convolutional neural network.

[0088] The fault characteristics of variable-speed rotating parts are often distributed in multiple frequency bands due to the change in speed and drift over time. The embodiment of the present application splices multiple one-dimensional sub-feature graphs into a two-dimensional time-frequency graph, and extracts features through a two-dimensional convolutional neural network. It can identify and detect the fault characteristics of adjacent frequency bands, realize cross-band feature extraction, facilitate the analysis of correlation features between frequency bands, and thus improve the accuracy of fault detection.

[0089] The above embodiment first extracts the fault features after feature enhancement, and simultaneously captures the fault features at different frequency scales through multiple channels of the wavelet convolution layer, which can cover the complete fault feature frequency band. The two-dimensional time-frequency graph is recognized through a two-dimensional convolutional neural network to realize the extraction of correlated fault features across frequency scales, which can prevent the drift of the characteristic frequency band caused by the change of speed, thereby improving the fault detection accuracy of variable speed rotating parts.

[0090] In some embodiments, the two-dimensional convolutional neural network may include a two-dimensional convolution layer, a nonlinear activation function layer, a pooling layer, a fully connected layer, and a classification layer. S204 may include:

[0091] The two-dimensional time-frequency graph is input into the two-dimensional convolutional layer, and the two-dimensional time-frequency graph is subjected to feature extraction through the two-dimensional convolutional layer to obtain a feature map;

[0092] The feature map is input into the nonlinear activation function layer, and the feature map is nonlinearly transformed by the activation function of the nonlinear activation function layer to obtain an activated feature map;

[0093] Input the activation feature map into the pooling layer, perform dimensionality reduction on the activation feature map, and obtain the pooled feature map;

[0094] Flatten the pooled feature map into a one-dimensional vector and input it into the fully connected layer, generating a feature vector through the preset weight matrix transformation;

[0095] The feature vector is input into the classification layer, the probability distribution of each fault classification is calculated, and the fault detection result of the variable speed rotating part is determined according to the probability distribution.

[0096] The cross-band scanning of the two-dimensional convolutional neural network in the embodiment of the present application can capture the drift characteristics of the fault frequency band under variable speed. Through multi-layer processing of convolution, activation and pooling, the original features in the time-frequency graph are gradually abstracted into high-order semantic feature vectors. Finally, the feature vectors are mapped to category probability distribution, providing clear and interpretable classification results.

[0097] The 2D convolutional neural network is a network layer that performs convolution operations on the 2D time-frequency graph. The 2D convolution kernel can simultaneously capture the correlation characteristics of adjacent frequency scales and time points, thereby adapting to the frequency band drift characteristics of the variable speed signal and generating a feature map that includes combined features across frequency bands and time domains. The nonlinear activation function layer transforms the feature map using a nonlinear function, enhancing the fault characteristics manifested as impact components and generating an activation feature map.

[0098] In some embodiments, when the present application performs dimensionality reduction processing on the activation feature map through the pooling layer, maximum pooling, hybrid pooling or adaptive pooling methods can be used.

[0099] Max pooling preserves the maximum value in each region, highlighting the peak features caused by the impact component of the fault signal. Hybrid pooling, using both max pooling and average pooling in parallel, extracts peak features and global energy distribution from the feature map. Feature concatenation then fuses the two pooling results to form a more comprehensive representation of the fault signature. Adaptive pooling dynamically adjusts the pooling window size based on the local energy distribution of the time-frequency graph. For example, smaller windows are used to preserve details in high-frequency fault areas, ensuring that fault signatures are not lost while reducing dimensionality.

[0100] In one example, a faulty rolling bearing in a wind turbine test bench is used as the research object. The wind turbine test bench structure is as follows: Figure 3 As shown, the test bench consists of a drive motor 1, a coupling 2, a planetary gearbox 3, a parallel shaft gearbox 4, an encoder 5, a rolling bearing 6, a bearing seat 7, a vibration sensor 8, and a magnetic powder brake 9. The motor is used to provide power, the planetary gearbox and the parallel shaft gearbox are used to transmit power, the encoder is used to detect the speed of the output shaft, the faulty bearing is placed on the bearing seat, three acceleration sensors are used to collect the bearing's vibration signals, and the brake is used to load the equipment. By using wire cutting, different types of faults are created on the inner ring, outer ring, and rolling element of the bearing. The fault detection results can include four types of bearing vibration signals: inner ring fault, outer ring fault, rolling element fault, and no fault. The steps for fault detection of the faulty rolling bearing are as follows:

[0101] In step 1, three acceleration vibration sensors were attached to the bearing housing of the wind turbine test bench and sampled at a frequency of 20 kHz to obtain a raw signal dataset. The bearing speed for all fault types ranged from 300 rpm to 2000 rpm. For each signal type, 470 samples were collected, each lasting 0.1 seconds.

[0102] Step 2: Label the data in the original signal dataset with the corresponding fault category. 235 samples in each category are included in the training set, and 235 samples are included in the test set.

[0103] Step 3: Construct an initial fault detection model of the feature self-enhanced composite dimensional wavelet network and input the training set samples into the initial fault detection model. The network layer structure and parameters of the fault detection model are shown in Table 1.

[0104] Table 1 Network layer structure and parameters of the initial fault detection model

[0105]

[0106] Step 4: Input each group of training set samples into the initial fault detection model in batches to learn the fault data and train the initial fault detection model. After each group of training data is trained, the trained initial fault detection model is tested using the test set samples. The training is stopped when the shutdown criterion is met or the maximum number of training times is reached to obtain the fault detection model.

[0107] Step 5: Input the rolling bearing data collected subsequently during operation of the wind turbine test bench into the fault detection model to perform fault detection and obtain the fault detection result.

[0108] In some embodiments, three network models are used as a control group for testing, and the three network models include a composite-dimensional convolutional neural network, a one-dimensional wavelet convolutional network, and a composite-dimensional wavelet network without a feature enhancement layer.

[0109] The network layer structure and parameters of the composite dimensional convolutional neural network are shown in Table 2. The network layer structure includes feature self-enhancement layer, one-dimensional convolution layer, two-dimensional convolution layer, activation layer, pooling layer and fully connected layer 1.

[0110] Table 2 Network layer structure and parameters of convolutional neural network with composite dimensions

[0111]

[0112] The network layer structure and parameters of the one-dimensional wavelet convolutional network are shown in Table 3. The network layer structure includes feature self-enhancement layer, wavelet convolution layer, one-dimensional convolution layer, activation layer, pooling layer, fully connected layer 1, fully connected layer 2 and classification layer.

[0113] Table 3 Network layer structure and parameters of one-dimensional wavelet convolutional network

[0114]

[0115] The network layer structure and parameters of the composite dimensional wavelet network without feature enhancement layer are shown in Table 4. The network layer structure includes one-dimensional convolution layer, wavelet convolution layer, two-dimensional convolution layer, activation layer, pooling layer, fully connected layer 1, fully connected layer 2 and classification layer.

[0116] Table 4 Network layer structure and parameters of composite dimensional wavelet network

[0117]

[0118]

[0119] In one example, the present application trains a feature self-enhanced composite dimensional wavelet network, a composite dimensional convolutional neural network, a one-dimensional wavelet convolutional network, and a composite dimensional wavelet network without a feature enhancement layer. The training results are shown in Figure 2. Figure 4As shown, the feature-enhanced composite dimensional wavelet network of the present invention converged around the eighth iteration, with an accuracy rate of 93%. The accuracy rates of the composite dimensional convolutional neural network, the one-dimensional wavelet convolutional network, and the composite dimensional wavelet network without a feature enhancement layer after convergence were 80%, 68%, and 50%, respectively. It can be seen that when analyzing fault signals of variable-speed rotating parts, the feature-enhanced composite dimensional wavelet network of the present invention has a strong advantage in fault detection accuracy.

[0120] In one example, the present application randomly selects a fault signal from the original signal data set, and the schematic diagram of the fault signal is as follows: Figure 5 As shown, the time domain waveform of the signal is Figure 5 As shown in (a), the envelope spectrum of the fault signal is as follows Figure 5 As shown in (b), Figure 5 Fc and 2Fc in (b) represent the fault characteristic frequency band and its double frequency band of the fault signal respectively.

[0121] The signal is input into the composite dimensional wavelet network without adding the feature self-enhancement layer and the composite dimensional wavelet network with adding the feature sub-enhancement layer, and the obtained two-dimensional time-frequency diagram is as follows Figure 6 As shown, the two-dimensional time-frequency diagram output by the composite dimensional wavelet network without adding the feature self-enhancement layer is Figure 6 In (a), the two-dimensional time-frequency graph output by the composite dimensional wavelet network with the feature self-enhancement layer is Figure 6 After comparison, it can be seen that after adding the feature self-enhancement layer, the fault characteristics of the signal in the output two-dimensional features are significantly enhanced.

[0122] In the embodiment of the present application, after the feature self-enhancement layer is added to the network, the feature sub-enhancement layer performs an autocorrelation operation on the input signal to Figure 5 Taking the fault signal in as an example, after the signal is autocorrelated, the envelope spectrum of the signal is as follows Figure 7 As shown in the figure, the energy of the signal's fault feature frequency is significantly enhanced after autocorrelation. The feature self-enhancement layer uses the input signal itself as the convolution kernel, essentially autocorrelating each signal, enhancing the signal's energy and demonstrating the interpretability of the feature self-enhancement layer.

[0123] In some embodiments, Figure 5 The fault signals in the image are used as test data to input into the composite dimensional convolutional neural network, the one-dimensional wavelet convolutional network and the composite dimensional wavelet network without feature enhancement layer. The test results are as follows: Figure 8 As shown, the two-dimensional time-frequency diagram of the composite dimensional wavelet network is as follows Figure 8 As shown in (a), all the feature maps of the one-dimensional wavelet convolutional network are spliced ​​into a two-dimensional time-frequency map. The two-dimensional time-frequency map of the one-dimensional wavelet convolutional network is as follows: Figure 8As shown in (b), the two-dimensional time-frequency diagram of the composite dimensional convolutional neural network is as follows Figure 8 As shown in (c), it can be seen that compared with the other networks, the composite dimensional wavelet network retains the fault feature information more completely, demonstrating the interpretability of the wavelet convolution layer.

[0124] Figure 9 A device 900 for detecting a fault in a variable speed rotating component provided by an embodiment of the present application is shown. The device may include:

[0125] An acquisition module 901 is used to acquire a vibration signal of a variable speed rotating part within a preset time period;

[0126] Feature extraction module 902, configured to input the vibration signal into a feature enhancement layer of a pre-trained fault detection model, extract fault features from the vibration signal through the feature enhancement layer, obtain fault features of the vibration signal at different time points, and determine a one-dimensional feature map corresponding to a preset time period based on the fault features at different time points;

[0127] The feature extraction module 902 is further configured to input the one-dimensional feature map into the wavelet convolution layer of the fault detection model, perform multi-frequency scale feature extraction on the one-dimensional feature map through multiple channels of the wavelet convolution layer, and obtain multiple one-dimensional sub-feature maps at different frequency scales, where each channel corresponds to a Morley wavelet.

[0128] Determination module 903 is used to splice multiple one-dimensional sub-feature graphs into a two-dimensional time-frequency graph according to the frequency scale, and input the two-dimensional time-frequency graph into the two-dimensional convolutional neural network of the fault detection model, and determine the fault detection result of the variable speed rotating part through the two-dimensional convolutional neural network based on the two-dimensional time-frequency graph.

[0129] In some embodiments, the variable speed rotating component fault detection device 900 may further include:

[0130] The splicing module is used to arrange multiple one-dimensional sub-feature maps in the order of the frequency scale of the corresponding channels and splice them into a two-dimensional matrix to obtain a two-dimensional time-frequency map. The rows of the two-dimensional matrix correspond to different time points within a preset time period, and the columns correspond to the frequency scale of the one-dimensional sub-feature map.

[0131] In some embodiments, the variable speed rotating component fault detection device 900 may further include:

[0132] The feature extraction module 902 is further configured to input the two-dimensional time-frequency graph into the two-dimensional convolution layer, perform feature extraction on the two-dimensional time-frequency graph through the two-dimensional convolution layer, and obtain a feature map;

[0133] A transformation module is used to input the feature map into the nonlinear activation function layer, perform nonlinear transformation on the feature map through the activation function of the nonlinear activation function layer, and obtain an activated feature map;

[0134] A dimensionality reduction module is used to input the activation feature map into the pooling layer, perform dimensionality reduction processing on the activation feature map, and obtain a pooled feature map;

[0135] The transformation module is also used to flatten the pooled feature map into a one-dimensional vector and input it into the fully connected layer to generate a feature vector through the preset weight matrix transformation;

[0136] The determination module 903 is further configured to input the feature vector into the classification layer, calculate the probability distribution of each fault classification, and determine the fault detection result of the variable speed rotating part according to the probability distribution.

[0137] In some embodiments, the variable speed rotating component fault detection device 900 may further include:

[0138] The acquisition module 901 is further configured to acquire vibration signal samples of a variable speed rotating part within a preset period of time and corresponding fault labels;

[0139] The feature extraction module 902 is further configured to input the vibration signal samples into the feature enhancement layer of the initial fault detection model, extract fault features from the vibration signal samples through the feature enhancement layer, obtain fault features of the vibration signal samples at different time points, and determine an intermediate one-dimensional feature graph corresponding to a preset time period based on the fault features of the vibration signal samples at different time points;

[0140] The feature extraction module 902 is further configured to input the intermediate one-dimensional feature map into the wavelet convolution layer of the initial fault detection model, perform multi-frequency scale feature extraction on the intermediate one-dimensional feature map through multiple channels of the wavelet convolution layer, and obtain multiple intermediate one-dimensional sub-feature maps at different frequency scales;

[0141] The determination module 903 is further configured to concatenate the multiple intermediate one-dimensional sub-feature graphs into an intermediate two-dimensional time-frequency graph based on the frequency scale, input the intermediate two-dimensional time-frequency graph into a two-dimensional convolutional neural network of the initial fault detection model, and determine a predicted fault detection result of the variable speed rotating component based on the intermediate two-dimensional time-frequency graph through the two-dimensional convolutional neural network;

[0142] The training module is used to determine the loss function value of the initial fault detection model according to the predicted fault detection results and fault labels, and train the model parameters of the initial fault detection model according to the loss function value until the training stop condition is met to obtain a pre-trained fault detection model.

[0143] In some embodiments, the determination module 903 is further configured to determine the gradient of the model parameters of the initial fault detection model based on the relationship between the loss function value and the gradient of the model parameters;

[0144] The training module is also used to determine the target model parameters based on the gradient of the model parameters and the relationship between the model parameters, and update the initial fault detection model according to the target model parameters until the training stop condition is met to obtain a pre-trained fault detection model.

[0145] In some embodiments, the variable speed rotating component fault detection device 900 may further include:

[0146] a calculation module, configured to subtract the product of the gradient of the first frequency scale and a preset learning rate from the first frequency scale to obtain a target first frequency scale;

[0147] The calculation module is further configured to determine a target frequency scale increment obtained by subtracting a frequency scale increment from a gradient of the frequency scale increment and a preset learning rate, wherein initial frequency scales between adjacent channels are equidistantly distributed;

[0148] An updating module is used to update a first frequency scale of a first channel and a frequency scale increment between adjacent channels of a wavelet convolution layer of an initial fault detection model according to a target first frequency scale and a target frequency scale increment.

[0149] Figure 9 The various modules in the device shown can be implemented Figure 2 Each step in the embodiment achieves the corresponding technical effect, which will not be described here for the sake of brevity.

[0150] Figure 10 A schematic diagram of the hardware structure of the terminal device provided in an embodiment of the present application is shown.

[0151] The terminal device may include a processor 1001 and a memory 1002 storing computer program instructions.

[0152] Specifically, the processor 1001 may include a central processing unit (CPU) or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0153] The memory 1002 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 1002 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. In one example, the memory 1002 may include a removable or non-removable (or fixed) medium, or the memory 1002 may be a non-volatile solid-state memory. The memory 1002 may be inside or outside the integrated gateway disaster recovery device.

[0154] In one example, the memory 1002 may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk storage media device, an optical storage media device, a flash memory device, an electrical, optical, or other physical / tangible memory storage device. Thus, generally, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method for detecting a variable speed rotating member fault according to the present disclosure.

[0155] The processor 1001 reads and executes the computer program instructions stored in the memory 1002 to implement Figure 1 The method for detecting faults of a variable speed rotating member in the illustrated embodiment.

[0156] In one example, the terminal device may further include a communication interface 1003 and a bus 1004. Figure 10 As shown, the processor 1001, the memory 1002, and the communication interface 1003 are connected via a bus 1004 and communicate with each other.

[0157] The communication interface 1003 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0158] Bus 1004 includes hardware, software or both, couples the parts of terminal equipment to each other.For example, but not limitation, bus can include Accelerated Graphics Port (AGP) or other graphics buses, Enhanced Industry Standard Architecture (EISA) bus, Front Side Bus (FSB), Hyper Transport (HT) interconnection, Industry Standard Architecture (ISA) bus, InfiniBand interconnection, Low Pin Count (LPC) bus, memory bus, Micro Channel Architecture (MCA) bus, Peripheral Component Interconnect (PCI) bus, PCI-Express (PCI-X) bus, Serial Advanced Technology Attachment (SATA) bus, Video Electronics Standards Association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 1004 can include one or more buses. Although the present application embodiment describes and shows specific bus, the application considers any suitable bus or interconnection.

[0159] In addition, in conjunction with the variable speed rotating component fault detection method in the above-mentioned embodiment, the present application embodiment can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when the computer program instructions are executed by a processor, any of the variable speed rotating component fault detection methods in the above-mentioned embodiment is implemented.

[0160] An embodiment of the present application further provides a computer program product, including a computer program, which, when processed and executed, implements any one of the methods for detecting faults of variable-speed rotating parts in the above embodiments.

[0161] It should be understood that the present application is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted here. In the above embodiments, several specific steps are described and illustrated as examples. However, the method process of the present application is not limited to the specific steps described and illustrated. Those skilled in the art can make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present application.

[0162] The functional blocks shown in the above-described block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of the present application are programs or text segments used to perform the required tasks. The programs or text segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via a data signal carried in a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, read-only memories (ROMs), flash memory, erasable read-only memories (EROMs), floppy disks, compact disc read-only memories (CD-ROMs), optical discs, hard disks, optical fiber media, radio frequency (RF) links, etc. The text segments can be downloaded via computer networks such as the Internet and intranets.

[0163] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps. In other words, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0164] Aspects of the present disclosure have been described above with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor or a field programmable logic circuit. It is also understood that each box in the block diagram and / or flowchart and the combination of the boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0165] The above description is only a specific embodiment of the present application. Those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the scope of protection of the present application.

Claims

1. A method for detecting faults in a variable speed rotating part, characterized in that: The method comprises: Acquiring a vibration signal of a variable speed rotating part within a preset time period; Inputting the vibration signal into a feature enhancement layer of a pre-trained fault detection model, extracting fault features from the vibration signal through the feature enhancement layer to obtain fault features of the vibration signal at different time points, and determining a one-dimensional feature map corresponding to the preset time period based on the fault features at different time points; Inputting the one-dimensional feature map into the wavelet convolution layer of the fault detection model, performing multi-frequency scale feature extraction on the one-dimensional feature map through multiple channels of the wavelet convolution layer to obtain multiple one-dimensional sub-feature maps of different frequency scales, each channel corresponding to a Morley wavelet; Multiple one-dimensional sub-feature graphs are spliced ​​into a two-dimensional time-frequency graph according to the frequency scale, and the two-dimensional time-frequency graph is input into the two-dimensional convolutional neural network of the fault detection model. The fault detection result of the variable speed rotating part is determined by the two-dimensional convolutional neural network based on the two-dimensional time-frequency graph.

2. The method for detecting faults of a variable speed rotating part according to claim 1, wherein: The step of splicing the plurality of one-dimensional sub-feature graphs into a two-dimensional time-frequency graph according to a frequency scale includes: Multiple one-dimensional sub-feature maps are arranged and spliced ​​into a two-dimensional matrix according to the frequency scale of the corresponding channel to obtain a two-dimensional time-frequency map, where the rows of the two-dimensional matrix correspond to different time points within a preset time period, and the columns correspond to the frequency scale of the one-dimensional sub-feature map.

3. The method for detecting faults of a variable speed rotating part according to claim 1, wherein: The two-dimensional convolutional neural network includes a two-dimensional convolutional layer, a nonlinear activation function layer, a pooling layer, a fully connected layer, and a classification layer. The determining of the fault detection result of the variable speed rotating part by the two-dimensional convolutional neural network based on the two-dimensional time-frequency graph includes: Inputting the two-dimensional time-frequency graph into the two-dimensional convolutional layer, performing feature extraction on the two-dimensional time-frequency graph through the two-dimensional convolutional layer to obtain a feature map; Inputting the feature map into the nonlinear activation function layer, performing nonlinear transformation on the feature map using the activation function of the nonlinear activation function layer to obtain an activated feature map; Inputting the activation feature map into the pooling layer, performing dimensionality reduction processing on the activation feature map to obtain a pooling feature map; Flattening the pooled feature map into a one-dimensional vector and inputting it into the fully connected layer to generate a feature vector through a preset weight matrix transformation; The feature vector is input into the classification layer, the probability distribution of each fault classification is calculated, and the fault detection result of the variable speed rotating part is determined according to the probability distribution.

4. The method for detecting faults of a variable speed rotating component according to claim 1, wherein: Before inputting the vibration signal into a feature enhancement layer of a pre-trained fault detection model, the method further includes: Obtain vibration signal samples of a variable speed rotating part within a preset time period and their corresponding fault labels; Inputting the vibration signal samples into a feature enhancement layer of an initial fault detection model, performing fault feature extraction on the vibration signal samples through the feature enhancement layer to obtain fault features of the vibration signal samples at different time points, and determining an intermediate one-dimensional feature graph corresponding to the preset time period based on the fault features of the vibration signal samples at different time points; Inputting the intermediate one-dimensional feature map into the wavelet convolution layer of the initial fault detection model, performing multi-frequency scale feature extraction on the intermediate one-dimensional feature map through multiple channels of the wavelet convolution layer, and obtaining multiple intermediate one-dimensional sub-feature maps of different frequency scales; splicing a plurality of the intermediate one-dimensional sub-feature graphs into an intermediate two-dimensional time-frequency graph according to a frequency scale, inputting the intermediate two-dimensional time-frequency graph into a two-dimensional convolutional neural network of the initial fault detection model, and determining a predicted fault detection result of the variable speed rotating component through the two-dimensional convolutional neural network based on the intermediate two-dimensional time-frequency graph; The loss function value of the initial fault detection model is determined according to the predicted fault detection result and the fault label, and the model parameters of the initial fault detection model are trained according to the loss function value until a training stop condition is met, thereby obtaining the pre-trained fault detection model.

5. The method for detecting faults of a variable speed rotating component according to claim 4, wherein: The model parameters of the initial fault detection model are trained according to the loss function value until a training stop condition is satisfied to obtain a pre-trained fault detection model, comprising: Determining the gradient of the model parameters of the initial fault detection model according to the relationship between the loss function value and the gradient of the model parameters; Target model parameters are determined according to the relationship between the gradient of the model parameters and the model parameters, and the initial fault detection model is updated according to the target model parameters until a training stop condition is met to obtain a pre-trained fault detection model.

6. The method for detecting faults of a variable speed rotating component according to claim 5, wherein: The model parameters include a first frequency scale of a first channel of a wavelet convolution layer and a frequency scale increment between adjacent channels. The target model parameters are determined based on the relationship between the gradient of the model parameters and the model parameters, and the initial fault detection model is updated based on the target model parameters, including: Subtract the product of the gradient of the first frequency scale and the preset learning rate from the first frequency scale to obtain the target first frequency scale; The target frequency scale increment is obtained by subtracting the product of the gradient of the frequency scale increment and the preset learning rate from the frequency scale increment, where the initial frequency scales between adjacent channels are equidistantly distributed; The first frequency scale of the first channel and the frequency scale increments between adjacent channels of the wavelet convolution layer of the initial fault detection model are updated according to the target first frequency scale and the target frequency scale increment.

7. A device for detecting faults of variable speed rotating parts, characterized in that: The device comprises: An acquisition module, configured to acquire a vibration signal of a variable speed rotating part within a preset time period; a feature extraction module, configured to input the vibration signal into a feature enhancement layer of a pre-trained fault detection model, extract fault features from the vibration signal through the feature enhancement layer, obtain fault features of the vibration signal at different time points, and determine a one-dimensional feature map corresponding to the preset time period based on the fault features at different time points; The feature extraction module is further configured to input the one-dimensional feature map into the wavelet convolution layer of the fault detection model, perform multi-frequency scale feature extraction on the one-dimensional feature map through multiple channels of the wavelet convolution layer, and obtain multiple one-dimensional sub-feature maps at different frequency scales, where each channel corresponds to a Morley wavelet; A determination module is used to splice the multiple one-dimensional sub-feature graphs into a two-dimensional time-frequency graph according to the frequency scale, and input the two-dimensional time-frequency graph into the two-dimensional convolutional neural network of the fault detection model, and determine the fault detection result of the variable speed rotating part through the two-dimensional convolutional neural network based on the two-dimensional time-frequency graph.

8. A terminal device, characterized in that: The device includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the method for detecting a variable speed rotating component fault according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the method for detecting faults of a variable speed rotating component according to any one of claims 1 to 6.

10. A computer program product, characterized in that When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device is caused to execute the method for detecting faults of a variable speed rotating component according to any one of claims 1 to 6.

Images