Mechanical rotating part fault diagnosis method, device, equipment and medium

Abstract

The invention discloses a mechanical rotating part fault diagnosis method and device, equipment and a medium, and relates to the technical field of fault diagnosis, and the method comprises the steps: monitoring and obtaining the current operation data of a mechanical rotating part, and carrying out the Fourier transform and continuous wavelet transform of the current operation data, obtaining a target frequency domain image and a target time frequency image under the same pixel size; inputting the target frequency domain image and the target time frequency image into a pre-training fault diagnosis model, so that a trained one-dimensional convolutional network and a trained two-dimensional convolutional network of the model carry out feature extraction and fusion on the target frequency domain image and the target time frequency image respectively, and carrying out fault classification prediction on fused image features. Outputting probability distribution results of the current operation data relative to various standard fault data sets; and taking the standard fault category corresponding to the maximum probability value in the probability distribution result as a predicted fault diagnosis result. And through similarity calculation, the fault type of the mechanical rotating part is accurately judged.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a method, device, equipment and medium for diagnosing faults of mechanical rotating parts. Background Art

[0002] In the field of fault diagnosis, the types of fault data are also increasing. For new mechanical faults, there is a risk of missing them if they are identified solely by traditional methods. Therefore, it is necessary to study the similarities between different fault data in order to better understand the patterns and laws of faults, so as to identify and prevent faults in a more targeted manner. Traditional fault diagnosis mainly relies on regular physical inspections and the experience and judgment of maintenance personnel. This method is not only time-consuming, but also inefficient, and it is difficult to achieve early detection and prevention of faults. With the development of information technology and data science, especially the application of machine learning and deep learning technologies, new solutions are provided for current fault diagnosis. Through real-time monitoring and analysis of component operation data, abnormal patterns can be discovered in a timely manner, and potential faults can be predicted and prevented.

[0003] At present, in the field of fault diagnosis, existing research has mainly focused on using traditional statistical methods and machine learning algorithms to analyze sensor data, such as vibration data, temperature data, etc. These methods have improved the accuracy of fault diagnosis to a certain extent, but still face challenges such as complex data processing and diverse fault modes. In recent years, deep learning technology has been introduced into fault diagnosis due to its outstanding performance in image recognition, speech processing and other fields, demonstrating its powerful ability to process complex data and pattern recognition. Although current technology is able to handle some common fault diagnosis problems, there are still great challenges in predicting complex and hidden faults. In addition, existing fault diagnosis systems often lack in-depth analysis of the similarity of fault data, which limits the breadth and depth of fault diagnosis. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a mechanical rotating component fault diagnosis method, device, equipment and medium, which can quickly and accurately perform fault diagnosis and prediction of mechanical rotating components. The specific scheme is as follows:

[0005] In a first aspect, the present application discloses a method for diagnosing a fault of a mechanical rotating component, comprising:

[0006] Monitor and obtain current operation data generated by the mechanical rotating component during operation, and perform Fourier transform and continuous wavelet transform on the current operation data to obtain a target frequency domain image and a target time-frequency image at the same pixel size;

[0007] The target frequency domain image and the target time-frequency image are respectively input into a pre-trained fault diagnosis model, so that the trained one-dimensional convolutional network and the trained two-dimensional convolutional network of the pre-trained fault diagnosis model respectively extract features from the target frequency domain image and the target time-frequency image, perform feature fusion on the extracted target frequency domain features and target time-frequency features, and perform fault classification prediction on the fused image features, so as to output the probability distribution results of the current operating data relative to various standard fault data sets;

[0008] The standard fault category corresponding to the maximum probability value in the probability distribution result is used as the predicted fault diagnosis result of the mechanical rotating component.

[0009] Optionally, performing Fourier transform on the current running data to obtain an initial frequency domain image includes:

[0010] The current running data with the DC component removed is subjected to Fourier transformation to obtain corresponding frequency domain data, and an initial frequency domain image is generated based on the frequency domain data.

[0011] Optionally, performing Fourier transform on the current running data with the DC component removed to obtain corresponding frequency domain data, and generating an initial frequency domain image based on the frequency domain data, includes:

[0012] Obtaining an average value of the current operating data to obtain a DC component of the current operating data;

[0013] removing the DC component from the current operation data to obtain target current operation data;

[0014] Performing a fast Fourier transform on the current running data of the target, and obtaining the absolute value after the fast Fourier transform to obtain amplitude information corresponding to each frequency component;

[0015] The amplitude information is length-normalized to obtain frequency domain data that eliminates the influence of the data length of the current running data, so as to generate an initial frequency domain image based on the frequency domain data.

[0016] Optionally, performing continuous wavelet transform on the current running data to obtain an initial time-frequency image includes:

[0017] Performing continuous wavelet transform on the current running data based on preset wavelet transform parameters to extract time-frequency information of the current running data to obtain corresponding time-frequency data, so as to generate an initial time-frequency image based on the time-frequency data;

[0018] Accordingly, obtaining the target frequency domain image and the target time-frequency image at the same pixel size includes:

[0019] The initial frequency domain image and the initial time-frequency image are resized to obtain a target frequency domain image and a target time-frequency image with the same pixel size.

[0020] Optionally, the mechanical rotating component fault diagnosis method further includes:

[0021] The time-frequency data and the frequency domain data are subjected to image color conversion to obtain an initial time-frequency image in RGB format and an initial frequency domain image in RGB format.

[0022] Optionally, before the target frequency domain image and the target time-frequency image are respectively input into the pre-trained fault diagnosis model, the method further includes:

[0023] The two-stream convolutional neural network model is trained using the historical standard fault data sets of each mechanical rotating component, so that the one-dimensional convolutional network of the two-stream convolutional neural network model extracts the global features of each historical frequency domain image in the historical standard fault data set, and the two-dimensional convolutional network extracts the image space features of the corresponding historical time-frequency images in the historical standard fault data set, and then the global features and the image space features are feature fused through the pooling layer to obtain fused features;

[0024] The fused features and corresponding fault labels are used to train each model network layer of the two-stream convolutional neural network model to update the model weights of the two-stream convolutional neural network model, thereby obtaining a pre-trained fault diagnosis model that has learned the discriminative power of feature representation.

[0025] Optionally, the using the fused features and the corresponding fault labels to train each model network layer of the two-stream convolutional neural network model to update the model weights of the two-stream convolutional neural network model to obtain a pre-trained fault diagnosis model that has learned the feature representation discriminative power, including:

[0026] The fused features are discarded and fully connected through the Dropout layer and the fully connected layer of the two-stream convolutional neural network model to train and adjust the weight factor of the fully connected layer, the first convolution kernel weight of the one-dimensional convolutional network, and the second convolution kernel weight of the two-dimensional convolutional network;

[0027] Inputting the target features output by the fully connected layer into the classification layer to train the classification layer to convert the target features into fault category scores under a preset dimension;

[0028] The fault category score is converted into a corresponding predicted probability distribution through a Softmax layer, and the cross entropy loss result is calculated based on the predicted probability distribution and the label encoding of the corresponding standard fault category. The model weights of the two-stream convolutional neural network model are reversely updated according to the cross entropy loss result to obtain a current fault diagnosis model that meets the preset update end conditions as a pre-trained fault diagnosis model.

[0029] Optionally, the mechanical rotating component fault diagnosis method further includes:

[0030] A preset update end condition is constructed based on the preset training batch, training batch size, and learning rate of the optimizer of the pre-trained fault diagnosis model.

[0031] Optionally, converting the fault category score into a corresponding predicted probability distribution through a Softmax layer includes:

[0032] The Softmax layer is used to calculate the proportion of each fault category score to the total fault category score, so as to generate a corresponding prediction probability distribution based on each proportion result.

[0033] Optionally, the step of taking the standard fault category corresponding to the maximum probability value in the probability distribution result as the predicted fault diagnosis result of the mechanical rotating component includes:

[0034] The probability means of the probability distribution results corresponding to all samples of the current operating data are calculated according to each standard fault category, and the standard fault category corresponding to the maximum probability value in the probability means is used as the predicted fault diagnosis result of the mechanical rotating component.

[0035] In a second aspect, the present application discloses a mechanical rotating component fault diagnosis device, comprising:

[0036] A data processing module is used to monitor and obtain current operation data generated by the mechanical rotating parts during operation, and to perform Fourier transform and continuous wavelet transform on the current operation data to obtain a target frequency domain image and a target time-frequency image at the same pixel size;

[0037] A feature extraction and prediction module, used to input the target frequency domain image and the target time-frequency image into a pre-trained fault diagnosis model, respectively, so that the trained one-dimensional convolutional network and the trained two-dimensional convolutional network of the pre-trained fault diagnosis model respectively extract features from the target frequency domain image and the target time-frequency image, perform feature fusion on the extracted target frequency domain features and target time-frequency features, perform fault classification prediction on the fused image features, and output the probability distribution results of the current operating data relative to various standard fault data sets;

[0038] The result output module is used to use the standard fault category corresponding to the maximum probability value in the probability distribution result as the predicted fault diagnosis result of the mechanical rotating component.

[0039] In a third aspect, the present application discloses an electronic device, comprising:

[0040] Memory, used to store computer programs;

[0041] The processor is used to execute the computer program to implement the steps of the above-disclosed method for diagnosing faults of mechanical rotating parts.

[0042] In a fourth aspect, the present application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the steps of the aforementioned disclosed method for diagnosing faults of mechanical rotating parts are implemented.

[0043] It can be seen that the present application discloses a fault diagnosis method for a mechanical rotating component, comprising: monitoring and acquiring the current operation data generated by the mechanical rotating component during operation, and performing Fourier transform and continuous wavelet transform on the current operation data respectively, so as to obtain a target frequency domain image and a target time-frequency image at the same pixel size; inputting the target frequency domain image and the target time-frequency image into a pre-trained fault diagnosis model respectively, so that the trained one-dimensional convolutional network and the trained two-dimensional convolutional network of the pre-trained fault diagnosis model respectively extract features from the target frequency domain image and the target time-frequency image, so as to perform feature fusion on the extracted target frequency domain features and target time-frequency features, and perform fault classification prediction on the fused image features, so as to output the probability distribution result of the current operation data relative to various standard fault data sets; and taking the standard fault category corresponding to the maximum probability value in the probability distribution result as the predicted fault diagnosis result of the mechanical rotating component. It can be seen that by simultaneously performing Fourier transform and continuous wavelet transform on the operation data, information can be obtained from the two dimensions of frequency domain and time-frequency to generate a target frequency domain image and a target time-frequency image. Through the pre-trained fault diagnosis model, the trained one-dimensional convolutional network and two-dimensional convolutional network process the frequency domain image and the time-frequency image respectively, extract the corresponding features, and the two network stream outputs are fused before the fully connected layer to integrate various information, significantly improve the model's ability to recognize complex fault modes, and effectively deal with the various faults of mechanical rotating parts. Then, the model performs fault classification prediction on the fused image features, outputs the probability distribution results corresponding to each standard fault category, and can more accurately determine the fault type of mechanical rotating parts. Reduce interference from human factors, determine the fault type based on the essence of the data, improve diagnostic accuracy, and provide a reliable basis for equipment maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0045] Figure 1 A flow chart of a fault diagnosis method for a mechanical rotating component disclosed in the present application;

[0046] Figure 2 A flow chart of a specific mechanical rotating component fault diagnosis method disclosed in this application;

[0047] Figure 3 This is a pre-trained fault diagnosis model architecture diagram disclosed in this application;

[0048] Figure 4 This is a schematic diagram of the structure of a mechanical rotating component fault diagnosis device disclosed in this application;

[0049] Figure 5 This is a structural diagram of an electronic device disclosed in this application. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0051] In the field of fault diagnosis, existing research has mainly focused on using traditional statistical methods and machine learning algorithms to analyze sensor data, such as vibration data, temperature data, etc. These methods have improved the accuracy of fault diagnosis to a certain extent, but still face challenges such as complex data processing and diverse fault modes. In recent years, deep learning technology has been introduced into fault diagnosis due to its outstanding performance in image recognition, speech processing and other fields, demonstrating its powerful ability to process complex data and pattern recognition. Although current technology is able to handle some common fault diagnosis problems, there are still great challenges in predicting complex and hidden faults. In addition, existing fault diagnosis systems often lack in-depth analysis of the similarity of fault data, which limits the breadth and depth of fault diagnosis.

[0052] To this end, the present invention provides a mechanical rotating component fault diagnosis solution, which can quickly and accurately perform fault diagnosis and prediction of the mechanical rotating component.

[0053] Reference Figure 1 As shown, an embodiment of the present invention discloses a fault diagnosis method for a mechanical rotating component, comprising:

[0054] Step S11: monitor and obtain current operation data generated by the mechanical rotating component during operation, and perform Fourier transform and continuous wavelet transform on the current operation data to obtain a target frequency domain image and a target time-frequency image at the same pixel size.

[0055] In this embodiment, multiple sensors, such as vibration sensors and speed sensors, are installed at different positions of the rotating parts of the locomotive running gear. The vibration sensor can monitor the vibration signal of the component and obtain the impact characteristics related to the fault; the speed sensor records the speed of the rotating component, and the speed fluctuation can also indicate a fault. These sensors convert physical quantities into electrical signals or digital signals to realize the monitoring and preliminary collection of current operating data.

[0056] In this embodiment, the data collected from the sensor may contain interference information such as noise and abnormal values. Therefore, it is necessary to preprocess the current operation data to remove invalid information so as to perform subsequent frequency domain analysis and time-frequency analysis on the preprocessed current operation data. Specifically, time series analysis, spectrum analysis and other methods can be used to filter out feature data related to fault diagnosis. For example, in the impact sample data, key information such as impact frequency and amplitude that can reflect the fault can be filtered out to improve data availability.

[0057] In this embodiment, the current operation data with the DC component removed is subjected to Fourier transformation to obtain corresponding frequency domain data, and an initial frequency domain image is generated based on the frequency domain data. Specifically, the average value of the current operation data is obtained to obtain the DC component of the current operation data; the DC component is removed from the current operation data to obtain the target current operation data; the target current operation data is subjected to fast Fourier transformation, and the absolute value after the fast Fourier transformation is obtained to obtain the amplitude information corresponding to each frequency component; the amplitude information is subjected to length normalization processing to obtain the frequency domain data that eliminates the influence of the data length of the current operation data, so as to generate an initial frequency domain image based on the frequency domain data.

[0058] It is understandable that in mechanical signal processing, the DC component will mask the dynamic feature information related to the fault. In order to highlight the AC component related to the fault in the signal, so that the subsequent analysis can focus on the more valuable dynamic change part and improve the accuracy of fault feature extraction, the DC component is removed from each pre-processed current operation data before Fourier transform. Specifically, the data average value of the current operation data is used as the DC component, and the DC component is removed from each pre-processed current operation data to obtain the target current operation data. Further, the target current operation data is converted by FFT (Fast Fourier Transform). The formula for fast Fourier transform conversion is as follows:

[0059] ;

[0060] in, represents the frequency domain data obtained after processing, is the current running data (time domain data), yes The average value of represents the fast Fourier transform, for The data length, In this way, the frequency domain data obtained by FFT can observe the signal from the frequency perspective and more intuitively find the frequency components related to the fault, such as abnormal changes in amplitude at a specific frequency, etc., providing key frequency domain feature basis for fault diagnosis.

[0061] Since the result of FFT is usually symmetrical, only half of the spectrum is needed to represent the entire spectrum. , just take the processed front The information of each frequency point is sufficient, that is:

[0062] ;

[0063] In this embodiment, the current operation data is subjected to continuous wavelet transform based on preset wavelet transform parameters to extract the time-frequency information of the current operation data and obtain corresponding time-frequency data, so as to generate an initial time-frequency image based on the time-frequency data. Perform continuous wavelet transform (CWT), where the mother wavelet of the continuous wavelet transform is the Morlet wavelet, which can provide good time-frequency localization characteristics when analyzing vibration signals. Set the scale range of CWT and the total scale number to 256 to determine the fineness of the transformation, where the continuous wavelet transform is shown in the following formula:

[0064] ;

[0065] in, For the mother wavelet, for The complex conjugate of is the scale parameter, when hour is stretched for low frequency analysis, when hour is compressed and used for high-frequency analysis; is the translation parameter, Used to normalize the signal. is the time-frequency data, and then the corresponding initial time-frequency image is generated based on each time-frequency data , intuitively displaying the energy distribution of the signal at different times and frequencies.

[0066] It should be noted that the scale parameter Determines the frequency range of the analysis, the larger Corresponding to low-frequency signals, such as the ability to detect bearing outer ring wear fault signals, smaller Corresponding to high-frequency signals, such as roller microcrack fault signals, the translation parameter It can locate the specific time point when the fault occurs. In this way, by adjusting the scale parameters and translation parameters and combining the characteristics of the mother wavelet, a comprehensive analysis of the time-frequency characteristics of the signal is achieved, which can not only capture high-frequency transient events, but also analyze low-frequency long-term trends, and accurately locate the time and frequency range of the fault.

[0067] In this embodiment, the time-frequency data and the frequency domain data are subjected to image color conversion to obtain an initial time-frequency image in RGB (Red Green Blue) format and an initial frequency domain image in RGB format. It can be understood that the processed FFT frequency domain data and the initial time-frequency image generated based on the time-frequency data obtained by continuous wavelet transform The images are converted into RGB format and adjusted to a uniform image (pixel) size to obtain an initial time-frequency image in RGB format and an initial frequency domain image in RGB format.

[0068] Step S12: input the target frequency domain image and the target time-frequency image into the pre-trained fault diagnosis model respectively, so that the trained one-dimensional convolutional network and the trained two-dimensional convolutional network of the pre-trained fault diagnosis model respectively extract features from the target frequency domain image and the target time-frequency image, fuse the extracted target frequency domain features and target time-frequency features, and perform fault classification prediction on the fused image features, so as to output the probability distribution results of the current operating data relative to various types of standard fault data sets.

[0069] In this embodiment, the target frequency domain image and the target time-frequency image are respectively input into two independent feature extraction channels of the pre-trained fault diagnosis model, wherein the frequency domain feature channel is the trained one-dimensional convolutional network in the trained two-stream CNN (Convolutional Neural Network) model, and then the preset feature extraction module of the frequency domain feature channel is used to perform deep feature analysis on the target frequency domain image to extract key features that characterize the fault frequency components, such as spectrum peaks, energy distribution, etc. The other independent feature extraction channel is the time-frequency feature channel, wherein the time-frequency feature channel is the trained two-dimensional convolutional network in the trained two-stream CNN model, and the preset feature analysis module is used to capture local features of the target time-frequency image to extract key features that characterize the time-domain dynamic characteristics of the fault, such as the time distribution of the impulse signal and the time-frequency energy concentration area.

[0070] Furthermore, the extracted target frequency domain features and target time-frequency features are subjected to feature fusion processing. The specific process of the fusion processing is as follows: dimension alignment and standardization of the target frequency domain features and the target time-frequency features are performed to ensure the comparability of the features; specifically, a preset fusion strategy (such as weighted superposition, splicing or attention mechanism) is used to generate fused composite features, that is, fused image features. The fused image features contain both global regularities in the frequency domain and local details in the time-frequency domain.

[0071] The fused composite features are input into the preset classification decision module, and the prediction results of the input data and various standard fault data sets are generated through nonlinear mapping and probability normalization. The prediction results are converted into probability distribution using a normalized exponential function (such as Softmax), and the confidence value of each fault label is output. Example output: Assuming that the standard fault categories of the standard fault data set include bearing inner ring fault, outer ring fault, roller fault, etc., the model output probability distribution is [0.85, 0.10, 0.05], indicating that the probability that the input data belongs to the inner ring fault is 85%.

[0072] Step S13: taking the standard fault category corresponding to the maximum probability value in the probability distribution result as the predicted fault diagnosis result of the mechanical rotating component.

[0073] In this embodiment, the probability mean values ​​of the probability distribution results corresponding to all samples of the current operating data are calculated according to each standard fault category, and the standard fault category corresponding to the largest probability value in the probability mean value is used as the predicted fault diagnosis result of the mechanical rotating component. It can be understood that if the current operating data is a single sample, the standard fault category corresponding to the largest probability value in the probability distribution result is directly taken as the predicted fault diagnosis result of the mechanical rotating component; if the current operating data is a batch sample, the probability mean value of each sample under each standard fault category is calculated. The standard fault label corresponding to the largest probability value in the probability mean value is selected as the final diagnosis result. Example: If the probability mean value of all samples calculated for the current operating data and the fault data set of "bearing inner ring fault" is 0.85, and the probability mean value of all samples calculated for the current operating data and the fault data set of "outer ring fault" is 0.10, it is determined to be an inner ring fault.

[0074] It can be seen that the present application discloses a fault diagnosis method for a mechanical rotating component, comprising: monitoring and acquiring the current operation data generated by the mechanical rotating component during operation, and performing Fourier transform and continuous wavelet transform on the current operation data respectively, so as to obtain a target frequency domain image and a target time-frequency image at the same pixel size; inputting the target frequency domain image and the target time-frequency image into a pre-trained fault diagnosis model respectively, so that the trained one-dimensional convolutional network and the trained two-dimensional convolutional network of the pre-trained fault diagnosis model respectively extract features from the target frequency domain image and the target time-frequency image, so as to perform feature fusion on the extracted target frequency domain features and target time-frequency features, and perform fault classification prediction on the fused image features, so as to output the probability distribution result of the current operation data relative to various standard fault data sets; and taking the standard fault category corresponding to the maximum probability value in the probability distribution result as the predicted fault diagnosis result of the mechanical rotating component. It can be seen that by simultaneously performing Fourier transform and continuous wavelet transform on the operation data, information can be obtained from the two dimensions of frequency domain and time-frequency to generate a target frequency domain image and a target time-frequency image. Through the pre-trained fault diagnosis model, the trained one-dimensional convolutional network and two-dimensional convolutional network process the frequency domain image and the time-frequency image respectively, extract the corresponding features, and the two network stream outputs are fused before the fully connected layer to integrate various information, significantly improve the model's ability to recognize complex fault modes, and effectively deal with the various faults of mechanical rotating parts. Then, the model performs fault classification prediction on the fused image features, outputs the probability distribution results corresponding to each standard fault category, and can more accurately determine the fault type of mechanical rotating parts. Reduce interference from human factors, determine the fault type based on the essence of the data, improve diagnostic accuracy, and provide a reliable basis for equipment maintenance.

[0075] Reference Figure 2As shown, the embodiment of the present invention discloses a specific method for diagnosing faults of mechanical rotating parts. Compared with the previous embodiment, this embodiment further illustrates and optimizes the technical solution. Specifically:

[0076] Step S21: monitor and obtain current operation data generated by the mechanical rotating component during operation, and perform Fourier transform and continuous wavelet transform on the current operation data to obtain a target frequency domain image and a target time-frequency image at the same pixel size.

[0077] Step S22: Use the historical standard fault data sets of each mechanical rotating part to train the two-stream convolutional neural network model, so that the one-dimensional convolutional network of the two-stream convolutional neural network model extracts the global features of each historical frequency domain image in the historical standard fault data set, and the two-dimensional convolutional network extracts the image space features of the corresponding historical time-frequency images in the historical standard fault data set, and then the global features and the image space features are feature fused through the convergence layer to obtain the fused features.

[0078] In this embodiment, the wavelet time-frequency image data and spectrum data processed for various types of faults are labeled in sequence, wherein a total of 6 fault categories are set, and they are labeled as "0", "1", "2", "3", "4", and "5" respectively, wherein "0" can represent a normal data set, and the remaining labels can be labeled in the order of QJ bearing fault, roller fault, inner ring fault, tread fault, and outer ring fault, thereby obtaining a historical standard fault data set carrying fault labels;

[0079] Build / train a dual-stream CNN network model architecture. The model processing and analysis combines the complex input of time domain and frequency domain data, and considers the application scenarios of time domain image data and frequency domain feature data. The model uses frequency domain data and time-frequency domain data as model inputs to capture different aspects of fault sample data and realize fault data set recognition and classification. The specific model architecture is as follows: Figure 3As shown: First, 1D-CNN is the one-dimensional convolutional network of the present invention, which is responsible for processing frequency domain data. After the training, a trained one-dimensional convolutional network is obtained, wherein the one-dimensional convolutional network includes: a convolutional layer 1D-1 with a kernel size of 5×5 and a number of kernels of 6, which is used to extract local features in the frequency domain data through convolution operations, and the output dimension is 60×60; a pooling layer 1D-1 with a pooling kernel of 2×2 and a number of kernels of 6 is used to downsample the local features output by the convolutional layer, while retaining the main features and reducing the data dimension, and the output dimension becomes 30×30. The convolutional layer 1D-2 and the pooling layer 1D-2 repeat the above operations to further extract deep-level features and reduce the dimension. Finally, the features are integrated through the fully connected layer FC layer 1D, and the output dimension is 12704, converting the local features into global feature representations suitable for classification. 2D-CNN is a two-dimensional convolutional network of the present invention, which is used to process time-frequency image data, wherein the convolution layer 2D-1 convolves the image data through 5×5 convolution kernels (6), extracts the spatial hierarchical features in the image, and outputs a 60×60 feature map. The pooling layer 2D-1 uses a 2×2 pooling kernel for downsampling, reduces the amount of data and enhances the model's ability to abstract image features, and outputs a 30×30 feature map. The convolution layer 2D-2 and the pooling layer 2D-2 continue to mine deep spatial features, and the fully connected layer FC layer 2D integrates these features and outputs 12704-dimensional data to provide a basis for subsequent classification. The outputs of 1D-CNN and 2D-CNN are fused in the convergence layer, integrating two different types of features (frequency domain and time domain features) into 16320-dimensional fused features, so that the model can comprehensively use multi-faceted information for fault judgment.

[0080] Step S23: Use the fused features and corresponding fault labels to train each model network layer of the two-stream convolutional neural network model to update the model weights of the two-stream convolutional neural network model to obtain a pre-trained fault diagnosis model that has learned the feature representation discriminative power.

[0081] In this embodiment, the fused features are discarded and fully connected through the Dropout layer and the fully connected layer of the two-stream convolutional neural network model to train and adjust the weight factor of the fully connected layer, the first convolution kernel weight of the one-dimensional convolutional network, and the second convolution kernel weight of the two-dimensional convolutional network; the target features output by the fully connected layer are input to the classification layer to train the classification layer to convert the target features into fault category scores under the preset dimension; the fault category scores are converted into corresponding predicted probability distributions through the Softmax layer, and the cross entropy loss results are calculated based on the predicted probability distribution and the label encoding of the corresponding standard fault category, and the model weights of the two-stream convolutional neural network model are reversely updated according to the cross entropy loss results to obtain the current fault diagnosis model that meets the preset update end conditions as a pre-trained fault diagnosis model. It can be understood that the Dropout layer randomly discards some neuron connections to prevent the model from overfitting and improve the generalization ability of the model. The subsequent FC-A1 and FC-A2 layers further process the fused features. By adjusting the weights of the fully connected layer FC layer 1D and FC layer 2D, the model can better learn the characteristic patterns of frequency domain and time-frequency image data. The weights of the FC-A1 and FC-A2 layers are adjusted so that the model can learn more discriminative feature representations, so as to more accurately distinguish different fault categories in subsequent classification tasks. The weights of the convolution kernels in the convolution layers (such as 1D-1, 1D-2, 2D-1, 2D-2) are adjusted because the convolution layer is the basis of feature extraction, and the extracted features will be passed to the subsequent fully connected layers for processing. When the fully connected layer adjusts the weights according to the loss function, the error information will be passed back to the convolution layer through the back propagation algorithm, thereby also updating the weights of the convolution kernels, so that the convolution layer can better extract features related to fault diagnosis.

[0082] In this way, when the model learns a more discriminative feature representation, it outputs 1120-dimensional and 184-dimensional data, which are the target features. Then the classification layer and the Softmax layer are trained. Specifically, the classification layer is trained to map the above-mentioned high-dimensional target feature classifications to 16-dimensional data, corresponding to different standard fault categories. Specifically, the Softmax layer is used to calculate the proportion of each fault category score to the total fault category score, so as to generate the corresponding predicted probability distribution based on each proportion result. It can be understood that the Softmax layer converts the 16-dimensional data output by the classification layer into a predicted probability distribution to help the model make decisions in the classification task of multiple fault data sets, and calculates the similarity between the predicted probability distribution of the input data set and each type of known fault data set, thereby achieving fault diagnosis and prediction. The calculation formula of Softmax is shown as follows:

[0083] ;

[0084] in, is the output of the FC layer (fully connected layer), , is the probability value of the corresponding output.

[0085] Among them, the network parameter settings of the one-dimensional convolutional network and the two-dimensional convolutional network are shown in Table 1 below:

[0086] Table 1

[0087]

[0088] In this embodiment, a preset update end condition is constructed based on the training batch, training batch size, and learning rate of the optimizer of the pre-set pre-trained fault diagnosis model. It can be understood that the training batch is set to 200 times, the batch size is 64, and the cross entropy loss function is used to measure the difference between the model prediction result and the true label. The network parameters are continuously adjusted at a learning rate of 0.005 through the SGD optimizer (Stochastic Gradient Descent) to gradually converge the model and improve the accuracy of fault diagnosis.

[0089] It can be seen that the model architecture adopts a dual-stream structure, which includes two independent convolutional networks, one for processing two-dimensional time-frequency image data and the other for processing one-dimensional spectrum data. Finally, a feature fusion layer is used to integrate the two types of features for the final classification task.

[0090] It should be noted that after the pre-trained fault diagnosis model is trained, the remaining case data set labeled as outer ring fault is used as the verification set to verify the trained model, except for the locomotive running gear bearing and tread fault case data for model training. The similarity calculation results of no fault, QJ bearing fault, roller fault, inner ring fault, tread fault, and outer ring fault are 0.33%, 5.26%, 6.82%, 0.00%, 0.02%, and 87.58%, respectively. In this way, the fault diagnosis result of the verification sample is outer ring fault, which is consistent with the corresponding true fault label. Take other validation sample sets to validate the trained model. For example, select the wheelset tread fault dataset to validate the trained model. The similarity calculation results of no fault, QJ bearing fault, roller fault, inner ring fault, tread fault, and outer ring fault are 7.37%, 3.75%, 1.82%, 0.09%, 86.17%, and 0.80%, respectively. In this way, the fault diagnosis result of the validation sample is tread fault, which is consistent with the corresponding real fault label. It can be seen that through data verification, this scheme can realize the similarity calculation of different fault datasets within an effective time, and at the same time can ensure a good recognition accuracy.

[0091] Step S24: input the target frequency domain image and the target time-frequency image into the pre-trained fault diagnosis model respectively, so that the trained one-dimensional convolutional network and the trained two-dimensional convolutional network of the pre-trained fault diagnosis model respectively extract features from the target frequency domain image and the target time-frequency image, fuse the extracted target frequency domain features and target time-frequency features, and perform fault classification prediction on the fused image features, so as to output the probability distribution results of the current operating data relative to various types of standard fault data sets.

[0092] Step S25: taking the standard fault category corresponding to the maximum probability value in the probability distribution result as the predicted fault diagnosis result of the mechanical rotating component.

[0093] For more detailed processing steps in steps S21, S24, and S25, please refer to the aforementioned disclosed embodiment content, which will not be repeated here.

[0094] It can be seen that by using the historical standard fault data sets of various mechanical rotating parts to train the two-stream convolutional neural network model, the model can learn a rich variety of fault modes and characteristics. The Dropout layer randomly discards some neuron connections during the training process, which effectively prevents the model from overfitting and enables the model to maintain good performance when facing new and unseen data. The model optimized after multiple rounds of training can better adapt to the fault diagnosis needs of different mechanical rotating parts, has stronger generalization ability, and can be widely used in fault diagnosis scenarios of various mechanical rotating parts.

[0095] Reference Figure 4 As shown, the present invention also discloses a mechanical rotating component fault diagnosis device, comprising:

[0096] The data processing module 11 is used to monitor and obtain the current operation data generated by the mechanical rotating parts during operation, and perform Fourier transform and continuous wavelet transform on the current operation data to obtain a target frequency domain image and a target time-frequency image at the same pixel size;

[0097] The feature extraction and prediction module 12 is used to input the target frequency domain image and the target time-frequency image into the pre-trained fault diagnosis model respectively, so that the trained one-dimensional convolutional network and the trained two-dimensional convolutional network of the pre-trained fault diagnosis model respectively extract features from the target frequency domain image and the target time-frequency image, fuse the extracted target frequency domain features and target time-frequency features, and perform fault classification prediction on the fused image features, so as to output the probability distribution results of the current operation data relative to various standard fault data sets;

[0098] The result output module 13 is used to use the standard fault category corresponding to the maximum probability value in the probability distribution result as the predicted fault diagnosis result of the mechanical rotating component.

[0099] It can be seen that the present application discloses monitoring and obtaining the current operation data generated by the mechanical rotating parts during operation, and performing Fourier transform and continuous wavelet transform on the current operation data respectively to obtain the target frequency domain image and the target time-frequency image at the same pixel size; the target frequency domain image and the target time-frequency image are respectively input into the pre-trained fault diagnosis model, so that the trained one-dimensional convolutional network and the trained two-dimensional convolutional network of the pre-trained fault diagnosis model respectively extract the features of the target frequency domain image and the target time-frequency image, so as to perform feature fusion on the extracted target frequency domain features and target time-frequency features, and perform fault classification prediction on the fused image features, so as to output the probability distribution results of the current operation data relative to various standard fault data sets; the standard fault category corresponding to the maximum probability value in the probability distribution result is used as the predicted fault diagnosis result of the mechanical rotating parts. It can be seen that by performing Fourier transform and continuous wavelet transform on the operation data at the same time, information can be obtained from the two dimensions of frequency domain and time-frequency to generate the target frequency domain image and the target time-frequency image. Through the pre-trained fault diagnosis model, the trained one-dimensional convolutional network and two-dimensional convolutional network process the frequency domain image and the time-frequency image respectively, extract the corresponding features, and the two network stream outputs are fused before the fully connected layer to integrate various information, significantly improve the model's ability to recognize complex fault modes, and effectively deal with the various faults of mechanical rotating parts. Then, the model performs fault classification prediction on the fused image features, outputs the probability distribution results corresponding to each standard fault category, and can more accurately determine the fault type of mechanical rotating parts. Reduce interference from human factors, determine the fault type based on the essence of the data, improve diagnostic accuracy, and provide a reliable basis for equipment maintenance.

[0100] Furthermore, the present application also discloses an electronic device. Figure 5 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content in the diagram cannot be regarded as any limitation on the scope of use of the present application.

[0101] Figure 5 The present invention provides a schematic diagram of the structure of an electronic device 20 provided in an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the mechanical rotating component fault diagnosis method disclosed in any of the aforementioned embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0102] In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present application, and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0103] Among them, the processor 21 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 21 can be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 21 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0104] In addition, the memory 22, as a carrier for storing resources, can be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0105] Among them, the operating system 221 is used to manage and control the hardware devices and computer programs 222 on the electronic device 20, so as to realize the operation and processing of the massive data 223 in the memory 22 by the processor 21, which can be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program that can be used to complete the mechanical rotating component fault diagnosis method performed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 can further include a computer program that can be used to complete other specific tasks. In addition to data transmitted from an external device received by the electronic device, the data 223 can also include data collected by its own input and output interface 25, etc.

[0106] Furthermore, the present application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the aforementioned disclosed mechanical rotating component fault diagnosis method is implemented. For the specific steps of the method, reference may be made to the corresponding contents disclosed in the aforementioned embodiments, and no further description will be given here.

[0107] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0108] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application. The steps of the method or algorithm described in conjunction with the embodiments disclosed herein can be implemented directly with hardware, a software module executed by a processor, or a combination of the two. The software module can be placed in a random access memory RAM (Random Access Memory), memory, read-only memory ROM (Read Only Memory), electrically programmable EPROM (Electrically Programmable Read Only Memory), electrically erasable programmable EEPROM (ElectricErasable Programmable Read Only Memory), register, hard disk, removable disk, CD-ROM (CompactDisc-Read Only Memory), or any other form of storage medium known in the technical field.

[0109] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0110] The scheme provided by the present invention is introduced in detail above. Specific examples are used in this article to illustrate the principle and implementation mode of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation mode and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for diagnosing faults of mechanical rotating parts, characterized in that: include: Monitor and obtain current operation data generated by the mechanical rotating component during operation, and perform Fourier transform and continuous wavelet transform on the current operation data to obtain a target frequency domain image and a target time-frequency image at the same pixel size; The target frequency domain image and the target time-frequency image are respectively input into a pre-trained fault diagnosis model, so that the trained one-dimensional convolutional network and the trained two-dimensional convolutional network of the pre-trained fault diagnosis model respectively extract features from the target frequency domain image and the target time-frequency image, perform feature fusion on the extracted target frequency domain features and target time-frequency features, and perform fault classification prediction on the fused image features, so as to output the probability distribution results of the current operating data relative to various standard fault data sets; The standard fault category corresponding to the maximum probability value in the probability distribution result is used as the predicted fault diagnosis result of the mechanical rotating component.

2. The mechanical rotating component fault diagnosis method according to claim 1, characterized in that: Performing Fourier transform on the current running data to obtain an initial frequency domain image, including: The current running data with the DC component removed is subjected to Fourier transformation to obtain corresponding frequency domain data, and an initial frequency domain image is generated based on the frequency domain data.

3. The mechanical rotating component fault diagnosis method according to claim 2, characterized in that: The performing Fourier transform on the current running data with the DC component removed to obtain corresponding frequency domain data, and generating an initial frequency domain image based on the frequency domain data, comprises: Obtaining an average value of the current operating data to obtain a DC component of the current operating data; removing the DC component from the current operation data to obtain target current operation data; Performing a fast Fourier transform on the current running data of the target, and obtaining the absolute value after the fast Fourier transform to obtain amplitude information corresponding to each frequency component; The amplitude information is length-normalized to obtain frequency domain data that eliminates the influence of the data length of the current running data, so as to generate an initial frequency domain image based on the frequency domain data.

4. The mechanical rotating component fault diagnosis method according to claim 2, characterized in that: Performing continuous wavelet transform on the current running data to obtain an initial time-frequency image includes: Performing continuous wavelet transform on the current running data based on preset wavelet transform parameters to extract time-frequency information of the current running data to obtain corresponding time-frequency data, so as to generate an initial time-frequency image based on the time-frequency data; Accordingly, obtaining the target frequency domain image and the target time-frequency image at the same pixel size includes: The initial frequency domain image and the initial time-frequency image are resized to obtain a target frequency domain image and a target time-frequency image with the same pixel size.

5. The mechanical rotating component fault diagnosis method according to any one of claims 2 to 4, characterized in that: Also includes: The time-frequency data and the frequency domain data are subjected to image color conversion to obtain an initial time-frequency image in RGB format and an initial frequency domain image in RGB format.

6. The mechanical rotating component fault diagnosis method according to claim 1, characterized in that: Before the target frequency domain image and the target time-frequency image are respectively input into the pre-trained fault diagnosis model, the method further includes: The two-stream convolutional neural network model is trained using the historical standard fault data sets of each mechanical rotating component, so that the one-dimensional convolutional network of the two-stream convolutional neural network model extracts the global features of each historical frequency domain image in the historical standard fault data set, and the two-dimensional convolutional network extracts the image space features of the corresponding historical time-frequency images in the historical standard fault data set, and then the global features and the image space features are feature fused through the pooling layer to obtain fused features; The fused features and corresponding fault labels are used to train each model network layer of the two-stream convolutional neural network model to update the model weights of the two-stream convolutional neural network model, thereby obtaining a pre-trained fault diagnosis model that has learned the discriminative power of feature representation.

7. The mechanical rotating component fault diagnosis method according to claim 6, characterized in that: The method of training each model network layer of the two-stream convolutional neural network model by using the fused features and the corresponding fault labels to update the model weights of the two-stream convolutional neural network model and obtain a pre-trained fault diagnosis model that has learned the discriminative power of feature representation, includes: The fused features are discarded and fully connected through the Dropout layer and the fully connected layer of the two-stream convolutional neural network model to train and adjust the weight factor of the fully connected layer, the first convolution kernel weight of the one-dimensional convolutional network, and the second convolution kernel weight of the two-dimensional convolutional network; Inputting the target features output by the fully connected layer into the classification layer to train the classification layer to convert the target features into fault category scores under a preset dimension; The fault category score is converted into a corresponding predicted probability distribution through a Softmax layer, and the cross entropy loss result is calculated based on the predicted probability distribution and the label encoding of the corresponding standard fault category. The model weights of the two-stream convolutional neural network model are reversely updated according to the cross entropy loss result to obtain a current fault diagnosis model that meets the preset update end conditions as a pre-trained fault diagnosis model.

8. The mechanical rotating component fault diagnosis method according to claim 7, characterized in that: Also includes: A preset update end condition is constructed based on the preset training batch, training batch size, and learning rate of the optimizer of the pre-trained fault diagnosis model.

9. The mechanical rotating component fault diagnosis method according to claim 7, characterized in that: The step of converting the fault category score into a corresponding predicted probability distribution through a Softmax layer includes: The Softmax layer is used to calculate the proportion of each fault category score to the total fault category score, so as to generate a corresponding prediction probability distribution based on each proportion result.

10. The mechanical rotating component fault diagnosis method according to claim 1, characterized in that: The method of using the standard fault category corresponding to the maximum probability value in the probability distribution result as the predicted fault diagnosis result of the mechanical rotating component includes: The probability means of the probability distribution results corresponding to all samples of the current operating data are calculated according to each standard fault category, and the standard fault category corresponding to the maximum probability value in the probability means is used as the predicted fault diagnosis result of the mechanical rotating component.

11. A fault diagnosis device for a mechanical rotating component, characterized in that: include: A data processing module is used to monitor and obtain current operation data generated by the mechanical rotating parts during operation, and to perform Fourier transform and continuous wavelet transform on the current operation data to obtain a target frequency domain image and a target time-frequency image at the same pixel size; A feature extraction and prediction module, used to input the target frequency domain image and the target time-frequency image into a pre-trained fault diagnosis model, respectively, so that the trained one-dimensional convolutional network and the trained two-dimensional convolutional network of the pre-trained fault diagnosis model respectively extract features from the target frequency domain image and the target time-frequency image, perform feature fusion on the extracted target frequency domain features and target time-frequency features, perform fault classification prediction on the fused image features, and output the probability distribution results of the current operating data relative to various standard fault data sets; The result output module is used to use the standard fault category corresponding to the maximum probability value in the probability distribution result as the predicted fault diagnosis result of the mechanical rotating component.

12. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor is used to execute the computer program to implement the steps of the mechanical rotating component fault diagnosis method according to any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that: Used to store computer programs; wherein, when the computer program is executed by a processor, the steps of the mechanical rotating component fault diagnosis method according to any one of claims 1 to 10 are implemented.

Images