Rolling bearing fault diagnosis method, device and equipment and storage medium
By acquiring the vibration waveform data of the rolling bearing, using the similarity relay algorithm to build the training set and verification set, and combining the wavelet transform layer with the Transform Vision network for model construction, the problems of high data acquisition cost and poor generalization ability in the existing technology are solved. , improves the accuracy and noise robustness of fault diagnosis, and improves the intelligence level of operation and maintenance of industrial equipment.
Patent Information
- Application Number
- CN202411227301.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-03
- Publication Date
- 2024-11-15
AI Technical Summary
The existing rolling bearing fault prediction and classification methods have problems such as high data acquisition cost, difficult parameters to obtain, and manual entry is prone to errors. The generalization ability of deep learning methods is poor, resulting in poor practical application results.
By obtaining the vibration waveform data of the rolling bearing, performing data preprocessing and feature extraction, using the similarity relay algorithm to build the training set and verification set, combining the downsampled wavelet transformation layer and the Transform Vision network for model construction and verification, and generating the target fault diagnosis model.
It effectively reduces data acquisition costs, improves the noise robustness of the model, the accuracy and generalization capabilities of fault diagnosis, and improves the intelligence level of operation and maintenance of industrial equipment.
Smart Images

Figure CN118965080A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of rolling bearing intelligent fault diagnosis, and in particular to a rolling bearing fault diagnosis method, device, equipment and storage medium. Background Art
[0002] Rolling bearings are widely used basic mechanical components, and are applied in various fields of national economy and national defense. Realizing the prediction and classification of rolling bearing faults is of great value and significance to the safe and stable operation of major equipment and facilities, realizing predictive maintenance, and improving economic benefits and equipment safety.
[0003] There are two main methods for rolling bearing fault prediction and classification: one method requires obtaining multiple bearing parameter indicators and using classic formulas to calculate the fault characteristic frequency, and then making identification judgments. However, in actual work, the parameters of rolling bearings are not easy to obtain, and manual entry is costly and error-prone. The other rolling bearing diagnosis method based on deep learning faces the problem of inconsistent distribution of training data sets and test data sets, which leads to poor generalization ability of the diagnosis method and poor practical application effect. Summary of the invention
[0004] In view of this, the purpose of the present invention is to provide a rolling bearing fault diagnosis method, device, equipment and storage medium, which can effectively reduce the data acquisition cost, solve the adverse effects caused by the inconsistent data distribution between the training set and the test set, and improve the noise robustness of the model as well as the accuracy and generalization ability of fault diagnosis. The specific scheme is as follows:
[0005] In a first aspect, the present application provides a rolling bearing fault diagnosis method, comprising:
[0006] By acquiring the vibration waveform data corresponding to each time node of the rolling bearing and performing corresponding data preprocessing operations, corresponding time series data satisfying a first preset length is obtained;
[0007] Based on preset feature extraction and fusion rules, the time series data is subjected to corresponding data sliding window division, feature extraction, feature fusion and image conversion to obtain corresponding initial training sets and test sets;
[0008] Based on the preset validation set division rule, the test set, the preset division ratio and the similarity deviation algorithm, the initial training set is screened to obtain a corresponding validation set and a target training set;
[0009] The model is constructed by combining the down-sampled wavelet transform layer with the preset Tranform Vision network to obtain the corresponding initial neural network model;
[0010] After completing the model training corresponding to the initial neural network model based on the target training set, model verification is performed based on the obtained trained model and the verification set to determine the corresponding target fault diagnosis model when the verification passes, so as to trigger the corresponding fault diagnosis operation using the target fault diagnosis model.
[0011] Optionally, the performing corresponding data sliding window division, feature extraction, feature fusion and image conversion on the time series data based on preset feature extraction and fusion rules includes:
[0012] Based on a preset sliding window, a preset repetition rate and a second preset length, each of the time series data is divided into data to obtain a plurality of subsequence data corresponding to each of the time series data;
[0013] Based on the preset feature extraction rules, the time domain features, frequency domain features and time-frequency features of each of the subsequence data are respectively extracted to obtain corresponding feature extraction results;
[0014] Continuous wavelet transform and two-dimensional image conversion are performed on each of the subsequence data respectively, and feature fusion is performed with the corresponding two-dimensional image based on the corresponding feature extraction result to obtain feature graphs corresponding to each of the time series data respectively.
[0015] Optionally, the performing continuous wavelet transform and two-dimensional image conversion on each of the subsequence data respectively, and performing feature fusion with the corresponding two-dimensional image based on the corresponding feature extraction result, includes:
[0016] Performing continuous wavelet transform and two-dimensional image conversion on each of the subsequence data to obtain a corresponding two-dimensional image;
[0017] After normalizing the corresponding feature extraction results and the time series data, the obtained normalized data is written into a specific position in the corresponding two-dimensional image to complete the corresponding feature fusion operation.
[0018] Optionally, the screening of samples of the initial training set based on a preset validation set partitioning rule, the test set, a preset partitioning ratio, and a similarity deviation algorithm to obtain a corresponding validation set includes:
[0019] Calculating a similarity deviation value between each of the feature graphs in the test set and each of the feature graphs in the initial training set based on the corresponding feature extraction results to obtain a corresponding first similarity deviation value;
[0020] Determine, from the initial training set, a number of candidate samples corresponding to each of the feature graphs in the test set according to a preset sample screening condition and each of the first similarity deviation values;
[0021] For a plurality of candidate samples corresponding to each of the feature graphs in the test set, similarity deviation values are calculated based on the corresponding vibration waveform data, and target samples corresponding to each of the feature graphs in the test set are determined according to the obtained second similarity deviation values;
[0022] Based on a preset division ratio and in a random selection manner, each of the target samples is selected to obtain a corresponding verification set and a target training set.
[0023] Optionally, the model is constructed by combining the down-sampled wavelet transform layer with a preset Tranform Vision network to obtain a corresponding initial neural network model, including:
[0024] By replacing the down-sampling wavelet transform layer with the convolution layer with a step size of 1 and the pooling layer with a step size of 2 to complete the corresponding wavelet integration preset Tranform Vision network design operation, the corresponding initial neural network model is obtained.
[0025] Optionally, performing model validation based on the obtained trained model and the validation set includes:
[0026] For any target feature map in the validation set, corresponding low-frequency feature extraction is performed through the down-sampling wavelet transform layer in the trained model to obtain corresponding extraction results;
[0027] After deep feature mining is performed on the extraction results through several preset Tranform Vision network layers in the trained model, fault diagnosis and weighted calculation are performed through the multi-layer perceptron layer in the trained model and the obtained deep mining results to obtain a target fault diagnosis result corresponding to the target feature map, and complete the corresponding verification operation.
[0028] Optionally, the performing of fault diagnosis and weighted calculation by using the multi-layer perceptron layer in the trained model and the obtained deep mining results includes:
[0029] After performing fault diagnosis, a weighted result corresponding to the characteristic graph is determined based on an initial diagnosis result corresponding to the target characteristic graph and a preset weighting formula; wherein the initial diagnosis result includes a fault classification result of each of the subsequence data corresponding to the target characteristic graph;
[0030] The target fault diagnosis result corresponding to the target characteristic diagram is obtained by rounding down the weighted result.
[0031] In a second aspect, the present application provides a rolling bearing fault diagnosis device, comprising:
[0032] A data preprocessing module, used to obtain vibration waveform data corresponding to each time node of the rolling bearing and perform corresponding data preprocessing operations to obtain corresponding time series data that meets a first preset length;
[0033] A feature extraction and fusion module is used to perform corresponding data sliding window division, feature extraction, feature fusion and image conversion on the time series data based on preset feature extraction and fusion rules to obtain corresponding initial training sets and test sets;
[0034] A validation set acquisition module is used to screen samples of the initial training set based on a preset validation set division rule, the test set, a preset division ratio, and a similarity deviation algorithm to obtain a corresponding validation set and a target training set;
[0035] A model building module, used for building a model by combining a down-sampled wavelet transform layer with a preset Tranform Vision network to obtain a corresponding initial neural network model;
[0036] The target model acquisition module is used to perform model verification based on the obtained trained model and the verification set after completing the model training corresponding to the initial neural network model based on the target training set, so as to determine the corresponding target fault diagnosis model when the verification passes, so as to trigger the corresponding fault diagnosis operation using the target fault diagnosis model.
[0037] In a third aspect, the present application provides an electronic device, including:
[0038] Memory, used to store computer programs;
[0039] The processor is used to execute the computer program to implement the steps of the aforementioned rolling bearing fault diagnosis method.
[0040] In a fourth aspect, the present application provides a computer-readable storage medium for storing a computer program, wherein the computer program implements the steps of the aforementioned rolling bearing fault diagnosis method when executed by a processor.
[0041] It can be seen that in the present application, the vibration waveform data corresponding to each time node of the rolling bearing is obtained, and the corresponding data preprocessing operation is performed to obtain the corresponding time series data that meets the first preset length; based on the preset feature extraction and fusion rules, the time series data is subjected to corresponding data sliding window division, feature extraction, feature fusion and image conversion to obtain the corresponding initial training set and test set; based on the preset verification set division rule, the test set, the preset division ratio and the similarity deviation algorithm, the initial training set is subjected to sample screening to obtain the corresponding verification set and the target training set; the model is constructed by combining the down-sampling wavelet transform layer with the preset Tranform Vision network to obtain the corresponding initial neural network model; after completing the model training corresponding to the initial neural network model based on the target training set, the model verification is performed based on the obtained trained model and the verification set to determine the corresponding target fault diagnosis model when the verification passes, so as to trigger the corresponding fault diagnosis operation using the target fault diagnosis model. That is, in this application, the vibration waveform data of the rolling bearing is collected and the training set and verification set are constructed using the similarity deviation algorithm, and then the model is constructed through the preset Tranform Vision network, and then the model is trained and verified, so as to use the verified target fault diagnosis model for diagnosis. In this way, the data collection cost can be effectively reduced, the adverse effects caused by the inconsistent data distribution of the training set and the test set can be solved, the noise robustness of the model and the accuracy and generalization ability of fault diagnosis can be improved, and the intelligent level of industrial equipment operation and maintenance can be improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] 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.
[0043] Figure 1 A flow chart of a rolling bearing fault diagnosis method provided in this application;
[0044] Figure 2 A schematic diagram of the structure of a specific target fault diagnosis model provided in this application;
[0045] Figure 3 A specific rolling bearing fault diagnosis method flow chart provided in this application;
[0046] Figure 4 A schematic diagram of the structure of a rolling bearing fault diagnosis device provided in this application;
[0047] Figure 5 A structural diagram of an electronic device provided for this application. DETAILED DESCRIPTION
[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0049] There are two main types of existing rolling bearing fault prediction and classification methods: one method requires obtaining multiple parameter indicators of the bearing and using the classic formula to calculate the fault characteristic frequency, and then performing identification and judgment. However, in actual work, the parameters of rolling bearings are not easy to obtain and the manual entry cost is high and prone to errors. Another rolling bearing diagnosis method based on deep learning faces the problem that the distribution of the training data set and the test data set are inconsistent, resulting in poor generalization ability of the diagnosis method and poor actual application effect. To this end, the present application provides a rolling bearing fault diagnosis solution that can effectively reduce the cost of data acquisition, solve the adverse effects caused by the inconsistent data distribution of the training set and the test set, and improve the noise robustness of the model as well as the accuracy and generalization ability of fault diagnosis.
[0050] See also Figure 1 As shown, an embodiment of the present invention discloses a rolling bearing fault diagnosis method, comprising:
[0051] Step S11, obtaining the vibration waveform data corresponding to each time node of the rolling bearing and performing corresponding data preprocessing operations to obtain corresponding time series data that meets the first preset length.
[0052] Specifically, in this embodiment, the vibration waveform data corresponding to each time node of the rolling bearing needs to be collected first, and then pre-processed to continuously intercept the data into time series data with a length of a first preset length. The first preset length can be autonomously configured and adjusted based on actual needs.
[0053] Step S12: Based on preset feature extraction and fusion rules, the time series data is subjected to corresponding data sliding window division, feature extraction, feature fusion and image conversion to obtain corresponding initial training sets and test sets.
[0054] Specifically, in this embodiment, the corresponding data sliding window division, feature extraction, feature fusion and image conversion are performed on the time series data based on the preset feature extraction and fusion rules, including: data division is performed on each of the time series data based on the preset sliding window, the preset repetition rate and the second preset length to obtain a plurality of sub-sequence data corresponding to each of the time series data; time domain features, frequency domain features and time-frequency features of each of the sub-sequence data are extracted based on the preset feature extraction rules to obtain corresponding feature extraction results; continuous wavelet transform and two-dimensional image conversion are performed on each of the sub-sequence data, and feature fusion is performed with the corresponding two-dimensional image based on the corresponding feature extraction results to obtain feature maps corresponding to each of the time series data. It should be understood that the time series data is divided into multiple sub-sequence data of the same length by means of a sliding window according to a sliding window size of half the sequence length and a repetition rate of 50%. For example, if the first preset length is 4096, that is, when the length of the time series data is 4096, it is divided into 7 sub-sequences of length 1024 according to a sliding window size of 512 and a repetition rate of 50%. The sliding window size and repetition rate can be selected and adjusted based on actual needs.
[0055] At the same time, regarding the feature extraction of the subsequence data, the time domain features that need to be extracted include but are not limited to: mean, median, root mean square, variance, peak, kurtosis index, margin index, pulse index, line integral, peak-to-peak value, information entropy, peak index, waveform index and skewness index; the frequency domain features that need to be extracted include but are not limited to: spectrum peak, spectrum energy and power spectrum energy; time-frequency features are collected by calculating continuous wavelet transform.
[0056] It is further necessary to understand that, in this embodiment, when performing feature fusion, each of the subsequence data is subjected to continuous wavelet transform and two-dimensional image conversion respectively, and feature fusion is performed based on the corresponding feature extraction result and the corresponding two-dimensional image, including: each of the subsequence data is subjected to continuous wavelet transform and two-dimensional image conversion respectively to obtain the corresponding two-dimensional image; after normalizing the corresponding feature extraction result and the time series data, the obtained normalized data is written into a specific position in the corresponding two-dimensional image to complete the corresponding feature fusion operation. That is, after obtaining the feature extraction result, each of the subsequence data is subjected to continuous wavelet transform and converted into a two-dimensional image, and the corresponding feature extraction result and the original time series are written into a specific position of the image after normalization to generate a new feature map, i.e., the training set data, and the acquisition of the test set data is processed in the same way.
[0057] Step S13: based on the preset verification set division rule, the test set, the preset division ratio and the similarity deviation algorithm, the initial training set is screened to obtain a corresponding verification set and a target training set.
[0058] In this embodiment, considering that the data of the initial training set and the test set come from different mechanical equipment or different parts, there is a problem of inconsistent data distribution even within the same fault category. In order to improve the prediction accuracy and generalization ability of the model, this embodiment uses the similarity deviation method to obtain the verification set. The similarity deviation is a similarity criterion method that measures the degree of similarity from the perspective of shape and value. Its calculation expression C ij for:
[0059] ;
[0060] Wherein, i and j represent two data samples (i.e., the feature graphs, hereinafter referred to as samples), respectively; R ij Indicates the degree of dispersion of the data points of two samples from the average difference, which can reflect the similarity of the shapes between the two samples; D ij It represents the average distance between two samples, which can reflect the similarity between the values of the two samples. ij With D ij The expression for the table is as follows:
[0061] ;
[0062] ;
[0063] ;
[0064] ;
[0065] Among them, X ik and X jk are the kth data points of the two samples respectively; H ij Represents the total average difference between two samples. In summary, it can be seen that the similarity describes the similarity of the shape and value between two samples at the same time. The smaller the value, the higher the similarity of the two samples, and vice versa. In a specific embodiment, the scheme for screening the validation set based on the above method is as follows: (1) The similarity value between each sample in the test set and each sample in the initial training set is calculated by extracting time-frequency domain features, and the 100 candidate data samples with the highest similarity are retained; (2) For each candidate sample in the test set, the similarity value is calculated using the corresponding original vibration waveform data, and the sample with the highest similarity is retained as the target sample; (3) For the N extracted initial training set data that are most similar to the test set data, that is, the target sample, randomly select M (M≤N, M and N meet the preset division ratio) pieces as the validation set, and the remaining data are included in the training data to obtain the target training set.
[0066] Step S14: construct a model by combining the down-sampled wavelet transform layer with the preset Tranform Vision network to obtain a corresponding initial neural network model.
[0067] In this embodiment, it should be understood that the model construction by combining the down-sampled wavelet transform layer with the preset TranformVision network to obtain the corresponding initial neural network model includes: replacing the down-sampled wavelet transform layer with a convolution layer with a step size of 1 and a pooling layer with a step size of 2 to complete the corresponding wavelet integration preset Tranform Vision network design operation to obtain the corresponding initial neural network model. Figure 2 As shown in Figure 2, in order to suppress the influence of noise on the final prediction results and improve the classification accuracy of fault prediction, this embodiment integrates wavelet into Figure 2 The Tranform Vision (TrV) architecture shown in , that is, the preset Tranform Vision network. The downsampling wavelet transform layer (DWT, Downsampling Wavelet Transform) is converted into a general network layer and replaces the commonly used convolution layer, that is, the downsampling layer, and a wavelet integrated TrV network (WaveTrV model for short), that is, the initial neural network model, is designed. In this way, during downsampling, the WaveTrV model can eliminate the high-frequency components of the feature map to improve the noise robustness of the model, and then extract high-level features from the low-frequency components to achieve better classification accuracy. The calculation formula for the low-frequency component of image X is as follows:
[0068] ;
[0069] ;
[0070] Where, X ll represents the low-frequency component of image X, L is the low-frequency filter matrix of orthogonal wavelet, L T is the transposed matrix of L. In this embodiment, Haar wavelet is selected, which can be selected and adjusted based on actual needs. In addition, in the WaveTrV model, the convolution layer with a step size of 1 and the pooling layer with a step size of 2 are replaced by the wavelet transform layer as the first two layers of feature extraction after the image input, which can retain the low-frequency components, and then the fault classification is calculated by the multi-layer perceptron layer (MLP) after 12 TrV layers.
[0071] Step S15: After completing the model training corresponding to the initial neural network model based on the target training set, model verification is performed based on the obtained trained model and the verification set to determine the corresponding target fault diagnosis model when the verification passes, so as to trigger the corresponding fault diagnosis operation using the target fault diagnosis model.
[0072] Specifically, in this embodiment, the target training set is used to perform model training corresponding to the initial neural network model, and after the training is completed, the validation set is used to verify the performance of the trained model. In the process of verification, the classification result of a sample is obtained by weighted majority voting after the WaveTrV model training is completed. That is, the model verification based on the obtained trained model and the validation set includes: for any target feature map in the validation set, the corresponding low-frequency feature extraction is performed through the down-sampled wavelet transform layer in the trained model to obtain the corresponding extraction result; after the extraction result is deep-featured mined through several preset Tranform Vision network layers in the trained model, fault diagnosis and weighted calculation are performed through the multi-layer perceptron layer in the trained model and the obtained deep mining result to obtain the target fault diagnosis result corresponding to the target feature map, and complete the corresponding verification operation. And the fault diagnosis and weighted calculation are performed through the multi-layer perceptron layer in the trained model and the deep mining results obtained, including: after the fault diagnosis is performed, the weighted result corresponding to the feature graph is determined based on the initial diagnosis result corresponding to the target feature graph and a preset weighting formula; wherein the initial diagnosis result includes the fault classification result of each subsequence data corresponding to the target feature graph; and the target fault diagnosis result corresponding to the target feature graph is obtained by rounding down the weighted result.
[0073] In a specific embodiment, the classification result of a sample is obtained by weighting the classification results of its 7 subsequence data according to the weight of each category in the training set. The preset weighting formula is as follows:
[0074] ;
[0075] Among them, L i is the category label of sample i; c represents the category label; W c represents the weight of category c in the training set; N ic Indicates the number of samples in the 7 subsequences of sample i that are predicted to be category c; the weighted result is rounded down as the final prediction result.
[0076] It can be seen that in the embodiment of the present application, the vibration waveform data corresponding to each time node of the rolling bearing is obtained, and the corresponding data preprocessing operation is performed to obtain the corresponding time series data that meets the first preset length; based on the preset feature extraction and fusion rules, the time series data is subjected to corresponding data sliding window division, feature extraction, feature fusion and image conversion to obtain the corresponding initial training set and test set; based on the preset verification set division rules, the test set, the preset division ratio and the similarity deviation algorithm, the initial training set is subjected to sample screening to obtain the corresponding verification set and the target training set; the model is constructed by combining the down-sampling wavelet transform layer with the preset Tranform Vision network to obtain the corresponding initial neural network model; after completing the model training corresponding to the initial neural network model based on the target training set, the model verification is performed based on the obtained trained model and the verification set to determine the corresponding target fault diagnosis model when the verification passes, so as to trigger the corresponding fault diagnosis operation using the target fault diagnosis model. That is, in this application, the vibration waveform data of the rolling bearing is collected and the training set and the verification set are constructed using the similarity deviation algorithm, and then the model is constructed through the preset Tranform Vision network, and then the model is trained and verified, so as to use the verified target fault diagnosis model for diagnosis. In this way, the data collection cost can be effectively reduced, the adverse effects caused by the inconsistent data distribution of the training set and the test set can be solved, and the noise robustness of the model as well as the accuracy and generalization ability of fault diagnosis can be improved.
[0077] As can be seen from the previous embodiment, the present application determines the validation set from the training set by calculating the similarity deviation value between the test set and each feature graph in the initial training set. To this end, the present embodiment will describe in detail the process of determining the validation set based on the similarity deviation value. Figure 3 As shown, an embodiment of the present invention discloses a rolling bearing fault diagnosis method, comprising:
[0078] Step S21: Calculate the similarity deviation value between each feature graph in the test set and each feature graph in the initial training set based on the corresponding feature extraction result to obtain a corresponding first similarity deviation value.
[0079] Step S22: determining a plurality of candidate samples corresponding to each of the feature graphs in the test set from the initial training set according to a preset sample screening condition and each of the first similarity deviation values.
[0080] Step S23, for several candidate samples corresponding to each of the feature graphs in the test set, calculate similarity deviation values based on the corresponding vibration waveform data, and determine the target samples corresponding to each of the feature graphs in the test set according to the obtained second similarity deviation values.
[0081] Step S24: Select each of the target samples based on a preset division ratio and in a random selection manner to obtain a corresponding verification set and a target training set.
[0082] Among them, the specific process of the above steps S21 to S24 can refer to the corresponding content disclosed in the above embodiments, and will not be repeated here.
[0083] It can be seen that in the embodiment of the present application, the first screening is completed by calculating the similarity value between each feature graph in the test set and each feature graph in the initial training set, and then the similarity value is calculated based on the corresponding vibration waveform data for several candidate samples corresponding to each feature graph in the test set to complete the second screening, and finally the target samples obtained by the second screening are randomly selected based on the corresponding division ratio to obtain the final verification set. In this way, the problems of large data distribution differences and poor generalization ability can be solved, and the prediction accuracy and generalization ability of the model can be improved.
[0084] See also Figure 4 As shown, the embodiment of the present application also discloses a rolling bearing fault diagnosis device, including:
[0085] The data preprocessing module 11 is used to obtain the vibration waveform data corresponding to each time node of the rolling bearing and perform corresponding data preprocessing operations to obtain corresponding time series data that meets the first preset length;
[0086] A feature extraction and fusion module 12 is used to perform corresponding data sliding window division, feature extraction, feature fusion and image conversion on the time series data based on preset feature extraction and fusion rules to obtain corresponding initial training sets and test sets;
[0087] A verification set acquisition module 13 is used to screen samples of the initial training set based on a preset verification set division rule, the test set, a preset division ratio, and a similarity degree algorithm to obtain a corresponding verification set and a target training set;
[0088] A model building module 14 is used to build a model by combining the down-sampled wavelet transform layer with the preset Tranform Vision network to obtain a corresponding initial neural network model;
[0089] The target model acquisition module 15 is used to perform model verification based on the obtained trained model and the verification set after completing the model training corresponding to the initial neural network model based on the target training set, so as to determine the corresponding target fault diagnosis model when the verification passes, so as to trigger the corresponding fault diagnosis operation using the target fault diagnosis model.
[0090] Among them, for more specific working processes of the above-mentioned modules, please refer to the corresponding contents disclosed in the aforementioned embodiments, which will not be repeated here.
[0091] It can be seen that in this application, the vibration waveform data of the rolling bearing is collected and the training set and the verification set are constructed using the similarity deviation algorithm, and then the model is constructed through the preset Tranform Vision network, and then the model is trained and verified, so as to use the verified target fault diagnosis model for diagnosis. In this way, the data collection cost can be effectively reduced, the adverse effects caused by the inconsistent data distribution between the training set and the test set can be solved, the noise robustness of the model and the accuracy and generalization ability of fault diagnosis can be improved, and the intelligent level of industrial equipment operation and maintenance can be improved.
[0092] In some specific embodiments, the feature extraction and fusion module 12 may specifically include:
[0093] A subsequence acquisition unit, configured to divide each of the time series data based on a preset sliding window, a preset repetition rate, and a second preset length, so as to obtain a plurality of subsequence data corresponding to each of the time series data;
[0094] A feature extraction unit, used to extract the time domain features, frequency domain features and time-frequency features of each of the subsequence data based on a preset feature extraction rule to obtain a corresponding feature extraction result;
[0095] The feature fusion unit is used to perform continuous wavelet transform and two-dimensional image conversion on each of the subsequence data, and perform feature fusion based on the corresponding feature extraction results and the corresponding two-dimensional images to obtain feature maps corresponding to each of the time series data.
[0096] In some specific embodiments, the feature fusion unit may specifically include:
[0097] A two-dimensional picture acquisition subunit, used to perform continuous wavelet transform and two-dimensional picture conversion on each of the subsequence data to obtain a corresponding two-dimensional picture;
[0098] The data writing subunit is used to write the obtained normalized data into a specific position in the corresponding two-dimensional image after normalizing the corresponding feature extraction result and the time series data, so as to complete the corresponding feature fusion operation.
[0099] In some specific embodiments, the verification set acquisition module 13 may specifically include:
[0100] A first similarity deviation value calculation unit, used for calculating the similarity deviation value between each of the feature graphs in the test set and each of the feature graphs in the initial training set based on the corresponding feature extraction result, so as to obtain a corresponding first similarity deviation value;
[0101] A candidate sample acquisition unit, configured to determine, from the initial training set, a number of candidate samples corresponding to each of the feature graphs in the test set according to a preset sample screening condition and each of the first similarity deviation values;
[0102] a target sample determination unit, configured to calculate similarity deviation values for a plurality of candidate samples corresponding to the feature graphs in the test set based on the corresponding vibration waveform data, and determine target samples corresponding to the feature graphs in the test set according to the obtained second similarity deviation values;
[0103] The sample random selection unit is used to select each of the target samples based on a preset division ratio and in a random selection manner to obtain a corresponding verification set and a target training set.
[0104] In some specific embodiments, the model building module 14 may specifically include:
[0105] The model building unit is used to complete the corresponding wavelet integration preset Tranform Vision network design operation by replacing the down-sampling wavelet transform layer with the convolution layer with a step size of 1 and the pooling layer with a step size of 2 to obtain the corresponding initial neural network model.
[0106] In some specific embodiments, the target model acquisition module 15 may specifically include:
[0107] The extraction result obtains the anesthetic, which is used to extract the corresponding low-frequency features through the down-sampling wavelet transform layer in the trained model for any target feature map in the verification set to obtain the corresponding extraction result;
[0108] A verification unit is used to perform deep feature mining on the extraction result through several preset Tranform Vision network layers in the trained model, and then perform fault diagnosis and weighted calculation through the multi-layer perceptron layer in the trained model and the obtained deep mining result to obtain a target fault diagnosis result corresponding to the target feature map, and complete a corresponding verification operation.
[0109] In some specific embodiments, the verification unit may specifically include:
[0110] A weighted result acquisition subunit is used to determine, after fault diagnosis, a weighted result corresponding to the characteristic graph based on an initial diagnosis result corresponding to the target characteristic graph and a preset weighting formula; wherein the initial diagnosis result includes a fault classification result of each of the subsequence data corresponding to the target characteristic graph;
[0111] The rounding-down subunit is used to obtain the target fault diagnosis result corresponding to the target characteristic graph by rounding down the weighted result.
[0112] 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.
[0113] Figure 5 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 rolling bearing 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.
[0114] 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.
[0115] 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.
[0116] The operating system 221 is used to manage and control the hardware devices and computer program 222 on the electronic device 20, which can be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program that can be used to complete the rolling bearing 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.
[0117] 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 rolling bearing fault diagnosis method disclosed above is implemented. The specific steps of the method can refer to the corresponding contents disclosed in the above embodiments, and will not be repeated here.
[0118] 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.
[0119] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in 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 the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians 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.
[0120] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
[0121] 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.
[0122] The technical solution provided by the present application is introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for general technicians in this field, according to the idea of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.
Claims
1. A rolling bearing fault diagnosis method, It is characterized in that include: By acquiring the vibration waveform data corresponding to each time node of the rolling bearing and performing corresponding data preprocessing operations, corresponding time series data satisfying a first preset length is obtained; Based on preset feature extraction and fusion rules, the time series data is subjected to corresponding data sliding window division, feature extraction, feature fusion and image conversion to obtain corresponding initial training sets and test sets; Based on the preset validation set division rule, the test set, the preset division ratio and the similarity deviation algorithm, the initial training set is screened to obtain a corresponding validation set and a target training set; The model is constructed by combining the down-sampled wavelet transform layer with the preset Tranform Vision network to obtain the corresponding initial neural network model; After completing the model training corresponding to the initial neural network model based on the target training set, model verification is performed based on the obtained trained model and the verification set to determine the corresponding target fault diagnosis model when the verification passes, so as to trigger the corresponding fault diagnosis operation using the target fault diagnosis model.
2. The rolling bearing fault diagnosis method according to claim 1, It is characterized in that The corresponding data sliding window division, feature extraction, feature fusion and image conversion of the time series data based on the preset feature extraction and fusion rules include: Based on a preset sliding window, a preset repetition rate and a second preset length, each of the time series data is divided into data to obtain a plurality of subsequence data corresponding to each of the time series data; Based on the preset feature extraction rules, the time domain features, frequency domain features and time-frequency features of each of the subsequence data are respectively extracted to obtain corresponding feature extraction results; Continuous wavelet transform and two-dimensional image conversion are performed on each of the subsequence data respectively, and feature fusion is performed with the corresponding two-dimensional image based on the corresponding feature extraction result to obtain feature graphs corresponding to each of the time series data respectively.
3. The rolling bearing fault diagnosis method according to claim 2, It is characterized in that The performing continuous wavelet transform and two-dimensional image conversion on each of the subsequence data respectively, and performing feature fusion with the corresponding two-dimensional image based on the corresponding feature extraction result, comprises: Performing continuous wavelet transform and two-dimensional image conversion on each of the subsequence data to obtain a corresponding two-dimensional image; After normalizing the corresponding feature extraction results and the time series data, the obtained normalized data is written into a specific position in the corresponding two-dimensional image to complete the corresponding feature fusion operation.
4. The rolling bearing fault diagnosis method according to claim 2, It is characterized in that The sample screening of the initial training set based on the preset verification set division rule, the test set, the preset division ratio and the similarity deviation algorithm to obtain the corresponding verification set includes: Calculating a similarity deviation value between each of the feature graphs in the test set and each of the feature graphs in the initial training set based on the corresponding feature extraction results to obtain a corresponding first similarity deviation value; Determine, from the initial training set, a number of candidate samples corresponding to each of the feature graphs in the test set according to a preset sample screening condition and each of the first similarity deviation values; For a plurality of candidate samples corresponding to each of the feature graphs in the test set, similarity deviation values are calculated based on the corresponding vibration waveform data, and target samples corresponding to each of the feature graphs in the test set are determined according to the obtained second similarity deviation values; Based on a preset division ratio and in a random selection manner, each of the target samples is selected to obtain a corresponding verification set and a target training set.
5. The rolling bearing fault diagnosis method according to claim 1, It is characterized in that The model is constructed by combining the down-sampling wavelet transform layer with the preset Tranform Vision network to obtain a corresponding initial neural network model, including: By replacing the down-sampling wavelet transform layer with the convolution layer with a step size of 1 and the pooling layer with a step size of 2 to complete the corresponding wavelet integration preset Tranform Vision network design operation, the corresponding initial neural network model is obtained.
6. The rolling bearing fault diagnosis method according to claim 4, It is characterized in that The performing model verification based on the obtained trained model and the verification set includes: For any target feature map in the validation set, corresponding low-frequency feature extraction is performed through the down-sampling wavelet transform layer in the trained model to obtain corresponding extraction results; After deep feature mining is performed on the extraction results through several preset Tranform Vision network layers in the trained model, fault diagnosis and weighted calculation are performed through the multi-layer perceptron layer in the trained model and the obtained deep mining results to obtain a target fault diagnosis result corresponding to the target feature map, and complete the corresponding verification operation.
7. The rolling bearing fault diagnosis method according to claim 6, It is characterized in that The method of performing fault diagnosis and weighted calculation by using the multi-layer perceptron layer in the trained model and the obtained deep mining results includes: After performing fault diagnosis, a weighted result corresponding to the characteristic graph is determined based on an initial diagnosis result corresponding to the target characteristic graph and a preset weighting formula; wherein the initial diagnosis result includes a fault classification result of each of the subsequence data corresponding to the target characteristic graph; The target fault diagnosis result corresponding to the target characteristic graph is obtained by rounding down the weighted result.
8. A rolling bearing fault diagnosis device, It is characterized in that include: A data preprocessing module, used to obtain vibration waveform data corresponding to each time node of the rolling bearing and perform corresponding data preprocessing operations to obtain corresponding time series data that meets a first preset length; A feature extraction and fusion module is used to perform corresponding data sliding window division, feature extraction, feature fusion and image conversion on the time series data based on preset feature extraction and fusion rules to obtain corresponding initial training sets and test sets; A validation set acquisition module is used to screen samples of the initial training set based on a preset validation set partitioning rule, the test set, a preset partitioning ratio, and a similarity deviation algorithm to obtain a corresponding validation set and a target training set; A model building module, used for building a model by combining a down-sampled wavelet transform layer with a preset Tranform Vision network to obtain a corresponding initial neural network model; The target model acquisition module is used to perform model verification based on the obtained trained model and the verification set after completing the model training corresponding to the initial neural network model based on the target training set, so as to determine the corresponding target fault diagnosis model when the verification passes, so as to trigger the corresponding fault diagnosis operation using the target fault diagnosis model.
9. An electronic device, It is characterized in that include: Memory, used to store computer programs; A processor, used to execute the computer program to implement the rolling bearing fault diagnosis method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, It is characterized in that Used to store a computer program, which, when executed by a processor, implements the rolling bearing fault diagnosis method as described in any one of claims 1 to 7.
Citation Information
Cited By
Method, device and equipment for monitoring running state of electromagnetic suspension system and medium
CN119988912A
Electric energy quality disturbance identification method based on self-supervised contrast learning
CN120372254A
