CN116399592A - Bearing fault diagnosis method based on channel attention dual-path feature extraction - Google Patents
Bearing fault diagnosis method based on channel attention dual-path feature extraction Download PDFInfo
- Publication number
- CN116399592A CN116399592A CN202310402758.7A CN202310402758A CN116399592A CN 116399592 A CN116399592 A CN 116399592A CN 202310402758 A CN202310402758 A CN 202310402758A CN 116399592 A CN116399592 A CN 116399592A
- Authority
- CN
- China
- Prior art keywords
- feature extraction
- fault diagnosis
- bearing
- fault
- path
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000605 extraction Methods 0.000 title claims abstract description 80
- 238000003745 diagnosis Methods 0.000 title claims abstract description 72
- 238000000034 method Methods 0.000 title claims abstract description 49
- 230000004927 fusion Effects 0.000 claims abstract description 17
- 238000007781 pre-processing Methods 0.000 claims abstract description 5
- 230000009977 dual effect Effects 0.000 claims description 16
- 238000012549 training Methods 0.000 claims description 13
- 238000012360 testing method Methods 0.000 claims description 12
- 238000011176 pooling Methods 0.000 claims description 7
- 238000005070 sampling Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 4
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 238000012935 Averaging Methods 0.000 claims description 2
- 239000000284 extract Substances 0.000 abstract description 5
- 239000010410 layer Substances 0.000 description 39
- 230000006870 function Effects 0.000 description 20
- 238000010586 diagram Methods 0.000 description 18
- 238000013527 convolutional neural network Methods 0.000 description 15
- 238000013528 artificial neural network Methods 0.000 description 9
- 238000004458 analytical method Methods 0.000 description 8
- 230000015654 memory Effects 0.000 description 8
- 230000007246 mechanism Effects 0.000 description 7
- 238000012795 verification Methods 0.000 description 7
- 230000004913 activation Effects 0.000 description 6
- 238000013135 deep learning Methods 0.000 description 5
- 230000007547 defect Effects 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 5
- 230000009466 transformation Effects 0.000 description 5
- 238000012545 processing Methods 0.000 description 4
- 238000001514 detection method Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 238000012706 support-vector machine Methods 0.000 description 3
- 238000012952 Resampling Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000000354 decomposition reaction Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 230000007787 long-term memory Effects 0.000 description 2
- 238000005096 rolling process Methods 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 230000002123 temporal effect Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000012631 diagnostic technique Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000010892 electric spark Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 238000003754 machining Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 238000002759 z-score normalization Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- General Physics & Mathematics (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention relates to a bearing fault diagnosis method based on channel attention dual-path feature extraction, which comprises the following steps: and acquiring bearing vibration signals, preprocessing the bearing vibration signals, inputting the bearing vibration signals into a pre-trained fault diagnosis model, and acquiring output fault classification identification information to realize fault diagnosis. The fault diagnosis model comprises a feature extraction network for extracting double-path features, a feature fusion network and a secondary extraction and classification network. Compared with the prior art, the method aims at the problems of insufficient feature extraction and the like in bearing fault diagnosis, and extracts the time features and the space features of fault signals from two angles of a time domain and a frequency domain respectively, so that the capability of capturing and extracting the fault features of a model is effectively improved, and better diagnosis precision, generalization and robustness are achieved.
Description
Technical Field
The invention relates to the technical field of fault diagnosis, in particular to a bearing fault diagnosis method based on channel attention dual-path feature extraction.
Background
The operation and maintenance of mechanical devices tends to be automated, intelligent, which increases the difficulty of managing and fault monitoring the devices. The bearing is one of key components in the rotary machine, and is poor in working condition environment and easy to fail. Once it fails, it can cause serious system crashes, equipment damage, significant economic losses, etc. Therefore, the effective fault analysis of the bearing operation state is beneficial to timely finding out the mechanical equipment fault and ensuring the reliable and stable operation of the mechanical system.
Automated collection of bearing equipment operational state data has prompted the development of diagnostic techniques to data-based drives. At present, detection information of bearing fault diagnosis mainly comes from vibration signals, wherein the bearing vibration signals are periodically changed time sequence data and comprise various state information of the bearing. In recent decades, with vibration signals as detection signals, the fault diagnosis of bearings mainly goes through three development stages: (1) fault diagnosis based on signal analysis; (2) fault diagnosis based on traditional machine learning; (3) deep learning-based fault diagnosis.
Methods based on signal analysis and related theories have been widely studied in the field of fault diagnosis, and common time-frequency analysis methods include EMD (empirical mode decomposition), VMD (variational mode decomposition), STFT (short-time fourier transform) and WT (wavelet transform). The fourier transform acts on the time domain of the signal and only the frequency content of the signal can be obtained, but the local characteristics of the signal cannot be obtained. The method of windowing is adopted to improve Fourier transform to obtain short-time Fourier transform, so that local characteristics of signals can be extracted, and the window adopted by STFT is fixed and is not suitable for unsteady time-varying signals. The WT is based on STFT, overcomes the defect that an STFT window does not change along with frequency, and can automatically adapt to the change of signal frequency, so that the localization analysis of the signal frequency is realized, and the method is widely applied to the time-frequency analysis and processing of the signal.
The fault diagnosis method based on traditional machine learning is often used in combination with a signal analysis method, and usually, vibration data are preprocessed, artificial features are extracted, and then a classifier is used for completing fault classification. The machine learning algorithm is commonly used for fault recognition and comprises a Support Vector Machine (SVM), a Bayesian classifier (Bayesian classifier), an Artificial Neural Network (ANN) and the like. SVM, ANN, etc. are mostly shallow models, and diagnostic performance is limited by the quality of the extracted fault features. The data volume of the sensor which is dynamically collected in real time is very large, the traditional machine learning method cannot fully utilize the data, and the fault diagnosis capability is weakened along with the increase of the data volume.
Deep learning methods (such as convolutional neural networks, recurrent neural networks, deep confidence networks, generation of countermeasure networks, etc.) have shown great advantages in terms of image classification, natural language processing, target detection, etc. as a development extension of machine learning. The deep learning network has excellent characteristic information characterization capability, deep characteristic extraction of data can be realized by constructing a deep network structure, and complex nonlinear mapping from a preprocessing data set to a fault state can be completed. Among them, convolutional Neural Networks (CNNs) have achieved good results in the field of fault diagnosis. Although CNN can realize spatial feature extraction of data, vibration signals are time-series data, and feature extraction is only performed from a local spatial angle, so that fault features cannot be fully mined. The cyclic neural network (RNN) has better performance in processing time sequence data and is suitable for fault diagnosis. Although the characteristics of the bearing vibration signals cannot be fully extracted by the traditional single depth network structure, partial key information is lost; secondly, the model uses only time domain data or frequency domain data as input, resulting in insufficient extraction of fault space-time features by the model.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a bearing fault diagnosis method based on channel attention dual-path feature extraction, which extracts the time feature and the space feature of fault signals from two angles of a time domain and a frequency domain respectively so as to improve the diagnosis precision.
The aim of the invention can be achieved by the following technical scheme:
the invention provides a bearing fault diagnosis method based on channel attention dual-path feature extraction, which comprises the following steps:
the method comprises the steps of obtaining a bearing vibration signal, inputting the bearing vibration signal into a pre-trained fault diagnosis model after preprocessing, obtaining output fault classification identification information, and realizing fault diagnosis, wherein the fault diagnosis model comprises the following components:
the characteristic extraction network for extracting the double-path characteristics comprises a first path and a second path, wherein the first path acquires signal global time characteristics based on an input bearing vibration signal, the second path acquires two-dimensional time-frequency information based on the input bearing vibration signal, a channel weight is determined through an attention module according to the two-dimensional time-frequency information, and signal local space characteristics are acquired based on the two-dimensional time-frequency information and the channel weight;
the feature fusion network is used for acquiring signal fusion features based on the signal global time features and the signal local space features;
and the secondary extraction and classification network is used for carrying out secondary extraction based on the signal fusion characteristics and outputting fault classification identification information based on the signal characteristics after the secondary extraction.
As a preferred technical solution, the process for obtaining the pre-trained fault diagnosis model includes the following steps:
obtaining a bearing vibration fault signal, performing sample expansion and label setting, and obtaining a standardized and divided data set;
based on each sample in the data set, acquiring corresponding two-dimensional time-frequency information;
and training the fault diagnosis model based on the data set and the corresponding two-dimensional time-frequency information to obtain a pre-trained fault diagnosis model.
As a preferable technical solution, after training the fault diagnosis model, the method further includes:
and fine tuning the model parameters by using the test set in the divided data set.
As a preferred technical scheme, the standardization is Z-score standardization, and sample expansion is performed by adopting an overlapping sampling mode.
As a preferable technical scheme, the channel weight is obtained through adaptive one-dimensional convolution operation, and the convolution kernel size of the one-dimensional convolution is determined by adopting the following formula:
wherein k is the size of a one-dimensional convolution kernel, C is the dimension of an input channel, and gamma and b are preset parameters.
As a preferable technical scheme, the attention module comprises a global average pooling layer and a one-dimensional convolution layer which are sequentially connected.
As a preferred technical solution, the two-dimensional time-frequency diagram is obtained by continuous wavelet transformation.
As a preferred technical solution, in the feature extraction network for dual-path feature extraction, the first path includes two LSTM subnetworks connected in sequence.
As an preferable technical solution, in the feature extraction network for dual-path feature extraction, the second path includes two 2D-CNN subnetworks connected in sequence.
The secondary extraction and classification network is used for carrying out secondary extraction based on the signal fusion characteristics to obtain final signal characteristics, and outputting fault classification identification information through a classifier based on the final signal characteristics.
Compared with the prior art, the invention has the following advantages:
(1) Fault diagnosis precision, generalization and robustness are high: the method aims at bearing vibration signals to be diagnosed, and a feature extraction network for extracting dual-path features is used for respectively extracting time features and space features of fault signals from two angles of a time domain and a frequency domain, so that fusion features are obtained, and the influence of certain feature extraction on subsequent operation is avoided. Meanwhile, the fused characteristic information can represent global characteristics of signals to the greatest extent, and finally, the fused characteristics are extracted and classified for the second time, so that a final diagnosis result is obtained. In order to make up for the defect of single time domain and frequency domain characteristics of vibration signals in fault diagnosis, the model adopts a double-input structure, and not only considers the time dependence of the vibration signals, but also considers the characteristics that the signals can be expressed in the frequency domain when impacted. And after the signal fusion characteristics are acquired, performing secondary extraction to acquire final signal characteristics, and further extracting the characteristics of the bearing vibration signals. Compared with the prior art, the method aims at solving the problems of insufficient feature extraction and the like in bearing fault diagnosis, learns local correlation and global context information from vibration signals, aims to greatly improve the aspects of automatic fault feature extraction, diagnosis accuracy, generalization and the like of a model, extracts fault signals from two angles of a time domain and a frequency domain respectively, can effectively improve the capability of capturing and extracting the fault features of the model, and has better diagnosis accuracy, generalization and robustness.
(2) The data set is easy to acquire: sample expansion using overlapping sampling may be used to obtain a sufficient number of data sets with a small amount of fault data to train the model.
Drawings
FIG. 1 is a flow chart of a bearing fault diagnosis method based on channel attention dual path feature extraction in embodiment 1;
FIG. 2 is a schematic diagram of sample expansion using resampling;
FIG. 3 is a schematic diagram of a channel attention mechanism module structure;
FIG. 4 is a diagram of a fault diagnosis network;
fig. 5 is a one-dimensional time domain plot of vibration signals for different types of fault samples of the data set used in case verification;
fig. 6 is a schematic diagram of confusion matrix of test sets under different loads in case verification of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
The terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present application, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.
The invention aims to solve the problems of low fault diagnosis precision and low efficiency caused by insufficient extraction of time and space characteristics of fault information in the fault diagnosis of a bearing by using a deep learning method, and provides a bearing fault diagnosis method based on channel attention dual-path characteristic extraction
As shown in fig. 1, the technical scheme of the invention is as follows: a bearing fault diagnosis method based on channel attention dual-path feature extraction is characterized in that vibration data are preprocessed to obtain time domain data and a time-frequency diagram, space features and time features of vibration signals are extracted from a dual-path view angle, a fault diagnosis network based on the dual-path feature extraction of a convolutional neural network and a long-short-time memory network is constructed by fusing attention mechanisms, and the signals are subjected to feature extraction and feature fusion and then are subjected to fault classification by using a classifier, and the method mainly comprises the following steps:
step S1: and expanding a sample by using a data enhancement method according to the vibration signal of the bearing under a certain working condition, and setting a corresponding label for the sample. Dividing the data set into a training set, a verification set and a test set, and carrying out standardized processing on the data set;
the normalization method adopted in the step S1 is a Z-score normalization method, so that the processed data accords with standard normal distribution, and the formula is expressed as follows:
where x is the raw data, μ is the mean of the set of data, σ is the standard deviation of the set of data,
is a normalized value.
Step S2: converting the sample processed in the step S1 into a two-dimensional time-frequency diagram by wavelet transformation, and corresponding the two-dimensional time-frequency diagram to a one-dimensional data label;
step S3: constructing a feature extraction network for extracting the dual-path features, wherein an upper layer uses a long-short-time memory network to extract the global time features of signals, and a lower layer uses a two-dimensional convolutional neural network to extract the local space features of the signals;
step S4: introducing a channel attention module into the convolutional neural network, and distributing characteristic weights;
in the step S4, the channel attention module uses the global average pooling, the one-dimensional convolution layer and the sigmoid activation function to implement the weighted feature map.
Step S5: fusing the features extracted by the dual paths in the step S3;
step S6: and taking the fused features as input of a one-dimensional convolutional neural network for secondary extraction, and finally completing fault classification and identification through a softmax classifier.
In the fault diagnosis network constructed in the steps S3, S4, S5 and S6, the dual-path characteristic extraction network is composed of two branches of a double-layer LSTM network and a double-layer 2D-CNN network; the fault identification consists of a layer of convolution block, a flat layer and two full-connection layers. The convolution layer, the long-short-time memory network and the first full-connection layer use a Relu activation function, the second full-connection layer uses a softmax function as an activation function, and the softmax function converts a classification result into a probability to realize fault diagnosis of the bearing.
Step S7: and (3) utilizing the training set to perform fault diagnosis network constructed in the steps S3, S4, S5 and S6, and performing fine adjustment and storage on model parameters through the testing set to complete fault diagnosis model training under different working conditions of the bearing.
For vibration signal data of one-dimensional nonlinearity, single time domain feature analysis is not efficient for diagnosing faults. In addition, a single one-dimensional convolutional neural network can only extract the spatial characteristics of data, and has poor processing effect on serialized data. In order to make up for the defect that the feature extraction of the vibration signal in a single time domain and a single frequency domain is insufficient, the invention extracts the spatial feature and the temporal feature of the vibration signal from the double-path visual angle, provides a bearing fault diagnosis method based on the channel attention double-path feature extraction, and constructs a double-path feature extraction mechanism: a global temporal feature extraction channel and a local spatial feature extraction channel. The former takes one-dimensional original signals as input, and the latter takes a two-dimensional time-frequency diagram after CWT conversion as input. And carrying out feature fusion on the signals after the two-channel feature extraction, and completing fault identification diagnosis by utilizing the fused feature information.
Example 1
As shown in fig. 1, the present embodiment provides a bearing fault diagnosis method based on channel attention dual-path feature extraction. In order to improve the feature extraction capability of the vibration signals, the embodiment provides a dual-path fault diagnosis model, local correlation and global context information are learned from the vibration signals, a network structure extracted from space and time parallel features is constructed, feature fusion is carried out on the signals after the two-channel feature extraction, and fault identification diagnosis is completed by utilizing the fused feature information.
In the invention, the original vibration signal is expanded into a sample by using a data enhancement method, and the data enhancement mode is adopted for overlapped sampling, namely, when the sample is acquired from the original signal, each section of signal and the next section of signal are overlapped and sampled. With a window width w, a repeated sampling frequency r, a step length s, and overlapping sampling along a time axis, the process is as described in fig. 2, and the process is expressed as follows:
w＝r+s
the standardized method adopted after the data set is divided into a training set, a verification set and a test set is a Z-score standardized method, so that the processed data accords with standard normal distribution, and the formula is expressed as follows:
where x is the raw data, μ is the mean of the set of data, σ is the standard deviation of the set of data,
is a normalized value.
The continuous wavelet transformation is adopted to convert one-dimensional signals into time-frequency diagrams to extract fault characteristics, and the formula is as follows:
wherein x is t Is an input signal;
is a mother wavelet function; parameters a and b are two continuous variables, a being the scale parameter (controlling the scaling of the wavelet function) and b being the time parameter (controlling the translation of the wavelet function). The above equation can thus be understood as converting the original one-dimensional signal into a two-dimensional signal with a frequency parameter a and a time parameter b via a wavelet function.
In this embodiment, a cmor wavelet function is selected, the scale parameter a=3, and the time parameter b=3. And performing continuous wavelet transformation on the vibration signals in the training set, the verification set and the test set to obtain a wavelet coefficient matrix.
Aiming at the defect of insufficient extraction of space-time characteristics of bearing faults in the traditional fault diagnosis, the time and space characteristics of fault information are extracted in parallel by constructing a double-path characteristic extraction mechanism, and in order to obtain important characteristics, a channel attention module is integrated in a convolutional neural network, and the attention of the network to the important information is increased by updating weights.
Channel attention module: important information is screened from a large amount of information. As shown in fig. 3, the feature map with dimension of [ h×w×c ] is input, and is subjected to global averaging pooling to become a vector of [1×1×c ], then the weight of each channel of the feature map is obtained through adaptive one-dimensional convolution operation, and finally the normalized weight is multiplied by the weight of the original input feature map to generate a weighted feature map. The channel attention module completes information interaction among the cross channels through the self-adaptive one-dimensional convolution kernel, and can realize information interaction among multiple channels. GAP is the global average pooling layer in FIG. 3, sigmoid is the activation function. The output of the channel attention module is consistent with its input dimension. The above-mentioned mid-dimension convolution kernel size k is adaptively obtained by a function expressed as:
where C is the input channel dimension, γ=2, b=1.
The dual-path characteristic extraction mechanism consists of two branches of a double-layer long-short-time memory network and a double-layer convolution neural network, wherein a channel attention module is integrated in the convolution neural network and is positioned between two convolution blocks.
The network structure design of the bearing fault diagnosis method based on the channel attention dual-path feature extraction is as follows:
the bearing fault diagnosis model designed by the invention is shown in fig. 4. The model adopts a double-input structure, one-dimensional time domain data is input into a long-short-time memory network, and a two-dimensional time-frequency diagram is input into a two-dimensional convolutional neural network. The dual-path characteristic extraction network consists of two branches of a double-layer long-short-time memory network and a double-layer convolution neural network, and the channel attention module is positioned between the double-layer convolution neural network layers; the fault identification consists of a layer of convolution block, a flat layer and two full-connection layers. The convolution layer, the long-short-time memory network and the first full-connection layer use a Relu activation function, the second full-connection layer uses a softmax function as an activation function, and the softmax function converts a classification result into a probability to realize fault diagnosis of the bearing. The experiment sets the batch size to 128, the training times of all training samples are 100 times, a Dropout layer is added before a Softmax layer, the Dropout parameter is set to 0.5, finally an Adam optimizer is designated by a combile function, the optimization coefficient is 0.0001, and a cross entropy loss function is selected by a loss function.
In the model structure of fig. 4, the dual-path Feature extraction network is divided into a CNN path and an LSTM path, LSTM is a long and short term memory network neuron, conv1 and Conv2 are two-dimensional convolution layers, pool1 and Pool2 are two-dimensional pooling layers, feature fusion layer is a Feature fusion layer, BN is a batch layer, conv3 is a one-dimensional convolution layer, pool3 is a one-dimensional pooling layer, flame is a flat layer, and Dense is a full connection layer. CWT represents a continuous wavelet transform. Channel attention the channel attention module.
In the dual-path feature extraction mechanism, one-dimensional time domain data is subjected to feature extraction through a long-term and short-term memory network, and the output can be expressed as [ y ] 1 ,…,y n ]The method comprises the steps of carrying out a first treatment on the surface of the The time-frequency diagram is extracted by a convolutional neural network, and the output is expressed as [ y ] i ,…,y s ]The feature fusion of dual path extraction using concatenation can be represented as [ y ] 1 ,…,y n ,y i ,…,y s ]。
And after the feature fusion layer, performing secondary feature extraction by using a single-layer one-dimensional convolution module, and connecting the full-connection layer, and completing fault classification by a softmax function.
Case verification
The invention uses kesixi Chu Da bearing disclosure data set as shown in table 1. And selecting a driving end bearing with the model of SKF6205 and vibration data with the sampling frequency of 12kHz, and respectively introducing single-point faults with the diameters of 0.007inch, 0.014inch and 0.021inch into an inner raceway, an outer raceway and a rolling body of the bearing by using an electric spark machining technology, wherein the total number of the single-point faults are 9 fault damage state samples and 1 non-damage normal sample. The experiment contained four data sets, each containing 10 data samples, each containing 784 data points, corresponding to data set samples at different loads of 0HP, 1HP, 2HP, 3HP, respectively. The number of samples for each failure type was set to 600, while the training set: verification set: test set = 7:1.5:1.5, each data set is augmented with resampling techniques to sample, and a total of 6000 samples for 10 fault types, divided into 4200 training samples, 900 validation samples, and 900 test samples.
Table 1 kesixi Chu Da bearing dataset
Model data preprocessing: the dual-path feature network constructed in this embodiment requires two different types of input, so that one-dimensional time domain data is converted into a time-frequency diagram through wavelet transformation, as shown in fig. 5, which is a time-domain diagram of four fault type vibration signals in the data set and a corresponding time-frequency diagram. Fig. 5 (a), (c), (e), and (g) show time domain diagrams under normal, inner ring failure, outer ring failure, and rolling element failure, respectively, and (b), (d), (f), and (h) show corresponding time-frequency diagrams, respectively.
In fig. 5, the left graph is a vibration signal one-dimensional time domain graph of a normal state of a bearing and three different types of fault samples, and the right graph is a time-frequency graph of one-dimensional data after CWT conversion.
In order to verify the effectiveness of the model, the fault diagnosis model is realized by Python coding under a Tensorflow deep learning framework.
In fig. 6, (a), (b), (c), and (d) are test set diagnosis confusion matrix result diagrams of loads 0HP, 1HP, 2HP, and 3HP, respectively, with the ordinate true being the true type of fault and the abscissa guess being the model prediction result. As can be seen from fig. 6, the number of samples in each test set is 900, and the number of diagnostic error samples is at most 1, and the diagnostic accuracy of the proposed model under four different load test sets can be higher than 99.50%, which indicates that the model has higher diagnostic accuracy.
Aiming at the problems of insufficient feature extraction and the like in bearing fault diagnosis, the invention provides a bearing fault diagnosis method based on double-path feature extraction of a channel attention mechanism, which is used for extracting time features and space features of fault signals from two angles of a time domain and a frequency domain respectively. The invention can effectively improve the fault feature capturing and extracting capability of the model, and has better diagnosis precision, generalization and robustness.
Example 2
The present embodiment provides an electronic device, including: one or more processors and a memory having stored therein one or more programs including instructions for performing the bearing fault diagnosis method based on channel attention dual path feature extraction as described in embodiment 1.
Example 3
The present embodiment provides a computer-readable storage medium including one or more programs for execution by one or more processors of an electronic device, the one or more programs including instructions for performing the bearing failure diagnosis method based on channel attention dual path feature extraction as described in embodiment 1.
While the invention has been described with reference to certain preferred embodiments, it will be understood by those skilled in the art that various changes and substitutions of equivalents may be made and equivalents will be apparent to those skilled in the art without departing from the scope of the invention. Therefore, the protection scope of the invention is subject to the protection scope of the claims.
Claims (10)
1. A bearing fault diagnosis method based on channel attention dual-path feature extraction is characterized by comprising the following steps:
the method comprises the steps of obtaining a bearing vibration signal, inputting the bearing vibration signal into a pre-trained fault diagnosis model after preprocessing, obtaining output fault classification identification information, and realizing fault diagnosis, wherein the fault diagnosis model comprises the following components:
the characteristic extraction network for extracting the double-path characteristics comprises a first path and a second path, wherein the first path acquires signal global time characteristics based on an input bearing vibration signal, the second path acquires two-dimensional time-frequency information based on the input bearing vibration signal, a channel weight is determined through an attention module according to the two-dimensional time-frequency information, and signal local space characteristics are acquired based on the two-dimensional time-frequency information and the channel weight;
the feature fusion network is used for acquiring signal fusion features based on the signal global time features and the signal local space features;
and the secondary extraction and classification network is used for carrying out secondary extraction based on the signal fusion characteristics to obtain final signal characteristics, and outputting fault classification identification information based on the final signal characteristics.
2. The method for diagnosing bearing faults based on channel attention dual path feature extraction as claimed in claim 1, wherein the process of obtaining the pre-trained fault diagnosis model comprises the following steps:
obtaining a bearing vibration fault signal, performing sample expansion and label setting, and obtaining a standardized and divided data set;
based on each sample in the data set, acquiring corresponding two-dimensional time-frequency information;
and training the fault diagnosis model based on the data set and the corresponding two-dimensional time-frequency information to obtain a pre-trained fault diagnosis model.
3. The method for diagnosing bearing faults based on channel attention dual path feature extraction as claimed in claim 2, wherein the standardization is Z-score standardization, and sample expansion is carried out in an overlapping sampling mode.
4. The method for bearing fault diagnosis based on channel attention dual path feature extraction of claim 2, further comprising, after training the fault diagnosis model:
and fine tuning the model parameters by using the test set in the divided data set.
5. The method for diagnosing a bearing fault based on channel attention dual path feature extraction as recited in claim 1, wherein said channel weights are obtained by an adaptive one-dimensional convolution operation, and a convolution kernel size of said one-dimensional convolution is determined using the following formula:
wherein k is the size of a one-dimensional convolution kernel, C is the dimension of an input channel, and gamma and b are preset parameters.
6. The method for diagnosing a bearing failure based on channel attention dual path feature extraction of claim 1, wherein said attention module comprises a global averaging pooling layer and a one-dimensional convolution layer connected in sequence.
7. The method for diagnosing bearing faults based on channel attention dual path feature extraction as claimed in claim 1, wherein the two-dimensional time-frequency map is obtained through continuous wavelet transform.
8. The method for diagnosing bearing faults based on channel attention dual path feature extraction as claimed in claim 1, wherein the first path comprises two LSTM sub-networks connected in sequence in the feature extraction network of the dual path feature extraction.
9. The method for diagnosing bearing faults based on channel attention dual path feature extraction as claimed in claim 1, wherein the second path comprises two 2D-CNN sub-networks connected in sequence in the feature extraction network of the dual path feature extraction.
10. The method for diagnosing bearing faults based on channel attention dual path feature extraction as claimed in claim 1, wherein the secondary extraction and classification network comprises a secondary extraction sub-network and a classifier, the secondary extraction sub-network comprises a convolution block, and the classifier comprises a flat layer and two full connection layers which are connected in sequence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310402758.7A CN116399592A (en) | 2023-04-14 | 2023-04-14 | Bearing fault diagnosis method based on channel attention dual-path feature extraction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310402758.7A CN116399592A (en) | 2023-04-14 | 2023-04-14 | Bearing fault diagnosis method based on channel attention dual-path feature extraction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116399592A true CN116399592A (en) | 2023-07-07 |
Family
ID=87013962
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310402758.7A Pending CN116399592A (en) | 2023-04-14 | 2023-04-14 | Bearing fault diagnosis method based on channel attention dual-path feature extraction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116399592A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117056814A (en) * | 2023-10-11 | 2023-11-14 | 国网山东省电力公司日照供电公司 | Transformer voiceprint vibration fault diagnosis method |
-
2023
- 2023-04-14 CN CN202310402758.7A patent/CN116399592A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117056814A (en) * | 2023-10-11 | 2023-11-14 | 国网山东省电力公司日照供电公司 | Transformer voiceprint vibration fault diagnosis method |
| CN117056814B (en) * | 2023-10-11 | 2024-01-05 | 国网山东省电力公司日照供电公司 | Transformer voiceprint vibration fault diagnosis method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110059601B (en) | Intelligent fault diagnosis method for multi-feature extraction and fusion | |
| Chen et al. | Intelligent fault diagnosis method of planetary gearboxes based on convolution neural network and discrete wavelet transform | |
| CN110261109B (en) | Rolling bearing fault diagnosis method based on bidirectional memory cyclic neural network | |
| CN105738109B (en) | Bearing fault classification diagnosis method based on rarefaction representation and integrated study | |
| CN112200244B (en) | Intelligent detection method for anomaly of aerospace engine based on hierarchical countermeasure training | |
| CN111340238A (en) | Fault diagnosis method, device, equipment and storage medium of industrial system | |
| CN108256556A (en) | Wind-driven generator group wheel box method for diagnosing faults based on depth belief network | |
| CN111914883A (en) | Spindle bearing state evaluation method and device based on deep fusion network | |
| CN113076920B (en) | Intelligent fault diagnosis method based on asymmetric domain confrontation self-adaptive model | |
| CN114549925A (en) | Sea wave effective wave height time sequence prediction method based on deep learning | |
| CN114004252A (en) | Bearing fault diagnosis method, device and equipment | |
| CN110991471B (en) | Fault diagnosis method for high-speed train traction system | |
| CN112146882B (en) | Bearing fault diagnosis method based on transfer learning vibration signal image recognition | |
| CN111079861A (en) | Power distribution network voltage abnormity diagnosis method based on image rapid processing technology | |
| CN116399592A (en) | Bearing fault diagnosis method based on channel attention dual-path feature extraction | |
| CN112418175A (en) | Rolling bearing fault diagnosis method and system based on domain migration and storage medium | |
| CN116011507A (en) | Rare fault diagnosis method for fusion element learning and graph neural network | |
| CN115290326A (en) | Rolling bearing fault intelligent diagnosis method | |
| CN113884844A (en) | Transformer partial discharge type identification method and system | |
| CN117009916A (en) | Actuator fault diagnosis method based on multi-sensor information fusion and transfer learning | |
| Li et al. | Intelligent fault diagnosis of aeroengine sensors using improved pattern gradient spectrum entropy | |
| CN115114965A (en) | Wind turbine generator gearbox fault diagnosis model, method, equipment and storage medium | |
| Zhao et al. | Bearing fault diagnosis based on mel frequency cepstrum coefficient and deformable space-frequency attention network | |
| CN113758709A (en) | Rolling bearing fault diagnosis method and system combining edge calculation and deep learning | |
| CN117763494A (en) | Bearing diagnosis method and system based on multi-mode and multi-scale fusion network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |