CN114330096A - Deep learning-based rail transit motor bearing fault degree classification model construction method - Google Patents
Deep learning-based rail transit motor bearing fault degree classification model construction method Download PDFInfo
- Publication number
- CN114330096A CN114330096A CN202111409465.9A CN202111409465A CN114330096A CN 114330096 A CN114330096 A CN 114330096A CN 202111409465 A CN202111409465 A CN 202111409465A CN 114330096 A CN114330096 A CN 114330096A
- Authority
- CN
- China
- Prior art keywords
- fault
- network
- layer
- classification
- training
- 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.)
- Granted
Links
- 238000013135 deep learning Methods 0.000 title claims abstract description 32
- 238000013145 classification model Methods 0.000 title claims abstract description 15
- 238000010276 construction Methods 0.000 title description 6
- 238000000034 method Methods 0.000 claims abstract description 50
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 34
- 238000005070 sampling Methods 0.000 claims abstract description 33
- 230000001133 acceleration Effects 0.000 claims abstract description 29
- 230000015654 memory Effects 0.000 claims abstract description 17
- 238000012549 training Methods 0.000 claims description 55
- 238000012795 verification Methods 0.000 claims description 29
- 238000012360 testing method Methods 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 10
- 238000011176 pooling Methods 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000007787 long-term memory Effects 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 238000000605 extraction Methods 0.000 claims description 7
- 230000005540 biological transmission Effects 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 5
- 230000006403 short-term memory Effects 0.000 claims description 5
- 238000009432 framing Methods 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 2
- 230000004913 activation Effects 0.000 claims description 2
- 230000007774 longterm Effects 0.000 claims description 2
- 230000000737 periodic effect Effects 0.000 claims description 2
- 238000003745 diagnosis Methods 0.000 abstract description 15
- 238000012544 monitoring process Methods 0.000 abstract description 4
- 230000008901 benefit Effects 0.000 abstract description 3
- 238000001514 detection method Methods 0.000 abstract description 3
- 230000036541 health Effects 0.000 abstract description 3
- 238000005457 optimization Methods 0.000 abstract description 2
- 238000012423 maintenance Methods 0.000 description 12
- 238000010586 diagram Methods 0.000 description 8
- 238000004458 analytical method Methods 0.000 description 6
- 230000007547 defect Effects 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 230000003137 locomotive effect Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 238000004901 spalling Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- 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
Abstract
The invention discloses a method for constructing a classification model of fault degrees of a motor bearing of rail transit based on deep learning, and relates to the field of motor bearing fault diagnosis in rail transit. Specifically, a deep learning network is trained by using acceleration sensor signals of a rail transit motor bearing, and then a convolutional neural network or a long-time memory network is adopted in the deep learning network, so that a novel efficient and accurate solution scheme can be provided for rail transit motor bearing health monitoring. An acceleration sensor is installed on the traction motor, and vibration signals of the motor are collected in real time and intelligently analyzed, so that real-time normal state monitoring and fault degree detection of a motor bearing are realized. The invention trains the network model by using a deep learning method, thereby effectively improving the classification accuracy of the fault type and the fault degree; the data sampling is a single-axis acceleration sensor, so that the hardware cost of data acquisition can be effectively reduced, and economic benefits are brought; the model optimization can be carried out by selecting the type and the depth of the network according to the data characteristics.
Description
Technical Field
The invention relates to the field of motor bearing fault diagnosis in rail transit, in particular to a method for constructing a rail transit motor bearing fault degree classification model based on deep learning.
Background
In recent years, with the trend of rail transit development and construction in China, the total mileage and the total number of rail transit vehicles are increasing, the complexity and the integration of train vehicles are increasing, and higher requirements are made on the safety and the high efficiency of maintenance of rail transit equipment. However, the current common train maintenance mode cannot completely keep pace with the development of rail transit, the current maintenance mode not only consumes more human resources, but also causes equipment failure due to untimely maintenance and the like, so that economic loss is caused, and even passenger safety is endangered. The existing maintenance modes of 'planned maintenance' and 'after repair' of rail transit vehicles cannot meet the maintenance requirements of the rail transit vehicles, and the fault Prediction and Health Management (PHM) technology with the 'maintenance according to the situation' capability is receiving increasing attention.
The large and complex size of rail transit vehicles leads to a high incidence of failure. The rail transit vehicle motor is the most critical part of a transmission chain part, and the analysis on the fault and the severity of a motor bearing directly influences whether the vehicle normally runs. Predictive maintenance is one of the key innovation points proposed by industry 4.0. Based on the continuous measurements and analysis, predictive maintenance can predict critical indicators such as the remaining useful life of the machine part. The key operation parameter data can assist decision making, judge the operation state of the machine and optimize the maintenance opportunity of the machine. One notable direction of development is the application of deep learning in predictive maintenance. The method has the advantages that signals of the acceleration sensor of the rail transit motor bearing are monitored, and diagnosis and prediction of faults are achieved through analysis, statistics and classification of the signals. The method is particularly important for processing sensor data and constructing a fault diagnosis model of a deep learning algorithm.
In the prior art, there are two directions for the related art, respectively as follows:
one is to extract features from the time or frequency domain signals and then analyze these features using machine learning algorithms when processing acceleration data for rail transit motor bearings. When the bearing breaks down, the vibration amplitude of the bearing is greatly increased, and meanwhile, a corresponding impact signal is generated. The peak index is generally used to detect impact vibration caused by bearing spalling, cracks, and the like, and the kurtosis index is used for the earliest fault diagnosis of the bearing. The waveform index is often used for detecting mechanical faults of bearing components caused by discrete defects such as pits, nicks, peeling and scratches. In the bearing fault diagnosis process, the fault characteristics of the rolling bearing are usually modulation phenomena, the time domain waveform of the vibration signal is complex, and the fault signal characteristics cannot be expressed visually. The vibration frequency signal has objectivity and can better reflect the basic characteristics of the vibration signal. The time domain vibration signal is converted into the frequency domain vibration signal through Fourier transform, and the frequency spectrum can visually reflect the energy size, the frequency composition and the phase of the vibration signal. However, the method has the following defects: when time domain or frequency domain feature model training is carried out on motor bearing acceleration sensor data, 3-axis signals of the acceleration sensor are often required to extract features. And when the professional knowledge of the bearings of the algorithm engineers is deficient, the fault degrees of the bearings are different, noise interference and the like, the classification effect of the model is poor, the classification precision is low, the hidden dangers of misjudgment and missed judgment exist, the generalization capability of the model is poor, and the signal characteristics under different rotating speed conditions are difficult to respond simultaneously.
The other direction: both the time domain fault diagnosis method and the frequency domain fault diagnosis method have certain limitations, and the comprehensiveness and the locality of the vibration signal cannot be well reflected, so that a new diagnosis method is generated, namely the time-frequency fault diagnosis method. The time-frequency fault diagnosis method organically combines a time domain and a frequency domain into a time-frequency phase plane to obtain vibration signal frequency spectrograms at different moments. The time-frequency fault diagnosis method widely applied at present mainly comprises Hilbert-Huang transform (HHT) and an envelope modulation method. The envelope demodulation method is to utilize envelope detection and spectrum analysis of envelope signals, and then diagnose and identify faults according to demodulated spectrum peaks. The envelope demodulation method is particularly suitable for high-frequency impact vibration, and is still the only and effective important analysis means for high-frequency impact vibration until now. The envelope demodulation method is mainly used for high-precision fault diagnosis of the bearing, and not only can diagnose a fault part, but also can judge the fault type. However, the method has the following defects: the envelope analysis, wavelet transformation and the like used still depend on manual feature extraction and professional knowledge accumulation to a great extent, and higher requirements are put forward for modeling of a fault algorithm model. Moreover, when the fault classification categories are too many and the training data amount is small, the model is easy to generate an overfitting phenomenon, so that the accuracy of a test set is reduced, and the problem that the training model cannot be generalized is caused.
Therefore, based on the drawbacks of the above two methods, a new fault diagnosis method needs to be constructed.
Disclosure of Invention
The invention provides a method for constructing a classification model of the fault degree of a rail transit motor bearing based on deep learning, aiming at solving the defect problem of the fault diagnosis of the rail transit motor bearing in the method.
The invention aims to use signals of an acceleration sensor of a rail transit motor bearing to carry out deep learning network training, and form a method for diagnosing the fault type and the fault severity of the rail transit motor bearing. The deep learning network adopts a convolutional neural network or a long-term memory network, and can provide a new efficient and accurate solution for the health monitoring of the rail transit motor bearing. An acceleration sensor is installed on the traction motor, and vibration signals of the motor are collected in real time and intelligently analyzed, so that real-time normal state monitoring and fault degree detection of a motor bearing are realized. The invention is realized by the following technical scheme:
a rail transit motor bearing fault degree classification model construction method based on deep learning comprises the following steps:
a. the rail transit motor is arranged on the joint debugging test bed, and the motor runs in a no-load mode; acquiring Z-axis direction acceleration signals of a bearing at a transmission end by using a single-axis acceleration sensor, acquiring continuous Z-axis direction acceleration signals within a T time range at the rotating speed of 1000-4500 rpm under each bearing fault type to obtain A original signals, dividing each original signal into B parts, wherein the time interval T of each part is equal to T/B, adding a fault type label to each part, and enabling each original signal to form an original signal data set consisting of time sequence data of B labels with faults; the signal sampling frequency of the single-axis acceleration sensor is Fs, so that each time series data with the fault label comprises N sampling points, and N is T/B multiplied by Fs;
b. the method comprises the steps that N sampling points in each time series data with the fault tags of each original signal data set are subjected to Fast Fourier Transform (FFT), FFT results of the N sampling points are obtained, time domain values of the sampling points of the time series data with the fault tags correspond to a frequency domain value, and the time domain values and the corresponding frequency domain values are combined, so that the time series data with the fault tags are changed from time domain one-dimensional 1 multiplied by N into time domain two-dimensional 2 multiplied by N, before FFT, the time series data with the fault tags belong to one-dimensional time series, signals with only one time domain dimension are difficult to analyze, deep network structures are needed to be compared, and more computing resources are consumed to complete identification requirements. Some signals are difficult to see what features are in the time domain, but if transformed to the frequency domain, the features are easy to see, and FFT has once been considered an epoch-making progress in signal analysis techniques. Changing the time sequence data with the fault label from one dimension of time domain 1 XN to two dimensions of time domain frequency domain 2 XN, and decomposing the original N point sequence into a series of short sequences in sequence; the symmetric property and the periodic property of the exponential factors in the discrete Fourier transform calculation formula are fully utilized, and then discrete Fourier transform values corresponding to the short sequences are calculated and combined, so that the aims of deleting repeated calculation, reducing multiplication and simplifying the structure are fulfilled; in order to realize fast algorithm FFT of discrete Fourier transform, the requirement that the signal sampling frequency Fs of a single-axis acceleration sensor is more than 2 times of the maximum frequency of a signal, namely the maximum rotating speed (rps) of a motor, is met so as to meet the Shannon sampling theorem;
c. randomly distributing data in each original signal data set according to the proportion of a training set, a verification set and a test set being 8:1:1, and finally obtaining A multiplied by n multiplied by B multiplied by 80% as training samples, A multiplied by n multiplied by B multiplied by 10% as verification samples and the rest A multiplied by n multiplied by B multiplied by 10% as test samples;
training a deep learning network model for processing classification tasks, wherein the deep learning network model is a Convolutional Neural Network (CNN) or a long-term and short-term memory network (LSTM); CNNs typically include several key levels, including convolutional layers, ReLU active layers, and pooling layers. Long-and-short memory networks (LSTM) are the most classical structure in Recurrent Neural Networks (RNNs), and are also well suited to classification tasks involving sequence data:
a. the processing procedure of the convolutional neural network CNN network is as follows: as shown in fig. 2, two-dimensional sequence data is upscaled into three-dimensional image data input image input layers; extracting image features by utilizing a convolutional layer, wherein the convolutional layer, the batch normalization layer, the ReLU activation layer and the pooling layer are used as a complete feature extraction process; when needed, the method can be used as a whole for expansion to form a plurality of convolution characteristic extraction processes; after the feature extraction is completed, using a discarding layer to prevent the model from being over-fitted; then using the fully connected layer for combining previously found features; the classification function layer Softmax performs weighting and classification calculation on the output fully-connected features; the classification layer outputs the result of the network training as the type of the fault;
b. the processing process of the long-time memory network LSTM comprises the following steps: as shown in fig. 3, two-dimensional sequence data is input into a sequence input layer; the long-term memory layer learns the long-term dependence between the time step lengths of the sequence data; discarding the layer, and randomly ignoring hidden layer nodes with a certain probability; when needed, the long-time memory layer and the short-time memory layer and the discarding layer can be expanded as a whole to form a multi-layer memory dependence process; some hidden layer nodes can be randomly ignored during each network iteration, so that network overfitting is prevented; the full connection layer is used for combining the features found before; the classification function layer Softmax performs weighting and classification calculation on the output fully-connected features; the classification layer outputs the result of the network training as the type of the fault;
c. inputting the data in the training sample in the step 2 into a CNN or LSTM network for training, and specifying the training parameters of the network; verifying the identification accuracy of the current training network by the data in the verification sample in the step 2, reaching the required accuracy, and storing the CNN or LSTM network model at the moment;
step 4, verifying the classification result of the model:
the verification samples in the step 2 are led into the CNN model or the LSTM model formed in the step 3 for classification, so that a classification result of the fault type and the severity is obtained, and the total accuracy of the classification result is obtained; and if the classification total accuracy rate does not meet the user index requirement, returning to the step 3 to adjust the CNN or LSTM network depth or layer number, and modifying the training parameters until the classification total accuracy rate meets the requirement.
Compared with the prior art, the invention has the following beneficial effects: the invention provides a method for constructing a classification model of the fault degree of a rail transit motor bearing based on deep learning, which comprises the following steps of (1) training a network model by using the deep learning method, so that the classification accuracy of the fault type and the fault degree can be effectively improved; (2) data sampling is a single-axis acceleration sensor, so that the hardware cost of data acquisition can be effectively reduced, and considerable economic benefits are brought. (3) The model optimization can be carried out by selecting the type and the depth of the network according to the data characteristics.
Drawings
Fig. 1 is a diagram of a fault identification process based on deep learning.
Fig. 2 is a network structure diagram of a convolutional neural network.
Fig. 3 shows a network structure of the long and short term memory network.
Fig. 4 is a time domain waveform diagram of each type of fault type in two embodiments.
Fig. 5 is a schematic structural diagram of a convolutional neural network used in the first embodiment.
FIG. 6 shows a training process of convolutional neural network according to an embodiment.
FIG. 7 is a diagram of a validation sample classification confusion matrix according to one embodiment.
Fig. 8 is a schematic diagram of a long-term and short-term memory network used in the second embodiment.
FIG. 9 is a diagram illustrating a training process of the long and short term memory network according to the second embodiment.
FIG. 10 is a diagram of a validation sample classification confusion matrix according to the second embodiment.
Detailed Description
The present invention is further illustrated by the following specific examples.
Example one
A rail transit motor bearing fault degree classification model construction method based on deep learning is disclosed, a design framework is shown in figure 1, and the method comprises the following steps:
the fault type one: inner ring 1.1 mm deep (inner ring fault degree 1 level)
And (2) fault type II: inner ring 1.5 mm deep (inner ring fault degree 2 grade)
And (3) fault type three: inner ring 1.7 mm deep (inner ring fault degree 3 grade)
And (4) fault type four: inner ring 2.0 mm deep (inner ring fault degree 4 grade)
And (5) fault type five: roller 0.6 mm deep (roller fault degree 1 grade)
The fault type six: roller 1.2 mm deep (roller fault degree 2 grade)
The fault type is seven: roller 1.5 mm deep (roller fault degree 3 grade)
The fault type is eight: roller 1.7 mm deep (roller fault degree 4 grade)
a. a motor of a locomotive of a certain model is arranged on a joint debugging test bed, and the motor runs in an idle load mode; acquiring continuous Z-axis direction acceleration signals of each fault type bearing at a motor transmission end within 240 seconds at the motor rotating speed of 1000-4500 rpm to obtain 8 original signals, dividing each original signal into 2400 parts, wherein the time interval t of each part is equal to 0.1 second, and adding a fault type label to each part to enable each original signal to form an original signal data set consisting of 2400 time sequence data with fault labels; the signal sampling frequency of the single-axis acceleration sensor is 25.6 kHz, so that each time series data with the fault label contains 2560 sampling points;
b. 2560 sampling points in each time series data with the fault tag of each original signal data set are subjected to fast algorithm FFT of discrete Fourier transform to obtain FFT results of the 2560 sampling points, so that a time domain value of each sampling point of each time series data with the fault tag corresponds to a frequency domain value, and each time domain value and the corresponding frequency domain value are combined, thereby changing each time series data with the fault tag from a time domain one-dimension (1 × 2560) to a time domain two-dimension (2 × 2560); in order to realize fast algorithm FFT of discrete Fourier transform, the signal sampling frequency Fs of the single-axis acceleration sensor is more than 2 times of the maximum frequency of the signal, namely the maximum rotating speed of the motor;
c. randomly distributing data according to the proportion of 8:1:1 of the training set, the verification set and the test set, wherein 15360 (8 multiplied by 2400 multiplied by 80%) of the data are obtained as training samples, 1920 (8 multiplied by 2400 multiplied by 10%) of the data are obtained as verification samples, and the rest 1920 (8 multiplied by 2400 multiplied by 10%) of the data are obtained as test samples.
training a Convolutional Neural Network (CNN) model, wherein the CNN model mainly comprises 29 layers and has a structure shown in figure 5; and (3) the two-dimensional sequence data of the training sample in the step (2) is upgraded into three-dimensional image data to be input into an image input layer, and training parameters of a specified network are as follows: the characteristic parameter of the image input layer (imageInputLayer) is [2 × 2560 × 1 ]; the convolutional network layer uses 6 convolutional layers (convolutional layers), the sizes of convolutional kernels are all [1 × 8], 6 ReLU active layers (convolutional layers) and 6 batch normalization layers (batch normalization layers), 5 maximum pooling layers (maxPooling layers), the sizes of pooling kernels are all [1 × 2], 1 average pooling layer (maxPooling layer), and the sizes of pooling kernels are all [1 × 32 ]; drop layer (dropout layer) drop probability 0.5; the fully connected layer (fullyconnected layer), the classification function (softmaxLayer) layer and the classification (classificationLayer) layer define the output of the network as 8 fault types; the solver used a "random gradient descent" (SGD). And (3) verifying the identification accuracy and generalization ability of the network by using the verification sample obtained in the step (2) in the training process, wherein the accuracy is over 99 percent, and storing the Convolutional Neural Network (CNN) model at the moment. The accuracy of the verification sample is set as a network convergence condition, as shown in fig. 6, loss functions of the training sample and the verification sample are always reduced during network training, the accuracy is always increased, and the training efficiency is high.
Step 4, verifying the classification result of the model:
and (3) introducing the verification samples in the step (2) into the convolutional neural network in the step (3) for classification, and finally verifying 8 classification results, wherein the total verification accuracy rate exceeds 99.3%, as shown in fig. 7.
Example two
A rail transit motor bearing fault degree classification model construction method based on deep learning comprises the following steps:
the fault type one: the inner ring is 1.1 mm deep (the failure degree of the inner ring is 1 level),
and (2) fault type II: the inner ring is 1.5 mm deep (inner ring failure degree 2 level),
and (3) fault type three: the inner ring is 1.7 mm deep (inner ring failure degree 3 level),
and (4) fault type four: the inner ring is 2.0 mm deep (inner ring failure degree 4 level),
and (5) fault type five: the rollers are 0.6 mm deep (roller failure degree 1 level),
the fault type six: the rollers are 1.2 mm deep (roller failure degree 2 level),
the fault type is seven: the rollers are 1.5 mm deep (roller failure degree 3 grade),
the fault type is eight: the roller is 1.7 mm deep (roller failure degree 4 grade);
a. a motor of a locomotive of a certain model is arranged on a joint debugging test bed, and the motor runs in an idle load mode; acquiring continuous Z-axis direction acceleration signals of each fault type bearing at a motor transmission end within 240 seconds at the motor rotating speed of 1000-4500 rpm to obtain 8 original signals, dividing each original signal into 2400 parts, wherein the time interval t of each part is equal to 0.1 second, and adding a fault type label to each part to enable each original signal to form an original signal data set consisting of 2400 time sequence data with fault labels; the signal sampling frequency of the single-axis acceleration sensor is 25.6 kHz, so that each time series data with the fault label contains 2560 sampling points;
b. 2560 sampling points in each time series data with the fault tag of each original signal data set are subjected to fast algorithm FFT of discrete Fourier transform to obtain FFT results of the 2560 sampling points, so that a time domain value of each sampling point of each time series data with the fault tag corresponds to a frequency domain value, and each time domain value and the corresponding frequency domain value are combined, thereby changing each time series data with the fault tag from a time domain one-dimension (1 × 2560) to a time domain two-dimension (2 × 2560); in order to realize fast algorithm FFT of discrete Fourier transform, the signal sampling frequency Fs of the single-axis acceleration sensor is more than 2 times of the maximum frequency of the signal, namely the maximum rotating speed of the motor;
c. randomly distributing data according to the proportion of 8:1:1 of the training set, the verification set and the test set, wherein 15360 (8 multiplied by 2400 multiplied by 80%) of the data are obtained as training samples, 1920 (8 multiplied by 2400 multiplied by 10%) of the data are obtained as verification samples, and the rest 1920 (8 multiplied by 2400 multiplied by 10%) of the data are obtained as test samples.
training a long-time memory network LSTM model, wherein the LSTM mainly comprises 8 layers and the structure is shown in figure 8; inputting the two-dimensional sequence data of the training sample in the step 2 into a sequence input layer, and specifying the training parameters of the network as follows: the characteristic parameter of a sequence input layer (sequence input layer) is 2; the first long-short term memory layer (lstmLayer) contains 10 hidden layer units; the second long-short term memory layer (lstmLayer) contains 100 hidden layer units; two drop layers (dropout layer) drop probability 0.2; the fully connected layer (fullyconnected layer), the classification function (softmaxLayer) layer and the classification (classificationLayer) layer define the output of the network as 8 fault types; the solver uses a "moment estimate solution" (adam). And (3) verifying the identification accuracy and generalization ability of the network by using the verification sample obtained in the step (2) in the training process, wherein the accuracy is over 99 percent, and storing the LSTM model of the network at the moment. The accuracy of the verification sample is set as a network convergence condition, as shown in fig. 9, loss functions of the training sample and the verification sample are always reduced during network training, the accuracy is always increased, and the training efficiency is high.
Step 4, verifying the classification result of the model:
and (3) introducing the verification samples in the step (2) into the long-time memory network in the step (3) for classification, and finally verifying 8 classification results, wherein the total verification accuracy rate exceeds 98.3%, as shown in fig. 10.
The scope of the invention is not limited to the above two embodiments, and various modifications and changes may be made by those skilled in the art, and any modifications, improvements and equivalents within the spirit and principle of the invention should be included in the scope of the invention.
Claims (6)
1. A rail transit motor bearing fault degree classification model building method based on deep learning is characterized by comprising the following steps: the method comprises the following steps:
step 1, presetting bearing fault types and quantity: the preset number of types of bearing faults and severity is A;
step 2, original signal acquisition and framing treatment:
a. the rail transit motor is arranged on the joint debugging test bed, and the motor runs in a no-load mode; acquiring Z-axis direction acceleration signals of a bearing at a transmission end by using a single-axis acceleration sensor, acquiring continuous Z-axis direction acceleration signals within a T time range at the rotating speed of 1000-4500 rpm under each bearing fault type to obtain A original signals, dividing each original signal into B parts, wherein the time interval T of each part is equal to T/B, adding a fault type label to each part, and enabling each original signal to form an original signal data set consisting of time sequence data of B labels with faults; the signal sampling frequency of the single-axis acceleration sensor is Fs, so that each time series data with the fault label comprises N sampling points, and N is T/B multiplied by Fs;
b. obtaining an FFT result of the N sampling points through a fast algorithm FFT of a Discrete Fourier Transform (DFT) in each time series data with the fault label of each original signal data set, enabling a time domain value of each sampling point of each time series data with the fault label to correspond to a frequency domain value, and combining each time domain value and the corresponding frequency domain value, so that each time series data with the fault label is changed from a time domain one-dimension 1 multiplied by N to a time domain two-dimension 2 multiplied by N, namely, an original N point sequence is sequentially decomposed into a series of short sequences; the symmetric property and the periodic property of the exponential factors in the discrete Fourier transform calculation formula are fully utilized, and then discrete Fourier transform values corresponding to the short sequences are calculated and combined, so that the aims of deleting repeated calculation, reducing multiplication and simplifying the structure are fulfilled; in order to realize fast algorithm FFT of discrete Fourier transform, the requirement that the signal sampling frequency Fs of a single-axis acceleration sensor is more than 2 times of the maximum frequency of a signal, namely the maximum rotating speed (rps) of a motor, is met so as to meet the Shannon sampling theorem;
c. randomly distributing data in each original signal data set according to the proportion of a training set, a verification set and a test set being 8:1:1, and finally obtaining A multiplied by n multiplied by B multiplied by 80% as training samples, A multiplied by n multiplied by B multiplied by 10% as verification samples and the rest A multiplied by n multiplied by B multiplied by 10% as test samples;
step 3, establishing a deep learning network:
training a deep learning network model for processing classification tasks, wherein the deep learning network model is a Convolutional Neural Network (CNN) or a long-term and short-term memory network (LSTM);
a. the processing procedure of the convolutional neural network CNN network is as follows: the two-dimensional sequence data is converted into three-dimensional image data in an up-to-down mode and input into an image input layer; extracting image features by utilizing a convolutional layer, wherein the convolutional layer, the batch normalization layer, the ReLU activation layer and the pooling layer are used as a complete feature extraction process; when needed, the method can be used as a whole for expansion to form a plurality of convolution characteristic extraction processes; after the feature extraction is completed, using a discarding layer to prevent the model from being over-fitted; then using the fully connected layer for combining previously found features; the classification function layer Softmax performs weighting and classification calculation on the output fully-connected features; the classification layer outputs the result of the network training as the type of the fault;
b. the processing process of the long-time memory network LSTM comprises the following steps: inputting two-dimensional sequence data into a sequence input layer; the long-term memory layer learns the long-term dependence between the time step lengths of the sequence data; discarding the layer, and randomly ignoring hidden layer nodes with a certain probability; when needed, the long-time memory layer and the short-time memory layer and the discarding layer can be expanded as a whole to form a multi-layer memory dependence process; some hidden layer nodes can be randomly ignored during each network iteration, so that network overfitting is prevented; the full connection layer is used for combining the features found before; the classification function layer Softmax performs weighting and classification calculation on the output fully-connected features; the classification layer outputs the result of the network training as the type of the fault;
c. inputting the data in the training sample in the step 2 into a CNN or LSTM network for training, and specifying the training parameters of the network; verifying the identification accuracy of the current training network by the data in the verification sample in the step 2, reaching the required accuracy, and storing the CNN or LSTM network model at the moment;
step 4, verifying the classification result of the model:
the verification samples in the step 2 are led into the CNN model or the LSTM model formed in the step 3 for classification, so that a classification result of the fault type and the severity is obtained, and the total accuracy of the classification result is obtained; and if the classification total accuracy rate does not meet the user index requirement, returning to the step 3 to adjust the CNN or LSTM network depth or layer number, and modifying the training parameters until the classification total accuracy rate meets the requirement.
2. The rail transit motor bearing fault degree classification model building method based on deep learning of claim 1 is characterized in that: in the step 1, 8 types and numbers of bearing faults are preset, which are respectively as follows:
the fault type one: the inner ring is 1.1 mm deep, and the fault degree of the inner ring is 1 grade;
and (2) fault type II: the inner ring is 1.5 mm deep, and the fault degree of the inner ring is 2 levels;
and (3) fault type three: the inner ring is 1.7 mm deep, and the fault degree of the inner ring is 3 grades;
and (4) fault type four: the depth of the inner ring is 2.0 mm, and the fault degree of the inner ring is 4 grades;
and (5) fault type five: the depth of the roller is 0.6 mm, and the fault degree of the roller is 1 grade;
the fault type six: the depth of the roller is 1.2 mm, and the fault degree of the roller is 2 grades;
the fault type is seven: the depth of the roller is 1.5 mm, and the fault degree of the roller is 3 grades;
the fault type is eight: the roller is 1.7 mm deep, and the roller fault degree is 4 grades.
3. The rail transit motor bearing fault degree classification model building method based on deep learning of claim 1 is characterized in that: in the step 2:
a. acquiring continuous Z-axis direction acceleration signals of each fault type bearing at a motor transmission end within 240 seconds at the motor rotating speed of 1000-4500 rpm to obtain 8 original signals, dividing each original signal into 2400 parts, wherein the time interval t of each part is equal to 0.1 second, and adding a fault type label to each part to enable each original signal to form an original signal data set consisting of 2400 time sequence data with fault labels; the signal sampling frequency of the single-axis acceleration sensor is 25.6 kHz, so that each time series data with the fault label contains 2560 sampling points;
b. 2560 sampling points in each time series data with the fault tag of each original signal data set are subjected to fast algorithm FFT of discrete Fourier transform to obtain FFT results of the 2560 sampling points, so that a time domain value of each sampling point of each time series data with the fault tag corresponds to a frequency domain value, and each time domain value and the corresponding frequency domain value are combined, thereby changing each time series data with the fault tag from a time domain one-dimension of 1 × 2560 to a time domain two-dimension of 2 × 2560; in order to realize fast algorithm FFT of discrete Fourier transform, the signal sampling frequency Fs of the single-axis acceleration sensor is more than 2 times of the maximum frequency of the signal, namely the maximum rotating speed of the motor;
c. randomly distributing data according to the proportion of 8:1:1 of a training set, a verification set and a test set, wherein 15360=8 × 2400 × 80% of the data are obtained as training samples, 1920=8 × 2400 × 10% of the data are obtained as verification samples, and the rest 1920=8 × 2400 × 10% of the data are obtained as test samples.
4. The rail transit motor bearing fault degree classification model building method based on deep learning of claim 1 is characterized in that: in the step 3:
training a Convolutional Neural Network (CNN) model, wherein the CNN model consists of 29 layers, the two-dimensional sequence data of the training sample in the step 2 are upgraded into three-dimensional image data which is input into an image input layer, and the training parameters of the network are specified as follows: the image input layer imageInputLayer characteristic parameter is [2 × 2560 × 1 ]; the convolutional network layer uses 6 convolutional layers, the sizes of convolutional kernels are all [1 × 8], 6 ReLU active layers reluLayers and 6 batch normalization layers batchNormalizationLayers, 5 maximum value pooling layers maxPolingLayers are all [1 × 2], 1 average value pooling layer maxPolingLayer is provided, and the sizes of pooling kernels are all [1 × 32 ]; the drop probability of the drop layer is 0.5; the fully connected layer fullyconnected layer, the classification function softmaxLayer and the classification classifying layer define the output of the network as 8 fault types; the solver uses a "random gradient descent" SGD; verifying the identification accuracy and generalization ability of the network by using the verification sample in the step 2 in the training process, wherein the accuracy reaches over 99 percent, and storing the CNN model of the convolutional neural network at the moment; and setting the accuracy of the verification sample as a network convergence condition.
5. The rail transit motor bearing fault degree classification model building method based on deep learning of claim 1 is characterized in that: in the step 3:
training a long-time memory network LSTM model, wherein the LSTM mainly comprises 8 layers, inputting the two-dimensional sequence data of the training sample in the step 2 into a sequence input layer, and specifying the training parameters of the network as follows: the sequence input layer characteristic parameter is 2; the first long-short term memory lstmLayer comprises 10 hidden layer units; the second lstmLayer contains 100 hidden layer units; the two drop layers dropoutLayer drop probability is 0.2; the fully connected layer fullyconnected layer, the classification function softmaxLayer and the classification classifying layer define the output of the network as 8 fault types; solving adam by using a solver through moment estimation; verifying the identification accuracy and generalization ability of the network by using the verification sample in the step 2 in the training process, wherein the accuracy reaches over 99 percent, and storing the LSTM model of the network at the moment; and setting the accuracy of the verification sample as a network convergence condition.
6. The rail transit motor bearing fault degree classification model building method based on deep learning of claim 1 is characterized in that: in the step 4: and (3) introducing the verification sample in the step (2) into the convolutional neural network or the long-short time memory network in the step (3) for classification, and finally verifying 8 classification results to verify the overall accuracy.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111409465.9A CN114330096B (en) | 2021-11-25 | 2021-11-25 | Rail transit motor bearing fault degree classification model construction method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111409465.9A CN114330096B (en) | 2021-11-25 | 2021-11-25 | Rail transit motor bearing fault degree classification model construction method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114330096A true CN114330096A (en) | 2022-04-12 |
| CN114330096B CN114330096B (en) | 2024-04-19 |
Family
ID=81047517
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111409465.9A Active CN114330096B (en) | 2021-11-25 | 2021-11-25 | Rail transit motor bearing fault degree classification model construction method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114330096B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115115037A (en) * | 2022-08-30 | 2022-09-27 | 无锡沃尔德轴承有限公司 | Bearing fault probability analysis device |
| CN115758258A (en) * | 2022-11-22 | 2023-03-07 | 西南交通大学 | Traction transformer fault diagnosis method based on vibration signals and domain knowledge |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200200648A1 (en) * | 2018-02-12 | 2020-06-25 | Dalian University Of Technology | Method for Fault Diagnosis of an Aero-engine Rolling Bearing Based on Random Forest of Power Spectrum Entropy |
| CN111538947A (en) * | 2020-05-18 | 2020-08-14 | 中车永济电机有限公司 | Method for constructing wind power generator bearing fault classification model |
-
2021
- 2021-11-25 CN CN202111409465.9A patent/CN114330096B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200200648A1 (en) * | 2018-02-12 | 2020-06-25 | Dalian University Of Technology | Method for Fault Diagnosis of an Aero-engine Rolling Bearing Based on Random Forest of Power Spectrum Entropy |
| CN111538947A (en) * | 2020-05-18 | 2020-08-14 | 中车永济电机有限公司 | Method for constructing wind power generator bearing fault classification model |
Non-Patent Citations (2)
| Title |
|---|
| 王太勇;王廷虎;王鹏;乔卉卉;徐明达;: "基于注意力机制BiLSTM的设备智能故障诊断方法", 天津大学学报(自然科学与工程技术版), no. 06, 27 April 2020 (2020-04-27) * |
| 许爱华;杜洋;袁涛;: "基于深度学习的电机轴承故障诊断研究", 组合机床与自动化加工技术, no. 03, 20 March 2020 (2020-03-20) * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115115037A (en) * | 2022-08-30 | 2022-09-27 | 无锡沃尔德轴承有限公司 | Bearing fault probability analysis device |
| CN115758258A (en) * | 2022-11-22 | 2023-03-07 | 西南交通大学 | Traction transformer fault diagnosis method based on vibration signals and domain knowledge |
| CN115758258B (en) * | 2022-11-22 | 2023-12-12 | 西南交通大学 | Traction transformer fault diagnosis method based on vibration signals and domain knowledge |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114330096B (en) | 2024-04-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111709448B (en) | Mechanical fault diagnosis method based on migration relation network | |
| Shi et al. | Planetary gearbox fault diagnosis using bidirectional-convolutional LSTM networks | |
| Islam et al. | Automated bearing fault diagnosis scheme using 2D representation of wavelet packet transform and deep convolutional neural network | |
| Zou et al. | Bearing fault diagnosis method based on EEMD and LSTM | |
| CN113469060A (en) | Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method | |
| Purushotham et al. | Multi-fault diagnosis of rolling bearing elements using wavelet analysis and hidden Markov model based fault recognition | |
| CN114330096B (en) | Rail transit motor bearing fault degree classification model construction method based on deep learning | |
| CN107817106A (en) | Fault Diagnosis of Roller Bearings based on Bayes's residual transform singular value decomposition Gaussian Mixture HMM framework | |
| CN110110768B (en) | Rolling bearing fault diagnosis method based on parallel feature learning and multiple classifiers | |
| CN112541233A (en) | Rotary machine fault diagnosis method based on improved convolutional neural network | |
| Li et al. | Joint attention feature transfer network for gearbox fault diagnosis with imbalanced data | |
| CN116358871B (en) | Rolling bearing weak signal composite fault diagnosis method based on graph rolling network | |
| Zhou et al. | Remaining useful life prediction by distribution contact ratio health indicator and consolidated memory GRU | |
| CN116625683A (en) | Wind turbine generator system bearing fault identification method, system and device and electronic equipment | |
| Lu et al. | Early fault warning and identification in condition monitoring of bearing via wavelet packet decomposition coupled with graph | |
| CN112729825A (en) | Method for constructing bearing fault diagnosis model based on convolution cyclic neural network | |
| CN110044619B (en) | Multi-fault feature identification method based on sparse multi-cycle group lasso | |
| Zhang et al. | Fault Diagnosis with Bidirectional Guided Convolutional Neural Networks Under Noisy Labels | |
| CN117093938A (en) | Fan bearing fault detection method and system based on deep learning | |
| CN114861349A (en) | Rolling bearing RUL prediction method based on model migration and wiener process | |
| CN116401950A (en) | Rolling bearing performance degradation prediction method based on time chart convolution neural network | |
| CN113111752B (en) | Rolling mill fault diagnosis method for sample imbalance enhanced extended depth confidence network | |
| CN115062867A (en) | Bearing fault prediction method based on Teager energy spectrum and improved CNN model | |
| CN115659778A (en) | On-line construction method and on-line updating equipment of wind turbine generator bearing fault diagnosis model | |
| CN116910574A (en) | Mechanical equipment diagnosis method and device based on meta-learning and time convolution 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |