CN114707534A - Rolling bearing fault diagnosis method under small sample data - Google Patents

Rolling bearing fault diagnosis method under small sample data Download PDF

Info

Publication number
CN114707534A
CN114707534A CN202210217466.1A CN202210217466A CN114707534A CN 114707534 A CN114707534 A CN 114707534A CN 202210217466 A CN202210217466 A CN 202210217466A CN 114707534 A CN114707534 A CN 114707534A
Authority
CN
China
Prior art keywords
sample
data
fault
training
rolling bearing
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
Application number
CN202210217466.1A
Other languages
Chinese (zh)
Inventor
向玲
王凯伦
胡雅楠
胡爱军
杨鑫
刘冰
张兰昕
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
North China Electric Power University
Original Assignee
North China Electric Power University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by North China Electric Power University filed Critical North China Electric Power University
Priority to CN202210217466.1A priority Critical patent/CN114707534A/en
Publication of CN114707534A publication Critical patent/CN114707534A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Software Systems (AREA)
  • Evolutionary Biology (AREA)
  • Biomedical Technology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Signal Processing (AREA)
  • Acoustics & Sound (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

A fault diagnosis method for a rolling bearing under small sample data comprises the following steps: a. collecting original vibration signal data samples of a rolling bearing with faults to construct a task set T, and classifying the data samples in the task set T; b. performing wavelet transformation on the data sample, converting the data sample into a two-dimensional wavelet image sample with fault characteristics, and then dividing the wavelet image sample into a training set, a verification set and a test set; c. completing the expansion of the training set; d. inputting wavelet image samples in a training set into a modified MobileNetv3 convolution network to train the convolution network; e. testing the convolution network; f. and inputting the vibration signal of the rolling bearing to be diagnosed into a convolution network qualified in the test to finish bearing fault diagnosis. The method can not only extract the fault characteristics in the original vibration signal, but also improve the generalization capability of the diagnosis model, thereby accurately diagnosing the fault of the small sample data bearing and ensuring the safe operation of mechanical equipment.

Description

Rolling bearing fault diagnosis method under small sample data
Technical Field
The invention relates to a method for diagnosing faults of a rolling bearing, and belongs to the technical field of data processing.
Background
The rolling bearing is used as an important part of rotary equipment and plays an important role in the safe operation of the equipment, and any type of bearing fault can cause potential fault hidden danger and unexpected safety problem of the equipment, so that advanced fault diagnosis technology is required to be adopted to carry out state evaluation and quality monitoring on the rolling bearing so as to reduce economic loss to the maximum extent. However, the actually acquired fault signals are quite complex, wherein the fault signals comprise fundamental frequency vibration, harmonic vibration and various noise signals, and effective features are difficult to extract from the signals for fault diagnosis only depending on experience; and because it is very difficult to obtain sufficient fault data, the accuracy of the diagnosis result is difficult to ensure by the existing fault diagnosis method based on the neural network. Therefore, how to effectively extract the fault characteristics and complete the fault diagnosis of the rolling bearing under the small sample data is very important.
Disclosure of Invention
The invention aims to provide a fault diagnosis method for a rolling bearing under small sample data aiming at the defects of the prior art so as to accurately diagnose the fault of the bearing and ensure the safe operation of mechanical equipment.
The problems of the invention are solved by the following technical scheme:
a fault diagnosis method for a rolling bearing under small sample data is characterized in that a bearing fault signal is collected by a bearing fault experiment table, the bearing fault experiment table is composed of an acceleration sensor, a motor, a test bearing and a loading module, and collected data are processed according to the following steps:
a. collecting original vibration signal data samples of a rolling bearing with faults, constructing a task set T, and classifying the data samples in the task set T according to different working conditions and different fault types;
b. performing wavelet transformation on data samples in a task Set T by adopting a Morlet wavelet basis function, converting each data sample into a two-dimensional wavelet image sample with fault characteristics, and dividing the wavelet image samples in the task Set T into a training Set (Support Set), a verification Set (Validation Set) and a Test Set (Test Set);
c. inputting each wavelet image sample in the training set into an SSGAN (semi-supervised learning generation countermeasure network) model for training to obtain a corresponding generated sample, and adding the generated sample as an auxiliary training sample into the training set to complete the expansion of the training set;
d. inputting wavelet image samples in the expanded training set into an improved MobileNetv3 convolutional network to train the convolutional network, and adjusting parameters of the convolutional network by using a verification set in the training process;
e. inputting the wavelet image samples in the test set into a trained convolution network, and testing the convolution network;
f. and c, performing wavelet transformation on the vibration signal of the rolling bearing to be diagnosed by adopting the method in the step b, and inputting the vibration signal into a MobileNetv3 convolution network qualified in the test to finish the fault diagnosis of the small sample bearing.
According to the method for diagnosing the fault of the rolling bearing under the small sample data, when wavelet transformation is carried out on data in a task set, cmor wavelets are selected as basic functions of the wavelet transformation, and the expression is as follows:
Figure BDA0003535601240000021
its corresponding fourier representation is:
Figure BDA0003535601240000022
wherein t represents time, f represents frequency, fbIs the shape parameter, a is the transformation scale, fcIs the center frequency.
According to the method for diagnosing the fault of the rolling bearing under the small sample data, the generated sample is screened before being used as an auxiliary training sample to be added into a training set, and the specific screening steps are as follows:
a. the structural similarity SSIM (x, y) of each generated sample to the wavelet image samples of the respective original vibration signals in the training set is calculated using the following formula:
SSIM(x，y)＝[l(x，y)]α*[c(x，y)]β*[s(x，y)]γ
wherein:
Figure BDA0003535601240000023
Figure BDA0003535601240000024
Figure BDA0003535601240000025
where x denotes the pixel value in the original vibration signal wavelet image sample, y denotes the pixel value in the generated sample, μxIs the mean value of x, μyIs the average value of the values of y,
Figure BDA0003535601240000026
is the variance of x and is the sum of the differences,
Figure BDA0003535601240000027
variance of y, σxyIs the covariance of x and y, C1，C2，C3Three constants, alpha, beta and gamma are constants greater than 0, and in engineering application, alpha is set asβ＝γ＝1；
b. Calculating the average value of the structural similarity SSIM (x, y) of each generated sample and the wavelet image sample of each original vibration signal in the training set
Figure BDA0003535601240000031
c. Will generate a sample according to
Figure BDA0003535601240000032
Sorting the magnitudes of the values of (a);
d. screening according to a set proportion
Figure BDA0003535601240000033
And generating a sample with a larger value and adding the sample into the training set to complete the expansion of the training set.
According to the fault diagnosis method for the rolling bearing under the small sample data, the improved MobileNetv3 convolutional network is characterized in that an original lightweight attention mechanism is replaced by a self-attention mechanism on the basis of an original MobileNetv3 convolutional network model; and simultaneously, an Early stop mechanism is added, and within a certain training turn, if the loss of the verification set is unchanged, the training of the convolution network model is stopped.
F, the fault diagnosis method of the rolling bearing under the small sample databIs set to a value of 3, fcThe value of (d) is set to 3.
In the rolling bearing fault diagnosis method under the small sample data, the original vibration signal data samples of the rolling bearing with the fault in the task set T form a data sample matrix
Figure BDA0003535601240000034
Wherein x ismDenotes the mth sample, M denotes the number of samples,
Figure BDA0003535601240000035
the ith data representing the mth sample, i ═ 1,2,. n, represents the length of the sample.
The method adopts the method of combining the wavelet transformation and the structure similarity generation countermeasure network with the improved MobileNetv3 convolutional neural network to diagnose the bearing fault, not only can extract the fault characteristics in the original vibration signal, but also can improve the generalization capability of a diagnosis model, thereby accurately diagnosing the bearing fault of small sample data and ensuring the safe operation of mechanical equipment.
Drawings
The present invention will be described in further detail with reference to the accompanying drawings.
FIG. 1 is a schematic diagram of the experimental setup;
FIG. 2 is an overall flow diagram of the present invention;
FIG. 3 is a Loss variation curve of a discriminant model and a generative model;
FIG. 4 is a bar graph of experimental data classification accuracy;
FIG. 5 is a bar graph of the MobileNetv3 network versus data classification accuracy before and after refinement;
FIG. 6 is a two-dimensional visualization of classification results;
fig. 7 is a rolling bearing failure data set sample condition diagram.
The symbols in the text are: t represents a task set, X represents a data sample matrix, M represents the number of samples, XmDenotes the mth sample, n denotes the length of the sample,
Figure BDA0003535601240000036
i-th data representing the m-th sample, t representing time, f representing frequency, fbIs the shape parameter, a is the transformation scale, fcIs the center frequency, x represents the pixel value in the original vibration signal wavelet image sample, y represents the pixel value in the generated sample, μxIs the mean value of x, μyIs the average value of the y and is,
Figure BDA0003535601240000041
is the variance of x and is the sum of the differences,
Figure BDA0003535601240000042
variance of y, σxyIs the covariance of x and y, C1，C2，C3With three constants, alpha, beta, gamma being greater than 0A constant.
Detailed Description
The invention provides a rolling bearing fault diagnosis method under small sample data aiming at the defects of the prior art and combining a traditional fault diagnosis method and a deep network model so as to accurately diagnose the rolling bearing fault and ensure the safe operation of mechanical equipment.
The invention comprises the following steps:
a. classifying original vibration signal data samples of the rolling bearing with faults, which are acquired by a data acquisition system, according to different working conditions and different fault types to construct a task set T, and obtaining a data sample matrix consisting of the original vibration signal data samples
Figure BDA0003535601240000043
Wherein xmDenotes the mth sample, M denotes the number of samples,
Figure BDA0003535601240000044
i-th data representing an m-th sample, n representing a length of the sample;
b. performing wavelet transformation on data samples in the task Set T by using a Morlet wavelet basis function to obtain two-dimensional wavelet image samples with fault characteristics, and dividing a training Set (Support Set), a verification Set (Validation Set) and a Test Set (Test Set);
c. inputting each wavelet image sample in the training set into an SSGAN (semi-supervised learning generation countermeasure network) model for training to obtain a high-quality auxiliary training sample and complete the expansion of the training set;
d. inputting the expanded training set into an improved MobileNetv3 convolution network for model training, and adjusting network parameters through a verification set;
e. inputting the wavelet image samples in the test set into a trained convolution network, and testing the convolution network;
f. and c, performing wavelet transformation on the vibration signal of the rolling bearing to be diagnosed by adopting the method in the step b, and inputting the vibration signal into a MobileNetv3 convolution network qualified in the test to finish the fault diagnosis of the small sample bearing.
The selection of wavelet basis function is the key to influence the wavelet transformation effect, and from the waveforms of several commonly used wavelet basis functions, the waveform of Morlet wavelet is a cosine signal with bilateral exponential attenuation, which is very similar to the fault pulse characteristics generated by a rolling bearing, and the cmor wavelet in Morlet wavelet represents the negative number form, and has strong self-adaptability, so the invention selects the cmor wavelet as the basis function of wavelet transformation, and the expression is as follows:
Figure BDA0003535601240000045
its corresponding fourier representation is:
Figure BDA0003535601240000046
in the formula (1), t represents time, f represents frequency, and fbIs a shape parameter and determines the speed of waveform oscillation attenuation.
In the formula (2), a is a transformation scale, fcThe center frequency determines the oscillation frequency of the waveform. Since the components in the signal are different, f can be adjustedbAnd fcThe value of (c) changes the time-frequency resolution of the wavelet, where f is setb＝3，f c3 to satisfy the currently analyzed signal.
The invention combines the structural similarity analysis with the deep convolution generation countermeasure network, builds the SSGAN, and completes the specific process of the small sample training set expansion as follows:
a. sampling N types of samples from a task set, randomly extracting 100 samples from 300 samples of each type of sample to form a training set, randomly extracting 100 samples from 200 samples of the rest of each type of sample to form a verification set, and forming a test set by the rest 100 samples;
b. inputting 100 small sample training sets into a generated confrontation network model to obtain a generated sample;
c. from the brightness, contrast and structure three, by Structural Similarity analysis (Structural Similarity)Screening the generated samples, performing structural similarity analysis on the generated samples and each real sample in a training set, and taking the average value of structural similarity SSIM (x, y)
Figure BDA0003535601240000051
The structural similarity is a measure of the similarity between images, the similarity index is between 0 and 1, the larger the index is, the higher the similarity is, and the structural similarity is calculated by the following formula:
SSIM(x，y)＝[l(x，y)]α*[c(x，y)]β*[s(x，y)]γ (3)
brightness (Luminance), Contrast (Contrast) and texture (Structure) are three important modules in the measurement system, where equation (5) is the Luminance Contrast function, equation (6) is the Contrast function, and equation (7) is the texture Contrast function.
Figure BDA0003535601240000052
Figure BDA0003535601240000053
Figure BDA0003535601240000054
Where x represents the pixel value in the original vibration signal wavelet image sample, y represents the pixel value in the generated sample, μxIs the mean value of x, μyIs the average value of the values of y,
Figure BDA0003535601240000055
is the variance of x and is the sum of the differences,
Figure BDA0003535601240000056
variance of y, σxyIs the covariance of x and y, C1＝(k1L)2，C2＝(k2L)2，
Figure BDA0003535601240000057
Is three constants, default k1＝0.01,k2＝0.03,L＝2B-1(B is the bit depth), α, β, γ being constants larger than 0. In actual engineering calculations, α ═ β ═ γ ═ 1 is generally set.
d. Calculating the average value of the structural similarity SSIM (x, y) of each generated sample and the wavelet image sample of each original vibration signal in the training set
Figure BDA0003535601240000061
e. Will generate a sample according to
Figure BDA0003535601240000062
Sorting the magnitudes of the values of (c);
f. screening according to a set proportion
Figure BDA0003535601240000063
And generating a sample with a larger value and adding the sample into the training set to complete the expansion of the training set.
The improved MobileNetv3 network is optimized and improved on the basis of an original MobileNetv3 network model, and a self-attention mechanism is adopted to replace an original lightweight attention mechanism, so that the accuracy of the model is improved greatly; an Early stop mechanism is added, and within a certain training turn, if the loss of the verification set is unchanged, the model training is stopped and the recognition result is output, so that the training time is greatly shortened.
The generative countermeasure network is mainly composed of a generative model (Generator model) and a discriminant model (Discriminator model). The generation model learns the real image distribution by using random noise Z with known distribution so as to ensure that the image G (Z) generated by the generation model is more real; the discrimination model distinguishes the truth of the sample in the obtained data set, and the training process is equivalent to the game between the two models. Over time, the generative and discriminative models are continually competing, eventually reaching nash equilibrium in alternating training: the generated image of the model is close to the real image distribution, and the discriminator cannot distinguish the true and false of the sample. The loss function is as follows:
Figure BDA0003535601240000064
where E (-) represents the mathematical expectation, X is the pixel value of the wavelet image, Pdata(X) true sample data distribution, D (X) represents discrimination result of discrimination model, Z is noise data, PZ(Z) represents the noise data distribution, and G (Z) is the data generated by the generation model.
The method adopts a method of combining a wavelet-transformed 2D diagram and a structure similarity generation countermeasure network with an improved MobileNetv3 convolutional neural network to diagnose the bearing fault, can extract fault characteristics in an original vibration signal, fully exerts strong time-frequency characteristic extraction capability of WT, can improve the quality of a generated image, and improves the generalization capability of a MobileNetv3 diagnosis model, so that the bearing fault of small sample data can be accurately diagnosed, and the safe operation of mechanical equipment is ensured.
The invention has the following advantages:
a. the invention adopts Morlet wavelet to extract two-dimensional time-frequency image feature from one-dimensional original signal, which gives full play to the powerful time-frequency feature extraction capability of WT;
b. the invention combines SSGAN with structural similarity, and gives full play to the advantages of SSGAN in image generation; the structural similarity eliminates the generated samples with larger difference with the real samples, thereby greatly improving the quality of the auxiliary training samples;
c. according to the invention, the Mobile Net v3 convolutional network is improved, an Early-stop mechanism is introduced, and the training time of the network is greatly shortened; the self-attention mechanism is used for replacing a lightweight attention mechanism, the fault diagnosis precision of the small-sample bearing reaches 100%, and the classification accuracy of the diagnosis model is effectively improved.
The effectiveness of the invention is verified by analyzing the fault signal of the rolling bearing, and the flow of the method is shown in fig. 2.
The bearing fault signal acquisition device consists of an acceleration sensor, a motor, a test bearing and a loading module. The vibration signal was acquired using an accelerometer with a sampling frequency of 12.8 kHz. The experimental bearing model is 6205 deep groove ball bearing, 7 types of fault data including normal signals, inner ring faults, outer ring faults and rolling body faults are generated under the conditions of 1425rpm (working condition 1) and 1470rpm (working condition 2) by using an electric spark machining technology, 300 samples are collected for each type of fault, each sample comprises 2000 sampling points, the number of overlapped points between adjacent samples is 200, and the size of a single wavelet image is 224 multiplied by 224. According to the following steps of 1: 1: 1, dividing the data into a training set, a verification set and a test set, wherein the specific data description is shown in table 1.
Table 1 experimental bearing data
Figure BDA0003535601240000071
In the face of the problem of less fault data, the small sample image training set is input into the SSGAN model for countertraining, and the generated high-quality wavelet image sample can enrich the training set and improve the robustness and diagnosis accuracy of the classification model. The auxiliary sample generation is shown in fig. 7.
The loss curves of the discriminant model and the generated model in the training process are shown in fig. 3, and during the previous 500 times of training, the loss values of the two models are relatively large and vibrate repeatedly. When the iteration number reaches 500, the loss values of the two models start to fluctuate slightly and tend to be stable, which indicates that the network gradually reaches Nash balance.
To further verify the effect of the number of training samples on the improved MobileNetv3 network, adjusting the number of real samples and the number of added helper samples, experiments were designed as shown in table 2. And mixing different numbers of auxiliary samples and original samples, inputting the mixed samples into a classification network for training, and applying a test set to the trained network.
It can be seen from table 2 that when the number of the real samples is 100, the classification accuracy is improved from 53.95% to 100% with the addition of the auxiliary sample, and the accuracy is improved by 46.05% compared with that before the auxiliary sample is not added. When the number of each type of real samples is 80, 60, 40, 20 and 0, the fault diagnosis accuracy can be improved to different degrees by adding auxiliary samples. The classification results are shown in fig. 4. As can be seen from the figure, increasing the number of the auxiliary training samples has a significant effect on improving the accuracy and stability of the diagnostic model.
TABLE 2 comparison of diagnostic accuracy under different data sets
Figure BDA0003535601240000081
Fig. 5 compares the original MobileNetv3 network model with the modified MobileNetv3 network model of table 2 for experiments 4, 8, 12, 16, 20, and 24, with 10 reproducibility tests per experiment. The result shows that under the same input condition, the original MobileNetv3 network model testing precision has certain fluctuation, and the improved MobileNetv3 network model fault diagnosis is higher in accuracy and better in stability.
To further validate the diagnosis of the improved MobileNetv3 network, the results of experiment 4 in table 2 were visualized using a t-distribution random neighborhood embedding t-SNE. Fig. 6 shows a two-dimensional visualization of the classification results, after PCA (principal component analysis), the 7 operating states are well distinguished. The result shows that the improved MobileNetv3 network model can more effectively mine the characteristics in the data, so that the fault diagnosis accuracy is further improved.

Claims (6)

1. A fault diagnosis method for a rolling bearing under small sample data is characterized in that a bearing fault signal is collected by a bearing fault experiment table, the bearing fault experiment table is composed of an acceleration sensor, a motor, a test bearing and a loading module, and collected data are processed according to the following steps:
a. collecting original vibration signal data samples of a rolling bearing with faults to construct a task set T, and classifying the data samples in the task set T according to different working conditions and different fault types;
b. performing wavelet transformation on data samples in the task set T by using a Morlet wavelet basis function, converting each data sample into a two-dimensional wavelet image sample with fault characteristics, and then dividing the wavelet image samples in the task set T into a training set, a verification set and a test set;
c. inputting each wavelet image sample in the training set into an SSGAN model for training to obtain a corresponding generated sample, and adding the generated sample as an auxiliary training sample into the training set to complete the expansion of the training set;
d. inputting wavelet image samples in the expanded training set into an improved MobileNetv3 convolutional network to train the convolutional network, and adjusting parameters of the convolutional network by using a verification set in the training process;
e. inputting the wavelet image samples in the test set into a trained convolution network, and testing the convolution network;
f. and c, performing wavelet transformation on the vibration signal of the rolling bearing to be diagnosed by adopting the method in the step b, and inputting the vibration signal into a MobileNetv3 convolution network qualified in the test to finish the fault diagnosis of the small sample bearing.
2. The method for diagnosing the fault of the rolling bearing under the small sample data according to claim 1, wherein when wavelet transformation is carried out on data in a task set, cmor wavelets are selected as basic functions of the wavelet transformation, and the expression is as follows:
Figure FDA0003535601230000011
its corresponding fourier representation is:
Figure FDA0003535601230000012
wherein t represents time, f represents frequency, fbIs the shape parameter, a is the transformation scale, fcIs the center frequency.
3. The method for diagnosing the fault of the rolling bearing under the small sample data according to the claim 1 or 2, wherein before the generated sample is used as an auxiliary training sample and added into a training set, the generated sample is screened, and the specific screening steps are as follows:
a. the structural similarity SSIM (x, y) of each generated sample to the wavelet image samples of the respective original vibration signals in the training set is calculated using:
SSIM(x，y)＝[l(x，y)]α*[c(x，y)]β*[s(x，y)]γ
wherein:
Figure FDA0003535601230000021
Figure FDA0003535601230000022
Figure FDA0003535601230000023
where x denotes the pixel value in the original vibration signal wavelet image sample, y denotes the pixel value in the generated sample, μxIs the mean value of x, μyIs the average value of the values of y,
Figure FDA0003535601230000024
is the variance of x and is the sum of the variance of x,
Figure FDA0003535601230000025
variance of y, σxyIs the covariance of x and y, C1，C2，C3Three constants, alpha, beta and gamma are constants larger than 0;
b. calculating the junction of each generated sample and the wavelet image sample of each original vibration signal in the training setMean value of the structural similarity SSIM (x, y)
Figure FDA0003535601230000026
c. Will generate a sample according to
Figure FDA0003535601230000027
Sorting the magnitudes of the values of (a);
d. screening according to a set proportion
Figure FDA0003535601230000028
And generating a sample with a larger value and adding the sample into the training set to complete the expansion of the training set.
4. The method for diagnosing the fault of the rolling bearing under the small sample data as claimed in claim 3, wherein the improved MobileNetv3 convolutional network is characterized in that an original lightweight attention mechanism is replaced by a self-attention mechanism on the basis of an original MobileNetv3 convolutional network model; and simultaneously, an Early stop mechanism is added, and within a certain training turn, if the loss of the verification set is unchanged, the training of the convolution network model is stopped.
5. The method for diagnosing the fault of the rolling bearing under the small sample data as claimed in claim 4, wherein f isbIs set to a value of 3, fcThe value of (d) is set to 3.
6. The method for diagnosing the fault of the rolling bearing under the small sample data as claimed in claim 5, wherein the original vibration signal data samples of the rolling bearing with the fault in the task set T form a data sample matrix
Figure FDA0003535601230000029
Figure FDA00035356012300000210
Wherein xmRepresents the m-th sampleAnd M represents the number of samples,
Figure FDA00035356012300000211
the ith data representing the mth sample, i ═ 1,2,. n, represents the length of the sample.
CN202210217466.1A 2022-03-07 2022-03-07 Rolling bearing fault diagnosis method under small sample data Pending CN114707534A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210217466.1A CN114707534A (en) 2022-03-07 2022-03-07 Rolling bearing fault diagnosis method under small sample data

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210217466.1A CN114707534A (en) 2022-03-07 2022-03-07 Rolling bearing fault diagnosis method under small sample data

Publications (1)

Publication Number Publication Date
CN114707534A true CN114707534A (en) 2022-07-05

Family

ID=82169194

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210217466.1A Pending CN114707534A (en) 2022-03-07 2022-03-07 Rolling bearing fault diagnosis method under small sample data

Country Status (1)

Country Link
CN (1) CN114707534A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115639605A (en) * 2022-10-28 2023-01-24 中国地质大学（武汉） Automatic high-resolution fault identification method and device based on deep learning
CN116128882A (en) * 2023-04-19 2023-05-16 中汽数据(天津)有限公司 Motor bearing fault diagnosis method, equipment and medium based on unbalanced data set

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115639605A (en) * 2022-10-28 2023-01-24 中国地质大学（武汉） Automatic high-resolution fault identification method and device based on deep learning
CN116128882A (en) * 2023-04-19 2023-05-16 中汽数据(天津)有限公司 Motor bearing fault diagnosis method, equipment and medium based on unbalanced data set

Similar Documents

Publication Publication Date Title
CN109765053B (en) Rolling bearing fault diagnosis method using convolutional neural network and kurtosis index
CN104819846B (en) Rolling bearing sound signal fault diagnosis method based on short-time Fourier transform and sparse laminated automatic encoder
CN114707534A (en) Rolling bearing fault diagnosis method under small sample data
CN110595780B (en) Bearing fault identification method based on vibration gray level image and convolution neural network
CN110702411B (en) Residual error network rolling bearing fault diagnosis method based on time-frequency analysis
CN110657985B (en) Gearbox fault diagnosis method and system based on singular value spectrum manifold analysis
CN104198184A (en) Bearing fault diagnostic method based on second generation wavelet transform and BP neural network
CN105258789B (en) A kind of extracting method and device of vibration signal characteristics frequency band
CN109858352A (en) A kind of method for diagnosing faults based on compressed sensing and the multiple dimensioned network of improvement
CN110795843A (en) Method and device for identifying faults of rolling bearing
CN110879351A (en) Fault diagnosis method for non-linear analog circuit based on RCCA-SVM
CN107451515A (en) A kind of rotating machinery fault recognition method and system
CN107478418A (en) A kind of rotating machinery fault characteristic automatic extraction method
Sun et al. Data-driven fault diagnosis method based on second-order time-reassigned multisynchrosqueezing transform and evenly mini-batch training
CN114091539A (en) Multi-mode deep learning rolling bearing fault diagnosis method
CN112146882B (en) Bearing fault diagnosis method based on transfer learning vibration signal image recognition
CN117592543A (en) Aeroengine rolling bearing fault diagnosis method based on self-supervision learning
CN110044619B (en) Multi-fault feature identification method based on sparse multi-cycle group lasso
CN112183260A (en) One-way valve fault diagnosis method based on total variation noise reduction and RQA
CN116541771A (en) Unbalanced sample bearing fault diagnosis method based on multi-scale feature fusion
Han et al. Fault diagnosis of rolling bearings using dual-tree complex wavelet packet transform and time-shifted multiscale range entropy
Wan et al. Adaptive asymmetric real Laplace wavelet filtering and its application on rolling bearing early fault diagnosis
CN116610990A (en) Method and device for identifying hidden danger of breaker based on characteristic space differentiation
CN115982566A (en) Multi-channel fault diagnosis method for hydroelectric generating set
Chen et al. Gear fault diagnosis model combined with MED-LMD-Hypersphere multiclass SVM

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