CN111458142B - Sliding bearing fault diagnosis method based on generation of countermeasure network and convolutional neural network - Google Patents

Sliding bearing fault diagnosis method based on generation of countermeasure network and convolutional neural network Download PDF

Info

Publication number
CN111458142B
CN111458142B CN202010253907.4A CN202010253907A CN111458142B CN 111458142 B CN111458142 B CN 111458142B CN 202010253907 A CN202010253907 A CN 202010253907A CN 111458142 B CN111458142 B CN 111458142B
Authority
CN
China
Prior art keywords
fault
neural network
bearing
sliding bearing
convolutional neural
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.)
Active
Application number
CN202010253907.4A
Other languages
Chinese (zh)
Other versions
CN111458142A (en
Inventor
齐亮
黄晶
万振刚
袁文华
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.)
Suzhou Xinchuanpin Intelligent Technology Co ltd
Original Assignee
Suzhou Xinchuanpin Intelligent Technology Co ltd
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 Suzhou Xinchuanpin Intelligent Technology Co ltd filed Critical Suzhou Xinchuanpin Intelligent Technology Co ltd
Priority to CN202010253907.4A priority Critical patent/CN111458142B/en
Publication of CN111458142A publication Critical patent/CN111458142A/en
Application granted granted Critical
Publication of CN111458142B publication Critical patent/CN111458142B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
    • G01D21/00—Measuring or testing not otherwise provided for
    • G01D21/02—Measuring two or more variables by means not covered by a single other subclass
    • 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
    • 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
    • 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)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Acoustics & Sound (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The invention provides a sliding bearing fault diagnosis method based on generation of a countermeasure network and a convolutional neural network, which comprises the following steps: collecting vibration signals and bearing bush temperatures of the sliding bearing under the working conditions of no fault and different faults and carrying out pretreatment; carrying out overall average empirical mode decomposition on the vibration signal to obtain an optimal eigenmode function and obtain a corresponding characteristic frequency; constructing a state matrix by combining the temperature of the bearing bush, and respectively inputting the state matrix into a generation countermeasure network according to the category to generate samples; performing edge expansion on the data of the initial sample set, and setting a fault state label; constructing a sliding bearing convolutional neural network model, and training the model by using a data set and a corresponding fault label; and acquiring data of the current sliding bearing, constructing a state matrix, and inputting the state matrix into a trained neural network model for fault diagnosis and prediction. The invention effectively solves the problems of poor diagnosis effect due to unobvious vibration fault signals, insufficient utilization of shaft temperature and insufficient sample size.

Description

Sliding bearing fault diagnosis method based on generation of countermeasure network and convolutional neural network
Technical Field
The invention relates to the technical field of fault diagnosis, in particular to a sliding bearing fault diagnosis method based on generation of a countermeasure network and a convolutional neural network.
Background
In industrial manufacturing processes, sliding bearings are important components of machine parts, and many fault conditions of rotary machines are associated with sliding bearings. The sliding bearing is also one of the most easily damaged links in the process of manufacturing, and common faults are as follows: too large fit clearance of the bearing bush, oil film whirling, oil film oscillation, bearing bush abrasion, bush burning, tyre stripping and the like. These faults can degrade the rotational accuracy of the bearing, produce noise, vibration, and further jam the bearing, eventually rendering the entire rotating machine useless. Therefore, the method has very important significance in fault diagnosis of the sliding bearing.
The traditional sliding bearing fault diagnosis method is driven by data, mainly comprises the steps of collecting vibration signals of a bearing, extracting characteristics and carrying out fault diagnosis by combining a classifier according to the difference of fault characteristics. However, this method is not ideal in diagnostic effect when the vibration fault signal is weak. Meanwhile, the diagnosis method takes the temperature of the bearing bush as a simple alarm prompt instead of combining the temperature of the bearing bush. When the temperature of the bearing bush exceeds the warning value, the temperature of the bearing bush is increased due to a plurality of reasons, and accurate diagnosis cannot be achieved. The current popular fault diagnosis method based on deep learning combines feature extraction and pattern classification into one framework, but has high requirements on the number of samples. And when processing the vibration signal of high latitude, it needs to spend a great deal of manpower to carry out feature extraction on it. Increasing workload and failure diagnosis difficulty.
Disclosure of Invention
The invention provides a sliding bearing fault diagnosis method based on generation of a countermeasure network and a convolutional neural network, aiming at data of different fault types, the best eigenmode function is obtained by carrying out overall average empirical mode decomposition on a vibration signal, the characteristic frequency of the eigenmode function is further obtained, and a state matrix representing the state of a bearing is created by combining the temperature of a bearing bush. And then, a training sample set is created by generating a confrontation network, then the sample set is trained by using a proper convolutional neural network, and fault characteristics are learned layer by layer from a data set in a self-adaptive manner, so that accurate identification of different faults is realized.
The technical scheme of the invention is realized as follows: the sliding bearing fault diagnosis method based on the generation of the countermeasure network and the convolutional neural network comprises the following steps:
(a) acquiring vibration signals and bearing bush temperatures of different fault degrees of the sliding bearing under the fault-free and different fault working conditions, and preprocessing the vibration signals and the bearing bush temperatures to obtain an initial data set formed by the vibration signals and the corresponding bearing bush temperatures under the fault-free and multiple fault working conditions;
(b) carrying out overall average empirical mode decomposition on the vibration signals to obtain optimal eigenmode functions, and further obtaining the characteristic frequency corresponding to each optimal eigenmode function; then, a bearing state matrix is constructed by combining the corresponding bearing bush temperature, and the bearing state matrix is respectively input into a generation countermeasure network according to the category to generate samples; adding the generated samples into an original small-class sample training set to obtain a convolutional neural network initial sample training set;
(c) performing edge expansion on data of the initial sample set of the convolutional neural network to obtain a final sample training set; setting a fault state label according to the fault state corresponding to the data;
(d) constructing a sliding bearing convolution neural network model, taking the final sample set as the input of the sliding bearing convolution neural network model, taking the corresponding fault state label as the expected output of the sliding bearing convolution neural network model, and training the sliding bearing convolution neural network;
(e) acquiring a vibration signal of the current sliding bearing and the corresponding bearing bush temperature, and constructing a state matrix of the sliding bearing by adopting the same method in the step (b); and (d) inputting the matrix into the sliding bearing convolutional neural network model trained in the step (d) to obtain a fault diagnosis result of the current sliding bearing.
As a preferred technical scheme, in the step (a), when data of each fault working condition is collected, the data of the fault type is respectively collected according to three fault degrees, wherein the degrees are respectively primary formation, about to be formed and final formation; after collecting vibration signals and bearing bush temperatures of the bearing bush under the working conditions of no fault and different faults, preprocessing the data, and comprising the following steps of:
denoising the vibration signal;
representing the actual temperature of the bearing bush by an average value, and reserving three decimal places after the decimal place;
and the maximum value of the temperature change rate of the bearing bush represents the temperature change rate of the bearing bush at the moment.
As a preferred technical solution, in the step (b), before performing ensemble-averaged empirical mode decomposition on the vibration signal and obtaining an optimal eigenmode function through decomposition, performing autocorrelation processing on the vibration signal; when the best eigenmode function is selected from the eigenmode functions obtained after decomposition, the best eigenmode function is obtained by utilizing the correlation coefficient distribution diagram of each eigenmode function component and the original signal and combining the noise judgment criterion and the pseudo component judgment criterion; when the size and the amplitude of the characteristic frequency corresponding to the eigenmode function are solved, an envelope spectrogram is obtained by using Hilbert-Huang transform, and the envelope spectrogram is locally amplified to obtain accurate characteristic frequency and amplitude, and if no obvious characteristic frequency exists, the intermediate frequency is taken as the standard.
As a preferred solution, in step (b), a state matrix is constructed by combining the temperature of the bearing shell and the vibration signal of the bearing, as
Figure BDA0002436520680000031
Wherein N is the number of the optimal IMFs,
Figure BDA0002436520680000032
in order to represent the vector of the temperature,
Figure BDA0002436520680000033
the upper value of (1) is the bearing bush temperature, and the lower value is the change rate of the bearing bush temperature;
Figure BDA0002436520680000034
in order to represent a vector of frequencies, a,
Figure BDA0002436520680000035
the upper values are the characteristic frequency magnitudes of the IMF and the lower values are the corresponding magnitudes.
As a preferable technical solution, in the step (c), 0 is supplemented when the edge extension is performed.
As a preferred technical solution, in the step (d), the constructed convolutional neural network of sliding bearing has the following characteristics: it is composed of 3 convolution layers, 2 pooling layers and 3 full-connection layers; and adding an activation function correction linear function to 3 convolutional layers, 2 pooling layers and the first 2 full-connection layers; performing local response normalization processing on the pooling layer; overlapping pooling is used at the pooling level and dropout operation is used at the first two fully connected levels.
By adopting the technical scheme, the invention has the beneficial effects that:
the invention provides a sliding bearing state matrix which is constructed by combining the temperature of a bearing bush and a bearing vibration signal, wherein the state matrix comprises the temperature of the bearing bush, the change rate of the temperature of the bearing bush and the characteristic frequency and amplitude of the vibration signal, and the state matrix is analyzed, so that the problems of inaccurate diagnosis when a vibration fault signal is weak and insufficient utilization of the temperature of the bearing bush are better solved;
according to the invention, by constructing the bearing state matrix, the dimensionality of training data is greatly reduced, and the training efficiency for generating the countermeasure network is effectively improved. Meanwhile, the problems of lack of an original data set and unbalanced samples can be effectively solved;
the sliding bearing convolution neural network model constructed by the invention uses the activation functions which are all correction linear functions, thereby effectively relieving the problem of gradient disappearance. And the operation of correcting the linear function is simple, so that the training speed is accelerated. Meanwhile, the pooling layer uses local response normalization processing, and the performance of the network is further improved. And the overlapping pooling is used in the pooling layer and the lost output operation is used in the first two fully-connected layers, so that the overfitting is further reduced;
according to the invention, the data of a certain fault type are respectively collected according to the three fault degrees, so that the method can predict the fault.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flow chart of a fault diagnosis method constructed in accordance with a preferred embodiment of the present invention;
FIG. 2 is a diagram of a network model generated based on a bearing state matrix for generating a countermeasure network constructed in accordance with a preferred embodiment of the present invention;
figure 3 is a schematic diagram of a plain bearing convolutional neural network model constructed in accordance with a preferred embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without inventive step based on the embodiments of the present invention, are within the scope of protection of the present invention.
As shown in fig. 1, a sliding bearing fault diagnosis method based on generation of a countermeasure network and a convolutional neural network includes:
step 1, sampling frequency f 1 Collecting vibration signals of different fault degrees of the sliding bearing under the working conditions of no fault and different faults and using the frequency f 2 And collecting the temperature of the bearing bush corresponding to the vibration signal in the time period. Preprocessing each vibration signal, including denoising the vibration signal; the actual bearing temperature at that time is expressed as an average value of the bearing temperature, and the bearing temperature change rate at that time is expressed as a maximum value of the bearing temperature change rate.
Step 2: and performing autocorrelation processing on the preprocessed vibration signal to improve the signal-to-noise ratio.
And 3, step 3: and performing CEEMD decomposition on the signal subjected to autocorrelation processing, calculating a correlation coefficient of each IMF component and the original signal, drawing a correlation coefficient distribution graph, determining a noise IMF component by using a noise judgment criterion, determining a pseudo IMF component by using a pseudo component judgment criterion, and finally removing the noise IMF component and the pseudo component to obtain the optimal IMF.
And 4, step 4: and HHT is carried out on the optimal IMF to obtain an envelope spectrogram, and local amplification is carried out on the envelope spectrogram to obtain the characteristic frequency and amplitude corresponding to each IMF. And then combining the temperature of the bearing bush corresponding to the vibration signal to construct a state matrix. This stateThe specific size of the matrix is variable and is related to the number of optimal IMFs. But the state matrix size for the same fault should be the same. While the matrix is a 2 x (N +1) state matrix. And the matrix is in the form of
Figure BDA0002436520680000061
Where N is the number of optimal IMFs.
Figure BDA0002436520680000062
The upper value is the bushing temperature and the lower value is the bushing temperature rate.
Figure BDA0002436520680000063
The upper values are frequency magnitudes and the lower values are corresponding amplitude values.
Figure BDA0002436520680000064
The arrangement order of (2) is arranged in the order from small to large based on the corresponding frequency. And a failure data set is constructed.
And 5: for each fault data set, a new sample is generated by training with a GAN model as shown in fig. 2, and is added to the original data set to construct a CNN initial sample training set. And constructing samples with the same size for each type of data is beneficial to solving the problems of sample imbalance and insufficient quantity. The number of the training samples is 1000, 900 of which are used as training samples, and 100 of which are used as testing samples.
Step 6: and performing edge expansion on the CNN initial sample set sample, supplementing by 0, and constructing a final CNN training sample. This has the problem of reducing the loss of information during the convolution process. And the samples are all constructed to be 4 x n (wherein the size of n is selected as 10 according to specific conditions) so as to solve the problem that the samples cannot be trained due to different sizes. Meanwhile, for CNN training, a fault label state is set for each sample of each category.
And 7: and constructing a sliding bearing convolution neural network model, as shown in figure 3. And taking the final sample set as the input of the sliding bearing convolution neural network model, taking the corresponding fault label as the expected output of the sliding bearing convolution neural network, and training the sliding bearing convolution neural network.
And 8: and (3) acquiring a vibration signal of the current sliding bearing and the corresponding bearing bush temperature, and constructing a state matrix in the state by the state matrix constructing method in the steps 1,2, 3 and 4. And then, performing edge extension on the state matrix by using the extension method in the step 6, and inputting the state matrix into the final convolutional neural network model trained in the step 7 for fault diagnosis.
The terminology of the present embodiment is corresponded with, wherein: generating a countermeasure network (GAN), a Convolutional Neural Network (CNN), a global Empirical Mode Decomposition (CEEMD), an eigenmode Function (IMF), constructing a sliding bearing Convolutional Neural network model (modified LeNET) and correcting a linear Function (ReLU).
Further, in step 2, the autocorrelation function is used to describe the correlation degree between values of the random signal x (t) at any two different time instances s, t, and is defined as:
R(s,t)＝E(X(s)*X(t))
further, in step 3, the CEEMD decomposition is to add random white noise with opposite sign to the autocorrelation analyzed signal, and then perform Empirical Mode Decomposition (EMD) decomposition. The final formula is as follows:
Figure BDA0002436520680000071
Figure BDA0002436520680000072
wherein C is j (t) is the final IMF, C nj Is IMF, C obtained by decomposition after white noise is added -nj Is IMF obtained after white noise decomposition in the opposite direction is added. r is n (t) is the final remainder, r n And r -n Are respectively addedAnd decomposing the residual term by directional white noise.
Further, in step 4, HHT mainly obtains an envelope spectrogram, thereby obtaining a characteristic frequency and amplitude corresponding to each IMF. The transformation formula for HHT is as follows:
Figure BDA0002436520680000081
wherein PV is Cauchy principal value (Cauchy principal value). The formula denotes that Y (t) is X (t) and
Figure BDA0002436520680000082
is performed.
Further, in step 5, the countermeasure network is generated as shown in fig. 2, and there are one discriminator D and one generator G. The generator G receives a random noise z from which samples are generated, denoted G (z). And judging whether the network D is real or not. Its input parameter is x, x represents an input, and the output d (x) represents the probability that x is a true sample, if 1, it represents 100% true sample, and the output is 0, it represents a sample that is not possible to be true. In the training process, the goal of generating the network G is to generate a real sample as much as possible to deceive the discrimination network D. And the goal of D is to try to separate the G generated sample from the real sample. Thus, G and D constitute a dynamic "gaming process". In the most ideal state, G can generate enough samples G (z) to be "spurious". For D, it is difficult to determine whether the sample generated by G is real or not, so D (G (z)) is 0.5.
As used herein, the objective function for generating a competing network is:
Figure BDA0002436520680000083
in the formula, x represents a real sample, z represents noise input into a G network, G (z) represents a sample generated by the G network, and D (G (z)) represents the probability of judging whether a picture is real or not by the D network.
In a further step 6, the fault status label, is also the expected output of the CNN model. The fault status label is a one-dimensional vector with the format [0,0, …, a i ,…,0]Wherein the number of 0 is 3 times of the fault type and a i The value of (b) is 1. Starting with the second element, every third is a group, indicating a fault type. Here, taking 7 fault types as examples, the following are respectively: excessive fit clearance of the bearing bush, oil film whirling, oil film oscillation, bearing bush abrasion, bush burning, tire stripping and cracks. The labels are set as follows: (wherein the number of 0 s in each failure tag is 21) the failure-free failure tag is [1,0,0, …,0]]The fault label initially formed by fault type 1 is [0,1,0, …,0]]The fault type 1 is the fault label to be formed as [0,0,1,0, …,0]]The final fault label formed by fault type 1 is [0,0,0,1,0, …,0]]The fault label initially formed by fault type 2 is [0,0,0,0,1,0, …,0]]The fault label to be formed by the fault type 2 is [0,0,0,0,0,1,0, …,0]The final fault label formed by fault type 2 is [0,0,0,0,0,0,1,0, …,0]And the other labels are similar.
Further, in step 7, the convolutional neural network model is shown in fig. 3, and the final output layer outputs a one-dimensional vector, elements in the vector represent the probability that the fault is the fault type, and fault diagnosis and fault prediction are performed according to the vector. As exemplified here with 7 fault types, the final output one-dimensional vector is [0.02,0.01,0.92,0.01,0,0.01,0,0.02,0.01,0,0,0,0,0,0,0,0 ]. By combining the fault label in step 6, it can be known that the vector indicates that the diagnosis result is: the probability of no fault is 2%, the probability of the fault type 1 being initially formed is 1%, the probability of the fault type 1 being about to be formed is 92%, the probability of the fault type 1 being finally formed is 1%, the probability of the fault type 2 being initially formed is 0%, the probability of the fault type 2 being about to be formed is 1%, the probability of the fault type 2 being finally formed is 0%, the probability of the fault type 3 being initially formed is 2%, the probability of the fault type 3 being about to be formed is 1%, the probability of the fault type 3 being finally formed is 0%, and the probabilities of the fault types 4, 5, 6, and 7 are all 0%. It can therefore be finally determined that the fault type is that fault 1 is about to form. The operation process is (the initial settings are as follows: convolutional layer C1 has 20 convolutional kernels, convolutional layer C3 has 50 convolutional kernels, convolutional layer C5 has 50 convolutional kernels, fully-connected layer F6 contains 600 neurons, and fully-connected layer F7 contains 600 neurons):
7.1 input data
And inputting the state matrix with the same size after the edge is expanded.
7.2 perform the first convolution operation
The convolutional layer C1 has 20 convolutional kernels, 2 x 2 in size, and the input layers are convolved for step 1 and weight normalized. And then, using a ReLU activation function for the convolution result to obtain an activation result.
The convolution formula is as follows:
Figure BDA0002436520680000101
wherein Q is 1,2, Q, and when Q is 1,
Figure BDA0002436520680000102
represents the input data of the input layer, and when Q is 2.., Q represents the number of feature maps output in the upper layer,
Figure BDA0002436520680000103
denotes the first
Figure BDA0002436520680000104
A characteristic map is generated by the device,
Figure BDA0002436520680000105
representing the bias of the feature map obtained by the kth kernel in the l-th layer,
Figure BDA0002436520680000107
represents the weight of the feature map obtained by the kth convolution kernel in the l-th layer, and f (x) represents the ReLU activation function.
The ReLU formula is as follows:
ReLU(x)＝max(0,x)
7.3 performing a first pooling operation
The pooling layer S2 employs an average pooling operation and is an overlapping pooling. The average pooling operation with pooling kernel size of 2 x 2 and step size of 1 was first used for the results of C1. And then, using a ReLU activation function for the pooling result to obtain an activation result. And then using local response normalization operation on the activation result.
The local response normalization formula is as follows:
Figure BDA0002436520680000108
wherein
Figure BDA0002436520680000109
Represents the value at position (x, y) on the ith volume area,
Figure BDA00024365206800001010
and the value of the local response normalization is shown, N is the total number of convolution surfaces, N is the number of adjacent surfaces, and K, alpha and beta are adjustable parameters.
7.4 perform the second convolution operation
The convolutional layer C3 has 50 convolutional kernels, which are connected to S2 in the following relationship: in 50 convolution kernels, 20 neurons are locally connected with 9 feature maps of an S2 layer; there are local connections between 27 neurons and 12 signatures at the S2 level; there are 3 neurons connected to all profiles at the S2 level. First, a convolution operation with a convolution kernel size of 2 × 2 and a step size of 2 is performed on the result of S2. And then, using a ReLU activation function for the convolution result to obtain an activation result.
7.5 performing a second pooling operation
The pooling layer S4 employs an average pooling operation and is an overlapping pooling. The average pooling operation with pooling kernel size of 1 x 2 and step size of 2 was first used for the results of C3. And then, using a ReLU activation function for the pooling result to obtain an activation result. And then using local response normalization operation on the activation result.
7.6 carry out the third convolution operation
The convolutional layer C5 has 50 convolutional kernels. First, a convolution operation with a convolution kernel size of 1 × 2 and a step size of 1 is performed on the result of S4. And then, using a ReLU activation function for the convolution result to obtain an activation result.
7.7 carry out the first full-join operation
The fully connected layer F6 includes 600 neurons, each neuron is connected with all neurons of the C5 layer, the value of the neuron is obtained by inner product of the output vector of the C5 layer and the weight vector of the F6 layer, a bias is added, an activation result is obtained through a Linear correction function (RecU), a dropout operation is used for the activation result, and the parameter is 0.5, so that the result is obtained.
7.8 performing a second full join operation
The fully connected layer F7 includes 600 neurons, each neuron is connected with all neurons of the F6 layer, the value of the neuron is obtained by inner product of the output vector of the F6 layer and the weight vector of the F7 layer, a bias is added, an activation result is obtained through a Linear correction function (RecU), a dropout operation is used for the activation result, and the parameter is 0.5, so that the result is obtained.
7.9 for output
This layer is a 22-way soft maximum output layer that is used to generate a label distribution that covers 22 classes.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (5)

1. The sliding bearing fault diagnosis method based on the generation of the countermeasure network and the convolutional neural network is characterized by comprising the following steps of:
(a) collecting vibration signals and bearing bush temperatures of different fault degrees of the sliding bearing under the fault-free and different fault working conditions, and preprocessing the data to obtain an initial data set formed by the vibration signals and the corresponding bearing bush temperatures under the fault-free and multiple fault working conditions;
(b) ensemble averaging of vibration signalsDecomposing by empirical mode decomposition (CEEMD) to obtain N optimal IMFs, and obtaining characteristic frequency corresponding to each IMF; combining with the temperature of the corresponding bearing bush to construct a shape
Figure FDA0003738812670000011
And respectively inputting the state matrixes with the size of 2 x (N +1) into a generation countermeasure network (GAN) according to the categories to generate samples, wherein
Figure FDA0003738812670000012
In order to represent a vector of temperatures,
Figure FDA0003738812670000013
the first element of (a) is the bearing bush temperature, and the second element is the change rate of the bearing bush temperature;
Figure FDA0003738812670000014
in order to represent a vector of frequencies, a,
Figure FDA0003738812670000015
the first element is the characteristic frequency of the IMF, the second element is the amplitude corresponding to the characteristic frequency of the IMF, and N is the number of the optimal IMF; adding the generated samples into an original small-class sample training set to obtain a convolutional neural network initial sample training set;
(c) performing edge expansion on data of the convolutional neural network initial sample training set to obtain a final sample training set; setting a fault state label according to the fault state corresponding to the data;
(d) constructing a sliding bearing convolution neural network model, taking a final sample training set as the input of the sliding bearing convolution neural network model, taking a corresponding fault state label as the expected output of the sliding bearing convolution neural network model, and training the sliding bearing convolution neural network model;
(e) acquiring a vibration signal of the current sliding bearing and the corresponding bearing bush temperature, and constructing a current sliding bearing state matrix by adopting the same method in the step (b); and (d) inputting the matrix into the sliding bearing convolutional neural network model trained in the step (d) to obtain a fault diagnosis result of the current sliding bearing.
2. The method for diagnosing faults of sliding bearings based on generation of a countermeasure network and a convolutional neural network as claimed in claim 1, wherein in the step (a), after acquiring vibration signals of different fault degrees of the sliding bearings under no fault and different fault conditions and temperatures of bearing bushes, preprocessing the data, including:
denoising the vibration signal;
the actual temperature of the bearing bush is represented by the average value of the temperature of the bearing bush, and three decimal places after the decimal place are reserved;
and the bearing bush temperature change rate is expressed as the maximum value of the bearing bush temperature change rate.
3. The plain bearing failure diagnosis method based on generation of a countermeasure network and a convolutional neural network as claimed in claim 1, wherein in step (b), the vibration signal is subjected to autocorrelation processing before being subjected to ensemble-averaged empirical mode decomposition to obtain an optimal eigenmode function; when the best eigenmode function is selected from the eigenmode functions obtained after decomposition, the best eigenmode function is obtained by utilizing the correlation coefficient distribution diagram of each eigenmode function component and the original signal and combining the noise judgment criterion and the pseudo component judgment criterion; when the size and the amplitude of the characteristic frequency corresponding to the eigenmode function are solved, an envelope spectrogram is obtained by using Hilbert-Huang transform, and the envelope spectrogram is locally amplified to obtain accurate characteristic frequency and amplitude, and if no obvious characteristic frequency exists, the intermediate frequency is taken as the standard.
4. The sliding bearing failure diagnosis method based on generation of a countermeasure network and a convolutional neural network as claimed in claim 1, characterized in that in step (c), it is supplemented with 0 when edge extension is performed.
5. The plain bearing fault diagnosis method based on generation of a countermeasure network and a convolutional neural network as claimed in claim 1, characterized in that in step (d), the constructed plain bearing convolutional neural network model is composed of 3 convolutional layers, 2 pooling layers and 3 full-link layers; and adding an activation function correction linear function to 3 convolutional layers, 2 pooling layers and the first 2 full-connection layers; performing local response normalization processing on the pooling layer; overlapping pooling was used at the pooling level and dropout operation was used at the first 2 fully connected levels.
CN202010253907.4A 2020-04-02 2020-04-02 Sliding bearing fault diagnosis method based on generation of countermeasure network and convolutional neural network Active CN111458142B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010253907.4A CN111458142B (en) 2020-04-02 2020-04-02 Sliding bearing fault diagnosis method based on generation of countermeasure network and convolutional neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010253907.4A CN111458142B (en) 2020-04-02 2020-04-02 Sliding bearing fault diagnosis method based on generation of countermeasure network and convolutional neural network

Publications (2)

Publication Number Publication Date
CN111458142A CN111458142A (en) 2020-07-28
CN111458142B true CN111458142B (en) 2022-08-23

Family

ID=71677050

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010253907.4A Active CN111458142B (en) 2020-04-02 2020-04-02 Sliding bearing fault diagnosis method based on generation of countermeasure network and convolutional neural network

Country Status (1)

Country Link
CN (1) CN111458142B (en)

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112182490B (en) * 2020-09-01 2024-02-02 华中科技大学 Reactor state diagnosis method and system
CN112146880B (en) * 2020-09-17 2022-03-29 天津大学 Intelligent diagnosis method for internal structure faults of rolling bearing at different rotating speeds
CN112329520B (en) * 2020-09-23 2023-01-31 西南交通大学 Truck bearing fault identification method based on generation countermeasure learning
CN112395968B (en) * 2020-11-11 2021-08-27 北京京航计算通讯研究所 Mechanical rotating part fault diagnosis method and device based on neural network
CN112308038B (en) * 2020-11-25 2022-09-27 哈尔滨工业大学 Mechanical equipment fault signal identification method based on classroom type generation confrontation network model
CN112577745A (en) * 2020-12-02 2021-03-30 上海应用技术大学 Rolling bearing fault diagnosis method based on 1D-CNN
CN112649198B (en) * 2021-01-05 2023-04-18 西交思创智能科技研究院(西安)有限公司 Intelligent fault diagnosis method, system and equipment for quasi-unbalanced rolling bearing and application
CN112903294B (en) * 2021-01-07 2023-07-14 泰华宏业(天津)智能科技有限责任公司 Rolling bearing fault diagnosis method based on VMD and deep convolution neural network
CN113256443B (en) * 2021-04-29 2023-06-20 西安交通大学 Nuclear power water pump guide bearing fault detection method, system, equipment and readable storage medium
CN113379123A (en) * 2021-06-09 2021-09-10 中德（珠海）人工智能研究院有限公司 Fault prediction method, device, server and storage medium based on digital twin
CN113821888B (en) * 2021-09-23 2024-02-27 西安热工研究院有限公司 Vibration data fault diagnosis method based on periodic impact feature extraction and echo state network

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106874957A (en) * 2017-02-27 2017-06-20 苏州大学 A kind of Fault Diagnosis of Roller Bearings
CN108764341A (en) * 2018-05-29 2018-11-06 中国矿业大学 A kind of adaptive deep neural network model of operating mode and variable working condition method for diagnosing faults

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA2689615C (en) * 2007-06-04 2016-08-23 Eaton Corporation System and method for bearing fault detection using stator current noise cancellation
CN101430240B (en) * 2008-11-28 2010-08-18 华北电力大学 On-line real-time diagnosis method for parallel misalignment fault of coupling
CN101776508A (en) * 2010-01-19 2010-07-14 西安交通大学 Optical-mechanical-electrical multi-parameter accurate monitoring device of rotating machine
CN105298823A (en) * 2015-11-23 2016-02-03 武汉大学 Large pump unit on-line monitoring and diagnosis system
CN107526853B (en) * 2016-06-22 2018-09-21 北京航空航天大学 Rolling bearing fault mode identification method and device based on stacking convolutional network
CN106895975B (en) * 2017-01-17 2019-03-15 苏州大学 Bearing fault diagnosis method based on Stacked SAE deep neural network
CN109813551A (en) * 2019-01-28 2019-05-28 江苏科技大学 A kind of online quantitative detecting method of diesel engine main bearing breaking-in quality
CN110942101B (en) * 2019-11-29 2023-06-27 湖南科技大学 Rolling bearing residual life prediction method based on depth generation type countermeasure network

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106874957A (en) * 2017-02-27 2017-06-20 苏州大学 A kind of Fault Diagnosis of Roller Bearings
CN108764341A (en) * 2018-05-29 2018-11-06 中国矿业大学 A kind of adaptive deep neural network model of operating mode and variable working condition method for diagnosing faults

Also Published As

Publication number Publication date
CN111458142A (en) 2020-07-28

Similar Documents

Publication Publication Date Title
CN111458142B (en) Sliding bearing fault diagnosis method based on generation of countermeasure network and convolutional neural network
CN109918752B (en) Mechanical fault diagnosis method, equipment and medium based on migration convolution neural network
CN106874957A (en) A kind of Fault Diagnosis of Roller Bearings
CN206504869U (en) A kind of rolling bearing fault diagnosis device
CN111562108A (en) Rolling bearing intelligent fault diagnosis method based on CNN and FCMC
CN112254964A (en) Rolling bearing fault diagnosis method based on rapid multi-scale convolution neural network
CN112629851B (en) Offshore wind turbine gearbox fault diagnosis method based on data enhancement method and image recognition
CN104616033A (en) Fault diagnosis method for rolling bearing based on deep learning and SVM (Support Vector Machine)
CN114358123B (en) Generalized open set fault diagnosis method based on deep countermeasure migration network
CN112765890B (en) Dynamic domain adaptive network-based multi-working-condition rotating machine residual life prediction method
CN112257530A (en) Rolling bearing fault diagnosis method based on blind signal separation and support vector machine
CN114358124B (en) New fault diagnosis method for rotary machinery based on deep countermeasure convolutional neural network
CN113375941A (en) Open set fault diagnosis method for high-speed motor train unit bearing
CN112487890B (en) Bearing acoustic signal fault diagnosis method based on parallel sparse filtering
Islam et al. Time–frequency envelope analysis-based sub-band selection and probabilistic support vector machines for multi-fault diagnosis of low-speed bearings
CN113505655A (en) Bearing fault intelligent diagnosis method for digital twin system
CN115791174B (en) Rolling bearing abnormality diagnosis method, system, electronic equipment and storage medium
CN112729834B (en) Bearing fault diagnosis method, device and system
CN111783531A (en) Water turbine set fault diagnosis method based on SDAE-IELM
CN116593157A (en) Complex working condition gear fault diagnosis method based on matching element learning under small sample
CN115046766A (en) Small sample bearing fault diagnosis method based on two-dimensional gray image self-adaptive subspace
CN116152678A (en) Marine disaster-bearing body identification method based on twin neural network under small sample condition
CN115683620A (en) Gearbox fault diagnosis method based on group decomposition and improved multi-scale weight dispersion entropy
CN113642662B (en) Classification detection method and device based on lightweight classification model
Li et al. An Improved LeNet-5 Convolutional Neural Network Supporting Condition-Based Maintenance and Fault Diagnosis of Bearings

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
TA01 Transfer of patent application right

Effective date of registration: 20210118

Address after: 215300 room 511, 5 / F, block B, building 1, Zhongke innovation Plaza, Huaqiao Town, Kunshan City, Suzhou City, Jiangsu Province

Applicant after: Suzhou xinchuanpin Intelligent Technology Co.,Ltd.

Address before: Room 1702-7, building 1, paddy Internet Industrial Park, 399 Xiarong street, Taihu new town, Wujiang District, Suzhou City, Jiangsu Province 215000

Applicant before: Suzhou Zhichuan New Automation Technology Co.,Ltd.

TA01 Transfer of patent application right
GR01 Patent grant
GR01 Patent grant