CN116793682A - Bearing fault diagnosis method based on iCORAL-MMD and anti-migration learning - Google Patents
Bearing fault diagnosis method based on iCORAL-MMD and anti-migration learning Download PDFInfo
- Publication number
- CN116793682A CN116793682A CN202310839080.9A CN202310839080A CN116793682A CN 116793682 A CN116793682 A CN 116793682A CN 202310839080 A CN202310839080 A CN 202310839080A CN 116793682 A CN116793682 A CN 116793682A
- Authority
- CN
- China
- Prior art keywords
- domain
- fault diagnosis
- mmd
- model
- icoral
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000003745 diagnosis Methods 0.000 title claims abstract description 125
- 238000000034 method Methods 0.000 title claims abstract description 66
- 238000013508 migration Methods 0.000 title claims abstract description 55
- 238000012549 training Methods 0.000 claims abstract description 83
- 238000012360 testing method Methods 0.000 claims abstract description 19
- 230000006870 function Effects 0.000 claims description 45
- 238000009826 distribution Methods 0.000 claims description 29
- 230000005012 migration Effects 0.000 claims description 27
- 238000011176 pooling Methods 0.000 claims description 21
- 230000004913 activation Effects 0.000 claims description 17
- 235000014653 Carica parviflora Nutrition 0.000 claims description 12
- 241000243321 Cnidaria Species 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 11
- 238000007781 pre-processing Methods 0.000 claims description 11
- 230000006978 adaptation Effects 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 6
- 230000003044 adaptive effect Effects 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 5
- 230000036541 health Effects 0.000 claims description 4
- 238000012544 monitoring process Methods 0.000 claims description 4
- 238000000844 transformation Methods 0.000 claims description 4
- 238000013459 approach Methods 0.000 claims description 2
- 238000013527 convolutional neural network Methods 0.000 abstract description 6
- 238000005516 engineering process Methods 0.000 abstract description 3
- 238000000605 extraction Methods 0.000 description 11
- 230000008569 process Effects 0.000 description 9
- 238000005096 rolling process Methods 0.000 description 9
- 230000008901 benefit Effects 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 238000013135 deep learning Methods 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 238000001914 filtration Methods 0.000 description 4
- 230000001133 acceleration Effects 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000012423 maintenance Methods 0.000 description 3
- 238000011897 real-time detection Methods 0.000 description 3
- 238000013526 transfer learning Methods 0.000 description 3
- 238000003491 array Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000012512 characterization method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000002405 diagnostic procedure Methods 0.000 description 1
- 238000010892 electric spark Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000003754 machining Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
- G06F18/15—Statistical pre-processing, e.g. techniques for normalisation or restoring missing data
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06N3/094—Adversarial learning
-
- 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
- G06N3/096—Transfer learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- 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
Abstract
The invention belongs to the technical field of information technology service, and discloses a bearing fault diagnosis method based on iCORAL-MMD and deep convolution anti-migration learning, which is used for collecting vibration signals of a bearing under the working condition of A, B; after pretreatment, converting the time domain signal into a frequency domain signal based on fast Fourier transform, and dividing a training set and a testing set proportionally; establishing an initial source domain 1DCNN fault diagnosis model, and training through source domain data to obtain a source domain pre-training model; sharing model parameters to a target domain fault diagnosis model for initialization, establishing an initial domain discriminator, and training the discriminator through characteristic signals; training the initial fault diagnosis model of the target domain by using iCORAL-MMD to obtain a domain discriminator and a target domain feature extractor; and combining the two to obtain a fault diagnosis model of the target domain. The invention provides a method for combining an countermeasure network with a related alignment metric and adapting to a maximum mean difference domain based on a one-dimensional convolutional neural network so as to enhance the fault diagnosis capability of a model to rotary machinery under different working conditions.
Description
Technical Field
The invention belongs to the technical field of information technology service, and particularly relates to a bearing fault diagnosis method based on iCORAL-MMD (improved Correlation Alignment joint Maximum Mean discrepancy) and anti-migration learning.
Background
Rotary machines are very common in large machinery equipment, and as a key component of the machinery, fault diagnosis thereof plays an important role in modern industry. With the development of intelligent manufacturing, data can be collected faster, which brings new perspectives and challenges to the industry. Data-driven fault diagnosis has attracted much research in recent years as a typical fault diagnosis. Therefore, to avoid catastrophic failure of the rotating machine, it is important to find a more efficient data driven failure diagnostic method.
Machine learning techniques have been applied to data driven fault diagnosis. In the conventional machine learning method, manual feature extraction should be performed in advance, and it has been proven that these manual extracted features define the upper limit performance of the machine learning method. But it is difficult to design the artificial function in advance. In recent years, deep learning is a new field of machine learning, and can automatically extract deep characterization features of original data. By utilizing the advantage, the influence of manually extracting the characteristics can be avoided, and the method has good prospect in the aspect of fault diagnosis. However, the traditional machine learning method works well under the general assumption: the training data and the test data should come from the same distribution. When the distribution is different, the performance of these methods is degraded, and the deep method has the above problems. Because of different equipment, environmental conditions and the like, the distribution of laboratory and actual scene data is often different, and most of the methods cannot directly deal with the problem of fault diagnosis after scene switching. To address this problem, domain adaptation is a popular approach to migration learning that performs a learning task (called a source domain problem) on a laboratory ideal data set, and then performs the same task (called a target domain problem) on a test data set of the actual scenario-related distribution. Deep transfer learning is an example of performing transfer learning using deep learning. Deep migration learning provides greater flexibility in extracting advanced features that transition from source problems to target problems compared to shallow structures, and has proven to be generalized in a variety of scientific and engineering problems.
Through the above analysis, the problems and defects existing in the prior art are as follows: in the prior art, training and testing are mostly carried out in a large amount of labeled data collected in a laboratory, and effective training and diagnosis on actual scene monitoring data cannot be carried out; and because of different equipment, environmental conditions and the like, the distribution of laboratory and actual scene data is often different, and most of the methods cannot directly process the fault diagnosis problem after scene switching.
Disclosure of Invention
Aiming at the problems existing in the prior art, the invention provides a method, a system, equipment and a terminal for optimizing data synchronization among serial chips.
The invention is realized in such a way, a bearing fault diagnosis method based on iCORAL-MMD and anti-migration learning is realized, firstly, preprocessing and frequency domain signal transformation are carried out on the collected bearing vibration data; secondly, building a basic diagnosis model and training the model by using a large amount of labeled data; then, the model is migrated to the monitoring data by using the obtained pre-training model and combining the improved CORAL-MMD self-adaption (iCORAL-MMD) and the countermeasure method, so that fault diagnosis can be realized.
Further, the bearing fault diagnosis method based on iCORAL-MMD and anti-migration learning comprises the following steps:
step one, collecting vibration signals of a bearing under the working condition A, B;
step two, the signals under the working condition A, B are respectively used as sample data of a source domain and a target domain, after preprocessing, the time domain signals are converted into frequency domain signals based on fast Fourier transform, and a training set and a testing set are proportionally divided;
step three, an initial source domain 1DCNN fault diagnosis model is established, training is carried out through source domain data, and a source domain pre-training model is obtained;
step four, sharing the parameters of the source domain fault diagnosis model to the target domain fault diagnosis model for initialization, establishing an initial domain discriminator, and training the domain discriminator through characteristic signals;
training the initial fault diagnosis model of the target domain by using iCORAL-MMD to obtain a domain discriminator and a target domain feature extractor;
and step six, combining the source domain classifier with the target domain feature extractor to obtain a fault diagnosis model of the target domain.
Further, the second step includes:
working condition A data is used as a source domain signalIncluding N health states;
working condition B data is used as a target domain signalIncluding N health states;
normalizing the vibration signal to generate a normalized signal;
and carrying out Fourier transform on the normalized signal to generate a frequency domain signal.
Further, the normalization processing of the one-dimensional time sequence signal specifically comprises the following steps:
in the formula ,xi For the normalized signal at the i-th instant,the time sequence signal at the i-th moment in the one-dimensional time sequence signal is max (X) which is the maximum value in the one-dimensional time sequence signal, and min (X) which is the minimum value in the one-dimensional time sequence signal.
Further, performing fast fourier transform on the source domain sample and the target domain sample to obtain a frequency domain signal, which specifically includes:
wherein the vibration signal x n Obtaining X through FFT k ，X k Is in the form of a plurality.
Further, the fault diagnosis model uses cross entropy as a loss function as follows:
wherein ,the j-th element of the network output vector representing the last layer in the classifier takes as input the i-th source domain sample. />A corresponding transmission component status label is indicated, and N is the number of the associated component status.
Further, the source domain fault diagnosis model pre-training specifically comprises the following steps:
wherein ,θS The network model representing source domain fault diagnosis involves parameters, delta represents learning rate.
Further, the domain discrimination model and training thereof are specifically as follows:
the domain discrimination model adopts a full-connection layer, takes Softmax as an activation function output result as a source domain or a target domain, and the loss is as follows:
wherein , and />2 represents the first and second elements of the output vector of the last layer in the domain discriminator, respectively, taking as input the i-th source domain sample,/> and />Is the corresponding element of the ith target domain sample; assuming that the first output element represents the source domain and the second represents the target domain, the parametric stochastic gradient descent update is as follows:
wherein ,θD The representation domain discriminant model involves parameters, δ representing the learning rate.
Further, the difference of the output characteristic distribution of the target domain characteristic extractor network is measured, and the updating of the target domain characteristic extractor network by the minimized difference is specifically as follows:
introducing domain adaptive learning modules into FC1 and FC2 of the fully connected layer, calculating covariance distances of features at source domain and target domain FC1 and FC2 and measuring distribution differences by using MMD, which is defined as iCORAL-MMD loss and is expressed as L G The method comprises the steps of carrying out a first treatment on the surface of the CORAL is an effective and simple unsupervised adaptive method widely used to measure differences between source and target domains in model identification, for example by exploring second order statistics of source and target domains to align their input feature distributions; thus, the only calculation it requires is to calculate covariance statistics in each domain; MMD directly measures the distribution difference distance between FS and FT by reducing learning from different domainsThe distribution difference distance between the features achieves domain self-adaption; compared to MMD, the difference is that MMD-based methods generally apply the same transformations to the source and target domains, their asymmetric transformations are more flexible, and the combined application of the two yields better performance for domain adaptation tasks;
the calculation formula is as follows:
λ 1 and λ2 The coefficients of CORAL loss and MMD loss,is the Frobenius norm of the matrix, n s Is the number of training samples from the source domain, n t Is the number of training samples from the target domain, FS i Is output data of a source domain through a full connection layer FC, FT j Is the output data of the target domain through the full connection layer FC H Is a regenerated kernel hilbert space. C (C) S and CT Is the covariance matrix of the source domain and the target domain, and the calculation formula is as follows:
where 1 is the column vector with element 1, F S Is output data of source domain through full connection layer FC, F T Is output data of target domain through full connection layer FC, n S and nT The number of samples for the source domain and the target domain, respectively, the gradient is calculated as follows:
the feature extractor network parameters are updated as:
wherein ,θD Representing parameters involved in the target domain feature extractor, and delta represents the learning rate.
Further, based on the iCORAL-MMD and the deep convolution, the total loss function adopted by the target domain fault diagnosis model of the anti-migration learning network is as follows:
L T ＝L G -L D ；
related parameters:
another object of the present invention is to provide a bearing fault diagnosis system based on iCORAL-MMD and deep convolution anti-migration learning, which applies the bearing fault diagnosis method based on iCORAL-MMD and deep convolution anti-migration learning, the bearing fault diagnosis system based on iCORAL-MMD and deep convolution anti-migration learning includes an input layer, at least one convolution layer, at least one pooling layer, at least one fully connected layer, and an output layer;
the convolution layer comprises a plurality of convolution kernels of 3 multiplied by 3, the convolution step length is 1, and the activation function of the convolution layer is a Leakyrlu function;
the pooling mode of the pooling layer is maximum pooling, the pooling step length is 2, and the window size of the pooling layer is 2 multiplied by 2;
the activation function of the full connection layer is a Leakyrelu function, which is:
wherein f (x) is a LeakyRelu function; x is the input vector of the upper layer;
the activation function of the full-connection output layer of the classifier is a Softmax function, and the Softmax function is as follows:
in the formula ,zi An output value of the i-th node; z j A total output value for a single node; n is the number of output nodes; e is a natural constant.
In combination with the technical scheme and the technical problems to be solved, the technical scheme to be protected has the following advantages and positive effects:
firstly, through introducing iCORAL-MMD and deep convolution anti-migration learning, the accuracy of bearing fault diagnosis is improved, bearing fault diagnosis under different working conditions can be self-adapted, and model training cost is reduced. Meanwhile, the method has good universality and expansibility, is suitable for bearing fault diagnosis scenes of different types, can monitor and diagnose bearing faults in real time, is beneficial to timely finding and processing potential problems, and reduces equipment fault rate and maintenance cost.
Most of the domain self-adaption-based diagnosis methods are to match a source domain with a target domain by only adopting a domain adaptation method such as an countermeasure network or a Maximum Mean Difference (MMD) metric under the assumption that the source domain tag set is the same as the target domain tag set. The invention provides a method for resisting joint domain adaptation of a network, relative alignment measurement (CORAL) and Maximum Mean Difference (MMD) based on a one-dimensional convolutional neural network so as to enhance the fault diagnosis capability of a model to rotary machinery under different working conditions.
Secondly, in the fault diagnosis method based on the deep convolution domain anti-migration learning, in the iCORAL-MMD-DCAT, the gradient problems such as gradient disappearance, gradient divergence and the like in the iCORAL-MMD-DCAT training process can be avoided by utilizing the deep convolution feature extractor to perform high-layer feature extraction, and the nonlinear fitting capacity of the iCORAL-MMD-DCAT can be improved; performing domain countermeasure training between a source domain and a target domain by utilizing gradient inversion in a domain discriminator, wherein the domain countermeasure training is performed by using the characteristic distribution of a source domain labeled sample and a target domain unlabeled sample, so that the domain adaptability of the source domain labeled sample to the target domain can be enhanced, and meanwhile, the difference between the target domain and the depth characteristic distribution of the source domain is reduced by measuring the depth characteristic distribution of the target domain by iCORAL-MMD, so that the domain invariant characteristics of the source domain and the target domain are obtained through learning, and the migration performance of the iCORAL-MMD-DCAT is improved; feature migration and classification based on minimizing the domain adaptation overall loss function of iCORAL-MMD-DCAT may improve diagnostic accuracy after migration. The advantages of the iCORAL-MMD-DCAT enable the fault diagnosis method based on the iCORAL-MMD-DCAT to carry out high-precision fault diagnosis on the current sample to be tested of the rotary machine by utilizing the sample with the label of the source domain under the condition that the sample with the label of the target domain under the current working condition of the bearing does not exist.
Thirdly, as inventive supplementary evidence of the claims of the present invention, the following important aspects are also presented:
(1) The expected benefits and commercial values after the technical scheme of the invention is converted are as follows:
in various modern industrial fields such as intelligent transportation and construction machinery, huge mechanical equipment operates under severe conditions such as high temperature, extreme pressure, long-term overload and the like, which increases the possibility of failure. Deviations in critical rotating components such as bearings and gearboxes can lead to mechanical failure or potentially catastrophic accidents, thereby emphasizing the importance of equipment failure detection and maintenance, and accurate failure diagnosis can improve the predictability and maintainability of the equipment. The bearing fault diagnosis method and the system based on iCORAL-MMD and deep convolution anti-migration learning are used as a safe and reliable diagnosis technology, and can effectively fill the defect of fault diagnosis under the same traditional working condition.
(2) Whether the technical scheme of the invention solves the technical problems that people want to solve all the time but fail to obtain success all the time is solved:
according to the technical scheme, the problem of insufficient actual data quantity is solved by training a large amount of labeled data in a laboratory, and further, the migration method of iCORAL-MMD combined countermeasure is provided to realize migration among related data with different distribution.
(3) The technical scheme of the invention overcomes the technical bias:
in practical engineering application, the diagnosis is still carried out by using a traditional signal processing method, the interpretation of deep learning is still a problem to be researched, but the excellent performance of the deep learning in the technical scheme of the invention in feature extraction and classification is proved by experiments. In addition, in the face of actual untagged monitoring data, the combination of CORAL and MMD for migration learning is well applied to the situation of different distribution data.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the embodiments of the present invention will be briefly described below, and it is obvious that the drawings described below are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a bearing fault diagnosis method based on iCORAL-MMD and deep convolution challenge migration learning provided by an embodiment of the invention;
FIG. 2 is a flow chart of signal processing under the working condition A, B provided by an embodiment of the invention;
FIG. 3 is a graph of signal processing results under a working condition A, B provided by an embodiment of the present invention;
FIG. 4 is a flow chart of a bearing fault diagnosis system based on iCORAL-MMD and deep convolution challenge migration learning provided by an embodiment of the present invention;
FIG. 5 is a schematic diagram of a bearing failure diagnosis system based on iCORAL-MMD and deep convolution challenge migration learning provided by an embodiment of the present invention;
FIG. 6 is a domain arbiter training flow diagram provided by an embodiment of the present invention;
FIG. 7 is a schematic diagram of domain arbiter training provided by an embodiment of the present invention;
FIG. 8 is a physical diagram of an experimental device provided by an embodiment of the invention;
fig. 9 is a schematic diagram of experimental results provided in the examples of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the following examples in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
Aiming at the problems in the prior art, the invention provides a method, a system, equipment and a terminal for optimizing data synchronization among serial chips, and the invention is described in detail below with reference to the accompanying drawings.
Referring to fig. 1, a flow chart of a fault diagnosis method based on iCORAL-MMD and deep convolution challenge migration learning according to an embodiment of the present invention is shown, where the method includes:
s101, collecting vibration signals of a bearing under a A, B working condition;
s102, taking a sample under the working condition A as a source domain sample, collecting a sample under the working condition B as a target domain sample, preprocessing data, performing Fourier transformation to obtain a frequency domain signal, and dividing a training set and a testing set according to a proportion;
s103, an initial fault diagnosis model based on a deep convolutional neural network is established, and the fault diagnosis model is pre-trained by using a source domain signal, wherein the fault diagnosis model comprises a feature extractor and a classifier;
s104, sharing the parameters of the source domain fault diagnosis model to the target domain fault diagnosis model for initialization, establishing an initial domain discriminator, training the domain discriminator through signals obtained by the source domain and the target domain through the feature extractor, and simultaneously updating parameters of the target domain feature extractor and the domain discriminator;
s105, measuring feature distances of the output layer depth features of the feature extractor network by using a correlation alignment metric CORAL and measuring distribution differences by using a maximum mean difference MMD, and updating network parameters of the target domain feature extractor by using the minimized differences;
s106, combining the source domain pre-training classifier to obtain a target domain fault diagnosis model.
The fault diagnosis method based on iCORAL-MMD and deep convolution countermeasure migration learning provided by the embodiment of the invention has the following working principle:
1) Data preprocessing and feature extraction
Firstly, collecting vibration signals of the bearing under the A, B working condition, and preprocessing the signals, including denoising, filtering, downsampling and the like. Then, fourier transform is performed on the preprocessed signal to obtain a frequency domain signal. And then, dividing the data into a training set and a testing set in proportion, taking a sample under the working condition A as a source domain sample, and taking a sample under the working condition B as a target domain sample.
2) Fault diagnosis model establishment based on deep convolutional neural network
A fault diagnosis model based on a deep convolutional neural network is established, and the fault diagnosis model comprises a feature extractor and a classifier. The feature extractor serves to extract features of the input signal and the classifier serves to map the features to corresponding bearing fault state categories. And pre-training the fault diagnosis model by using the source domain signal so as to improve the accuracy of the initial model.
3) Migration learning model establishment
And sharing the parameters of the source domain fault diagnosis model to the target domain fault diagnosis model for initialization, then establishing a domain discriminator, training the domain discriminator through signals obtained by the source domain and the target domain through the feature extractor, and simultaneously updating parameters of the target domain feature extractor and the domain discriminator. The domain discriminator is used for judging which domain the current input signal belongs to by comparing the characteristic distribution difference of the source domain and the target domain.
4) Correlation alignment metric CORAL and maximum mean difference MMD
Feature distribution differences are measured on the output layer depth features of the feature extractor network using a correlation alignment metric CORAL and a maximum mean difference MMD, and the target domain feature extractor network parameters are updated with minimal differences. CORAL and MMD are efficient domain adaptation algorithms, the former can achieve feature alignment by measuring the difference between the feature covariance matrices of the source domain and the target domain, and the latter can achieve feature adaptation by measuring the feature distribution differences.
5) Fault diagnosis model establishment and optimization
Combining the source domain pre-training classifier to obtain a target domain fault diagnosis model, and continuously updating parameters of the domain discriminator and the feature extractor through an iterative optimization process to finally obtain a high-accuracy fault diagnosis model. The model can monitor and diagnose bearing faults in real time, is beneficial to timely finding and processing potential problems, and reduces equipment fault rate and maintenance cost.
In conclusion, the fault diagnosis method based on iCORAL-MMD and deep convolution anti-migration learning migrates knowledge of a source domain to a target domain through the migration learning thought, enhances generalization capability and accuracy of a model, is suitable for bearing fault diagnosis scenes under different working conditions, and has good expansibility and instantaneity.
According to the fault diagnosis method based on iCORAL-MMD and depth convolution anti-migration learning, the time domain signals are converted into the frequency domain signals based on the Fourier transform principle, and compared with the one-dimensional time sequence signals, the target fault diagnosis model can learn the time dimension information better, so that the depth features can be extracted from the target fault diagnosis model, and the reliability of the fault diagnosis result can be improved. Further, compared with a common deep learning method, the method adopts the initial source domain feature extraction model established by the 1DCNN principle to initialize the target domain feature extraction model. The iCORAL-MMD and the domain discrimination model are adopted, so that the feature extraction model of the target domain can be optimized, the feature distribution difference between the source domain and the target domain is reduced, the target domain feature extractor is combined with the classifier trained by the source domain to form the target domain fault diagnosis model, and the fault diagnosis accuracy under different working conditions can be improved.
In the embodiment of the present invention, as shown in fig. 2 and 3, S102 includes:
s201, the signals under the working condition A, B are source domain samples and target domain samples respectively;
specifically:
the source domain data collected under the working condition A is expressed as:
wherein ,is a source domain sample, +.>Is a corresponding fault type label, n s Representing the number of source domain samples.
The collected data as the target domain under the working condition B is expressed as:
wherein , and nt Representing the target domain samples, the corresponding fault type labels and the target sample numbers, respectively.
Due to different working conditions, the source domain D s And target domain D t The data distribution is different.
S202, normalizing the time domain signal to generate a normalized signal;
specifically:
in the formula ,xi For the normalized signal at the i-th instant,the time sequence signal at the i-th moment in the one-dimensional time sequence signal is max (X) which is the maximum value in the one-dimensional time sequence signal, and min (X) which is the minimum value in the one-dimensional time sequence signal.
S203, carrying out Fourier transform on the normalized signal to generate a frequency domain signal;
specifically:
wherein the vibration signal x n Obtaining X through FFT k ，X k Is in the form of a plurality.
In the embodiment of the present invention, as shown in fig. 4 and 5, S103 includes:
s401, dividing a source domain signal into a training set and a testing set according to a proportion;
s402, training an initial 1DCNN neural network model through a training set to obtain a source domain fault diagnosis model, and specifically:
the source domain fault diagnosis model comprises a feature extractor and a classifier, wherein the feature extractor comprises an input layer, at least one convolution layer, at least one pooling layer, at least one full-connection layer and an output layer by adopting 1D-CNN; the classifier adopts a full-connection output layer to use a Softmax function, and specifically comprises the following steps:
the convolution layer comprises a plurality of convolution kernels of 3×3, the convolution step length is 1, the activation function of the convolution layer is a LeakyRelu function, and the activation function is:
wherein f (x) is a Leakyrlu function; x is the input vector of the upper layer.
The pooling mode of the pooling layer is maximum pooling, the pooling step length is 2, and the window size of the pooling layer is 2 multiplied by 2;
the activation function of the full connection layer is a LeakyRelu function;
the activation function of the full-connection output layer of the classifier is a Softmax function:
in the formula ,zi An output value of the i-th node; z j A total output value for a single node; n is the number of output nodes; e is a natural constant.
The fault diagnosis model takes cross entropy as a loss function, and comprises the following steps:
wherein ,the j-th element of the network output vector representing the last layer in the classifier takes as input the i-th source domain sample. />A corresponding transmission component status label is indicated, and N is the number of the associated component status.
The source domain fault diagnosis model pre-training is carried out by minimizing the back propagation of a loss function, and random gradient descent updating network parameters in a feature extractor and a classifier:
wherein ,θS The network model representing source domain fault diagnosis involves parameters, delta represents learning rate.
S403, testing the source domain fault diagnosis model through the test set.
In the embodiment of the present invention, as shown in fig. 6 and 7, S104 and S105 include:
s601, copying parameters of a source domain fault diagnosis model to a target domain fault diagnosis model for initialization, and establishing an initial domain discriminator;
the domain discrimination model adopts a full connection layer, and takes softmax as an activation function output result as a source domain or a target domain. The loss is as follows:
wherein , and />The first and second elements of the output vector representing the last layer in the domain discriminator, respectively, take as input the i-th source domain sample. /> and />Is the corresponding element of the i-th target field sample. Note that it is assumed that the first output element represents the source domain and the second represents the target domain. The random gradient descent of the parameter is updated as follows:
wherein ,θD The representation domain discriminant model involves parameters, δ representing the learning rate.
In the domain discrimination, the smaller and better the updating parameters are, the better the classifying effect is indicated, and the parameter updating mode is counter-propagation; however, the domain arbiter does not differentiate for the feature extractor, i.e., the larger the domain loss, the better, and therefore, the domain is divided by-L in the feature extractor D The smaller the better the parameter update.
S602, an iCORAL-MMD domain self-adaption layer is established at a source domain and target domain feature extractor output layer;
the measuring of the difference of the output characteristic distribution of the target domain characteristic extractor network, and the updating of the target domain characteristic extractor network by the minimized difference is specifically as follows:
wherein λ1 and λ2 CORAL loss and MMD loss, respectivelyIs used for the coefficient of (a),is the Frobenius norm of the matrix, n s Is the number of training samples from the source domain, n t Is the number of training samples from the target domain, FS i Is output data of a source domain through a full connection layer FC, FT j Is the output data of the target domain through the full connection layer FC H Is a regenerated kernel hilbert space,representing covariance matrix>Representing a center matrix I n Is an n-dimensional vector with all elements being 1. CORAL is an efficient and simple unsupervised adaptive method widely used to measure differences in source and target domains in model identification, for example by exploring second order statistics of source and target domains to align their input feature distributions. Thus, the only calculation it requires is to calculate covariance statistics in each domain.
Introducing domain adaptive learning modules into FC1 and FC2 of the fully connected layer, calculating covariance distances of features at source domain and target domain FC1 and FC2 and measuring distribution differences by using MMD, which is defined as iCORAL-MMD loss and is expressed as L G . The calculation formula is as follows:
is the Frobenius norm of the matrix, C S and CT Is the covariance matrix of the source domain and the target domain. Their calculation formula is as follows:
where 1 is the column vector with element 1, F S Is output data of source domain through full connection layer FC, F T Is output data of target domain through full connection layer FC, n S and nT The number of samples for the source domain and the target domain, respectively. Their gradients were calculated as follows:
the feature extractor network parameters are updated as:
wherein ,θD Representing parameters involved in the target domain feature extractor, and delta represents the learning rate.
S603, training the model through target domain data to obtain a trained target domain feature extractor and domain discriminator;
the total loss function adopted by the target domain fault diagnosis model of the anti-migration learning network based on iCORAL-MMD and the depth convolution is as follows:
L T ＝L G -L D
related parameters:
s604, combining the target domain feature extractor with the source domain classifier to obtain a fault diagnosis model of the target domain.
Based on the above process, the second half of the iCORAL-MMD-DCAT transfer learning process aims at minimizing the iCORAL-MMD-DCAT domain self-adaptation and domain countermeasure overall loss function, thereby not only achieving the domain countermeasure purpose, completing the feature transfer, but also ensuring the high precision of the classifier on the classification of the target domain test sample.
The embodiment of the invention has a great advantage in the research and development or use process, and has the following description in combination with data, charts and the like of the test process.
1. The experimental device comprises:
the test uses the bearing data collected by a rolling bearing (rolling bearing is a typical rotary machine) fault simulation test bed of the electric engineering laboratory of Kassi university of America. The test stand, as shown in fig. 8, consisted of a 2 horsepower motor (left), torque sensor/encoder (center), load cell (right) and control electronics. The rolling bearing at the driving end to be detected (model SKF6205-2 RS) supports the rotating shaft of the motor. The laboratory respectively processes pits with the diameter of 0.1778mm on the outer ring, the rolling body and the inner ring of the 3 rolling bearings in an electric spark machining mode to simulate single-point cracks of the outer ring, the rolling body and the inner ring of the rolling bearings. An acceleration sensor is arranged on a bearing seat of the driving end, vibration acceleration signals of the rolling bearing caused by faults under different working conditions (namely different rotating speeds and different loads) monitored by the acceleration sensor are collected through a signal collector, and the sampling frequency is 12kHz. Setting the rotating speed of 1772r/min and the working condition A under the condition of 1 horsepower load; and collecting fault samples of the outer ring, the rolling body and the inner ring under each working condition under the working condition B under the conditions of the rotating speed 1750r/min and the 2 horsepower load.
Each sample is subjected to segmentation preprocessing to obtain a corresponding 1×960 matrix as an input sample of the iCORAL-MMD-DCAT.
icoral-MMD-DCAT network architecture design and parameter settings:
(1) The depth feature extractor is designed as follows, the first layer is a 1 st convolution layer, the second layer is a 1 st pooling layer, the third layer is a 2 nd convolution layer, the fourth layer is a 2 nd pooling layer, the fifth layer is a 3 rd convolution layer, the sixth layer is a 3 rd pooling layer, and the seventh layer is a flattened layer flat layer to output depth features:
(2) The classifier is designed with the structure shown in table 2, wherein the first layer is a fully connected layer connected with the deep convolution feature extractor, the activation function of the layer is a Leakyrlu function, the second layer is an output layer of the classifier, and the activation function of the layer is softmax.
(3) The structure of the domain classifier is shown in table 3, the first layer is a fully connected layer, the activation function of the layer is a Leakyrlu function, the layer is connected with the fully connected layer of the depth convolution feature extractor through a gradient inversion layer, the second layer is an output layer of the domain classifier, and the activation function of the layer is sigmoid.
3. And taking the sample under the working condition A as a sample with a label in a source domain, and taking the sample under the working condition B as a sample without a label in a target domain (namely, a current sample to be tested) to perform an experiment. The experimental result is shown in fig. 9, the migration model obtains more than 95% accuracy on the source domain and the target domain, and the target domain and the source domain characteristics can be obtained from the characteristic visual diagram, so that better self-adaption is realized.
The model effect of the invention is as follows:
according to the fault diagnosis method based on the depth convolution domain anti-migration learning, in the iCORAL-MMD-DCAT, the gradient problems such as gradient disappearance, gradient divergence and the like in the iCORAL-MMD-DCAT training process can be avoided by utilizing the depth convolution feature extractor to perform high-layer feature extraction, and the nonlinear fitting capacity of the iCORAL-MMD-DCAT can be improved; performing domain countermeasure training between a source domain and a target domain by utilizing gradient inversion in a domain discriminator, wherein the domain countermeasure training is performed by using the characteristic distribution of a source domain labeled sample and a target domain unlabeled sample, so that the domain adaptability of the source domain labeled sample to the target domain can be enhanced, and meanwhile, the difference between the target domain and the depth characteristic distribution of the source domain is reduced by measuring the depth characteristic distribution of the target domain by iCORAL-MMD, so that the domain invariant characteristics of the source domain and the target domain are obtained through learning, and the migration performance of the iCORAL-MMD-DCAT is improved; feature migration and classification based on minimizing the domain adaptation overall loss function of iCORAL-MMD-DCAT may improve diagnostic accuracy after migration. The advantages of the iCORAL-MMD-DCAT enable the fault diagnosis method based on the iCORAL-MMD-DCAT to carry out high-precision fault diagnosis on the current sample to be tested of the rotary machine by utilizing the sample with the label of the source domain under the condition that the sample with the label of the target domain under the current working condition of the bearing does not exist.
The invention discloses a bearing fault diagnosis method based on iCORAL-MMD (improved CORrelation Alignment joint Maximum Mean discrepancy) and deep convolution anti-migration learning. According to the method, the source domain data and the target domain data are subjected to combined training, so that bearing faults under different working conditions are effectively diagnosed. Specific examples are as follows:
example 1-automobile Engine bearing failure diagnosis
And (3) data acquisition: vibration signals of the automobile engine bearing are collected under different working conditions (such as different rotating speeds and loads).
Data preprocessing: and carrying out pretreatment such as denoising, filtering, normalization and the like on the acquired signals.
Signal analysis and feature extraction: and converting the preprocessed time domain signal into a frequency domain signal by adopting Fast Fourier Transform (FFT), and extracting the characteristics related to bearing faults.
Model building and source domain pre-training: a1 DCNN model is built, and training data of a source domain (such as bearing vibration signals under normal working conditions) is used for pre-training the model.
Domain arbiter training: and sharing parameters of the source domain pre-training model to the target domain model for initialization, establishing an initial domain discriminator, and training the discriminator through the source domain characteristics and the target domain characteristics.
iCORAL-MMD training and model migration: and (3) performing joint training on the source domain pre-training model and the target domain data by using an iCORAL-MMD method to realize migration of the model.
Building a target domain fault diagnosis model: and combining the source domain classifier with the target domain feature extractor to obtain a fault diagnosis model of the target domain, and performing fine adjustment on the model by using training data of the target domain.
Model evaluation and application: and evaluating the model by using test data of a target domain, calculating indexes such as diagnosis accuracy, recall rate and the like, and realizing real-time detection and diagnosis of the bearing faults of the automobile engine.
Example 2-wind turbine bearing failure diagnosis
And (3) data acquisition: and collecting vibration signals of the bearing of the wind generating set under different wind speed conditions.
Data preprocessing: and carrying out pretreatment such as denoising, filtering, normalization and the like on the acquired signals.
Signal analysis and feature extraction: and converting the preprocessed time domain signal into a frequency domain signal by adopting Fast Fourier Transform (FFT), and extracting the characteristics related to bearing faults.
Model building and source domain pre-training: a 1DCNN model is built and the model is pre-trained using training data of a source domain (e.g., bearing vibration signals at normal wind speeds).
Domain arbiter training: and sharing parameters of the source domain pre-training model to the target domain model for initialization, establishing an initial domain discriminator, and training the discriminator through the source domain characteristics and the target domain characteristics.
iCORAL-MMD training and model migration: and (3) performing joint training on the source domain pre-training model and the target domain data by using an iCORAL-MMD method to realize migration of the model.
Building a target domain fault diagnosis model: and combining the source domain classifier with the target domain feature extractor to obtain a fault diagnosis model of the target domain, and performing fine adjustment on the model by using training data of the target domain.
Model evaluation and application: and evaluating the model by using test data of a target domain, calculating indexes such as diagnosis accuracy, recall rate and the like, and realizing real-time detection and diagnosis of bearing faults of the wind generating set.
Example 3 train bearing failure diagnosis
And (3) data acquisition: vibration signals of the train bearings are collected under different speed and load conditions.
Data preprocessing: and carrying out pretreatment such as denoising, filtering, normalization and the like on the acquired signals.
Signal analysis and feature extraction: and converting the preprocessed time domain signal into a frequency domain signal by adopting Fast Fourier Transform (FFT), and extracting the characteristics related to bearing faults.
Model building and source domain pre-training: a 1DCNN model is built and pre-trained using training data from a source domain (e.g., bearing vibration signals at normal speed and load).
Domain arbiter training: and sharing parameters of the source domain pre-training model to the target domain model for initialization, establishing an initial domain discriminator, and training the discriminator through the source domain characteristics and the target domain characteristics.
iCORAL-MMD training and model migration: and (3) performing joint training on the source domain pre-training model and the target domain data by using an iCORAL-MMD method to realize migration of the model.
Building a target domain fault diagnosis model: and combining the source domain classifier with the target domain feature extractor to obtain a fault diagnosis model of the target domain, and performing fine adjustment on the model by using training data of the target domain.
Model evaluation and application: and evaluating the model by using test data of a target domain, calculating indexes such as diagnosis accuracy, recall rate and the like, and realizing real-time detection and diagnosis of the train bearing faults.
It should be noted that the embodiments of the present invention can be realized in hardware, software, or a combination of software and hardware. The hardware portion may be implemented using dedicated logic; the software portions may be stored in a memory and executed by a suitable instruction execution system, such as a microprocessor or special purpose design hardware. Those of ordinary skill in the art will appreciate that the apparatus and methods described above may be implemented using computer executable instructions and/or embodied in processor control code, such as provided on a carrier medium such as a magnetic disk, CD or DVD-ROM, a programmable memory such as read only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The device of the present invention and its modules may be implemented by hardware circuitry, such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., as well as software executed by various types of processors, or by a combination of the above hardware circuitry and software, such as firmware.
The foregoing is merely illustrative of specific embodiments of the present invention, and the scope of the invention is not limited thereto, but any modifications, equivalents, improvements and alternatives falling within the spirit and principles of the present invention will be apparent to those skilled in the art within the scope of the present invention.
Claims (10)
1. The bearing fault diagnosis method based on iCORAL-MMD and deep convolution anti-migration learning is characterized by comprising the following steps:
firstly, preprocessing collected bearing vibration data and converting frequency domain signals; secondly, building a basic diagnosis model and pre-training the model by using a large amount of labeled data; and then, using the obtained pre-training model, combining with the iCORAL-MMD and the countermeasure method, and migrating the model to the monitoring data to realize fault diagnosis.
2. The bearing fault diagnosis method based on iCORAL-MMD and deep convolution anti-migration learning according to claim 1, comprising the following steps:
step one, collecting vibration signals of a bearing under the working condition A, B;
step two, the signals under the working condition A, B are respectively used as sample data of a source domain and a target domain, after preprocessing, the time domain signals are converted into frequency domain signals based on fast Fourier transform, and a training set and a testing set are proportionally divided;
step three, an initial source domain 1DCNN fault diagnosis model is established, training is carried out through source domain data, and a source domain pre-training model is obtained;
step four, sharing the parameters of the source domain fault diagnosis model to the target domain fault diagnosis model for initialization, establishing an initial domain discriminator, and training the domain discriminator through characteristic signals;
training the initial fault diagnosis model of the target domain by using iCORAL-MMD to obtain a domain discriminator and a target domain feature extractor;
and step six, combining the source domain classifier with the target domain feature extractor to obtain a fault diagnosis model of the target domain.
3. The bearing fault diagnosis method based on iCORAL-MMD and deep convolution anti-migration learning according to claim 2, wherein the step two comprises:
working condition A data is used as a source domain signalIncluding N health states;
working condition B data is used as a target domain signalIncluding N health states;
normalizing the vibration signal to generate a normalized signal;
performing Fourier transform on the normalized signal to generate a frequency domain signal;
the normalization processing of the one-dimensional time sequence signal comprises the following steps:
in the formula ,xi For the normalized signal at the i-th instant,the time sequence signal at the i-th moment in the one-dimensional time sequence signal is max (X) which is the maximum value in the one-dimensional time sequence signal, and min (X) which is the minimum value in the one-dimensional time sequence signal.
4. The bearing fault diagnosis method based on iCORAL-MMD and deep convolution anti-migration learning according to claim 2, wherein the fast fourier transform is performed on the source domain sample and the target domain sample to obtain a frequency domain signal, specifically:
wherein the vibration signal x n Obtaining X through FFT k ，X k Is in the form of a plurality.
5. The bearing fault diagnosis method based on iCORAL-MMD and deep convolution anti-migration learning according to claim 2, wherein the fault diagnosis model uses cross entropy as a loss function as follows:
wherein ,the j-th element of the network output vector representing the last layer in the classifier takes the i-th source domain sample as input; />Representing a phaseThe corresponding drive member status label, and N is the number of the associated member status.
6. The bearing fault diagnosis method based on iCORAL-MMD and deep convolution anti-migration learning according to claim 2, wherein the source domain fault diagnosis model pre-training is specifically:
wherein ,θS The network model representing source domain fault diagnosis involves parameters, delta represents learning rate.
7. The bearing fault diagnosis method based on iCORAL-MMD and deep convolution anti-migration learning according to claim 2, wherein the domain discrimination model and training thereof are specifically as follows:
the domain discrimination model adopts a full-connection layer, takes Softmax as an activation function output result as a source domain or a target domain, and the loss is as follows:
wherein , and />2 represents the first and second elements of the output vector of the last layer in the domain discriminator, respectively, taking as input the i-th source domain sample,/> and />Is the ithCorresponding elements of the target domain sample; assuming that the first output element represents the source domain and the second represents the target domain, the parametric stochastic gradient descent update is as follows:
wherein ,θD The representation domain discriminant model involves parameters, δ representing the learning rate.
8. The bearing fault diagnosis method based on iCORAL-MMD and deep convolution anti-migration learning according to claim 2, wherein the target domain feature extractor network output feature distribution difference is measured, and the updating of the target domain feature extractor network by the minimized difference is specifically:
introducing a domain self-adaptive learning module into FC1 and FC2 of the full-connection layer, calculating covariance distance and maximum mean difference of features at FC1 and FC2 of a source domain and a target domain, and defining the covariance distance and the maximum mean difference as iCORAL-MMD loss; CORAL is an effective and simple unsupervised adaptive method widely used to measure differences between source and target domains in model identification, for example by exploring second order statistics of source and target domains to align their input feature distributions; thus, it is the covariance statistics in each domain that need to be calculated; compared to MMD, the MMD-based approach typically applies the same transformations to the source and target domains, its asymmetric transformations are more flexible, and typically yields better performance for domain adaptation tasks;
the calculation formula is as follows:
λ 1 and λ2 The coefficients of CORAL loss and MMD loss,is the Frobenius norm of the matrix, n s Is the number of training samples from the source domain, n t Is the number of training samples from the target domain, FS i Is output data of a source domain through a full connection layer FC, FT j Is the output data of the target domain through the full connection layer FC H Is a regenerated core Hilbert space, < >>Is the Frobenius norm of the matrix, C S and CT Is the covariance matrix of the source domain and the target domain, and the calculation formula is as follows:
where 1 is the column vector with element 1, F S Is output data of source domain through full connection layer FC, F T Is output data of target domain through full connection layer FC, n S and nT The number of samples for the source domain and the target domain, respectively, the gradient is calculated as follows:
the feature extractor network parameters are updated as:
wherein ,θD Representing parameters involved in the target domain feature extractor, and delta represents the learning rate.
9. The bearing fault diagnosis method based on iCORAL-MMD and deep convolution anti-migration learning of claim 2, wherein the total loss function adopted by the target domain fault diagnosis model of the network based on iCORAL-MMD and deep convolution anti-migration learning is as follows:
L T ＝L G -L D ；
related parameters:
10. bearing fault diagnosis system based on iCORAL-MMD and deep convolution challenge migration learning, which comprises an input layer, at least one convolution layer, at least one pooling layer, at least one fully connected layer and an output layer, applying the bearing fault diagnosis method based on iCORAL-MMD and deep convolution challenge migration learning according to any one of claims 1-9;
the convolution layer comprises a plurality of convolution kernels of 3 multiplied by 3, the convolution step length is 1, and the activation function of the convolution layer is a Leakyrlu function;
the pooling mode of the pooling layer is maximum pooling, the pooling step length is 2, and the window size of the pooling layer is 2 multiplied by 2;
the activation function of the full connection layer is a Leakyrelu function, which is:
wherein f (x) is a LeakyRelu function; x is the input vector of the upper layer;
the activation function of the full-connection output layer of the classifier is a Softmax function, and the Softmax function is as follows:
in the formula ,zi An output value of the i-th node; z j A total output value for a single node; n is the number of output nodes; e is a natural constant.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310839080.9A CN116793682A (en) | 2023-07-07 | 2023-07-07 | Bearing fault diagnosis method based on iCORAL-MMD and anti-migration learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310839080.9A CN116793682A (en) | 2023-07-07 | 2023-07-07 | Bearing fault diagnosis method based on iCORAL-MMD and anti-migration learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116793682A true CN116793682A (en) | 2023-09-22 |
Family
ID=88037448
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310839080.9A Pending CN116793682A (en) | 2023-07-07 | 2023-07-07 | Bearing fault diagnosis method based on iCORAL-MMD and anti-migration learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116793682A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117330315A (en) * | 2023-12-01 | 2024-01-02 | 智能制造龙城实验室 | Rotary machine fault monitoring method based on online migration learning |
| CN117349749A (en) * | 2023-10-09 | 2024-01-05 | 石家庄铁道大学 | Multi-source domain bearing fault diagnosis method based on mixed convolution |
| CN117407698A (en) * | 2023-12-14 | 2024-01-16 | 青岛明思为科技有限公司 | Hybrid distance guiding field self-adaptive fault diagnosis method |
-
2023
- 2023-07-07 CN CN202310839080.9A patent/CN116793682A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117349749A (en) * | 2023-10-09 | 2024-01-05 | 石家庄铁道大学 | Multi-source domain bearing fault diagnosis method based on mixed convolution |
| CN117349749B (en) * | 2023-10-09 | 2024-03-15 | 石家庄铁道大学 | Multi-source domain bearing fault diagnosis method based on mixed convolution |
| CN117330315A (en) * | 2023-12-01 | 2024-01-02 | 智能制造龙城实验室 | Rotary machine fault monitoring method based on online migration learning |
| CN117330315B (en) * | 2023-12-01 | 2024-02-23 | 智能制造龙城实验室 | Rotary machine fault monitoring method based on online migration learning |
| CN117407698A (en) * | 2023-12-14 | 2024-01-16 | 青岛明思为科技有限公司 | Hybrid distance guiding field self-adaptive fault diagnosis method |
| CN117407698B (en) * | 2023-12-14 | 2024-03-08 | 青岛明思为科技有限公司 | Hybrid distance guiding field self-adaptive fault diagnosis method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111709448B (en) | Mechanical fault diagnosis method based on migration relation network | |
| CN111651937B (en) | Method for diagnosing faults of in-class self-adaptive bearing under variable working conditions | |
| Hu et al. | Data-driven fault diagnosis method based on compressed sensing and improved multiscale network | |
| CN110751207B (en) | Fault diagnosis method for anti-migration learning based on deep convolution domain | |
| CN110059601B (en) | Intelligent fault diagnosis method for multi-feature extraction and fusion | |
| CN110849626B (en) | Self-adaptive sparse compression self-coding rolling bearing fault diagnosis system | |
| CN116793682A (en) | Bearing fault diagnosis method based on iCORAL-MMD and anti-migration learning | |
| CN111914883B (en) | Spindle bearing state evaluation method and device based on deep fusion network | |
| CN111562108A (en) | Rolling bearing intelligent fault diagnosis method based on CNN and FCMC | |
| CN113008559B (en) | Bearing fault diagnosis method and system based on sparse self-encoder and Softmax | |
| CN112113755B (en) | Mechanical fault intelligent diagnosis method based on deep convolution-kurtosis neural network | |
| CN112633339A (en) | Bearing fault intelligent diagnosis method, bearing fault intelligent diagnosis system, computer equipment and medium | |
| CN113188794B (en) | Gearbox fault diagnosis method and device based on improved PSO-BP neural network | |
| Iqbal et al. | CNC machine-bearing fault detection based on convolutional neural network using vibration and acoustic signal | |
| CN113139278A (en) | Deep learning-based bearing residual life prediction method and system | |
| CN112926728B (en) | Small sample turn-to-turn short circuit fault diagnosis method for permanent magnet synchronous motor | |
| CN113792758A (en) | Rolling bearing fault diagnosis method based on self-supervision learning and clustering | |
| CN114755017B (en) | Variable-speed bearing fault diagnosis method of cross-domain data driving unsupervised field shared network | |
| CN111076934A (en) | Method for diagnosing potential fault of bearing based on S transformation | |
| CN112393934A (en) | Wind turbine generator fault diagnosis method based on sparse self-coding and extreme learning machine | |
| CN116186641A (en) | Cross-equipment rolling bearing small sample fault diagnosis method based on feature fusion | |
| CN114241271B (en) | Multi-twin migration learning fusion multi-information mechanical fault intelligent diagnosis method | |
| Shaowu et al. | Study on the health condition monitoring method of hydraulic pump based on convolutional neural network | |
| CN114626625A (en) | Bearing life prediction method based on multi-module U-BiLSTM network | |
| Liu et al. | Rolling bearing fault diagnosis of train running gear based on optimized deep residual 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 |