CN112990259A - Early fault diagnosis method of rotary mechanical bearing based on improved transfer learning - Google Patents
Early fault diagnosis method of rotary mechanical bearing based on improved transfer learning Download PDFInfo
- Publication number
- CN112990259A CN112990259A CN202110153282.9A CN202110153282A CN112990259A CN 112990259 A CN112990259 A CN 112990259A CN 202110153282 A CN202110153282 A CN 202110153282A CN 112990259 A CN112990259 A CN 112990259A
- Authority
- CN
- China
- Prior art keywords
- aligned
- data set
- diagnosed
- data
- eigenvalue
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000003745 diagnosis Methods 0.000 title claims abstract description 32
- 238000013526 transfer learning Methods 0.000 title claims abstract description 12
- 239000013598 vector Substances 0.000 claims abstract description 171
- 239000011159 matrix material Substances 0.000 claims abstract description 72
- 238000000605 extraction Methods 0.000 claims abstract description 25
- 238000009826 distribution Methods 0.000 claims description 21
- 238000000354 decomposition reaction Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000001228 spectrum Methods 0.000 claims description 4
- 235000014653 Carica parviflora Nutrition 0.000 claims 1
- 241000243321 Cnidaria Species 0.000 claims 1
- 230000006870 function Effects 0.000 description 11
- 239000008186 active pharmaceutical agent Substances 0.000 description 9
- 238000013508 migration Methods 0.000 description 8
- 230000005012 migration Effects 0.000 description 8
- 238000012360 testing method Methods 0.000 description 7
- 238000013135 deep learning Methods 0.000 description 6
- 238000010801 machine learning Methods 0.000 description 6
- 238000012549 training Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000000513 principal component analysis Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 239000000306 component Substances 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000003064 k means clustering Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 239000000047 product Substances 0.000 description 2
- 239000004576 sand Substances 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component 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
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06N3/088—Non-supervised learning, e.g. competitive learning
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Probability & Statistics with Applications (AREA)
- Acoustics & Sound (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The application relates to an early fault diagnosis method of a rotary mechanical bearing based on improved transfer learning, which belongs to the technical field of computers and comprises the following steps: performing data alignment on the labeled data set and the data set to be diagnosed and then performing feature extraction; clustering the eigenvalue vector matrix corresponding to the aligned labeled data set; carrying out eigenvalue vector alignment on an eigenvalue vector matrix corresponding to the label of each fault type and an eigenvalue vector matrix corresponding to the aligned data set to be diagnosed; for each eigenvalue vector to be diagnosed in the second aligned eigenvalue vectors, determining n nearest eigenvalue vectors from the first aligned eigenvalue vector matrix; determining a label of a fault type corresponding to the eigenvalue vector to be diagnosed according to the labels of the fault types to which the n eigenvalue vectors belong; and the non-supervision mode is adopted to realize the classification of the target domain non-label early-stage rotating mechanical bearing fault data, and the classification accuracy is improved.
Description
[ technical field ] A method for producing a semiconductor device
The application relates to an early fault diagnosis method of a rotary mechanical bearing based on improved transfer learning, and belongs to the technical field of computers.
[ background of the invention ]
Bearings are important components in rotating mechanical devices. Bearings in mechanical devices generally assume the function of transmitting power, and are therefore the core components of rotating machines. Typically, during operation of the mechanical apparatus, the bearings rotate at speeds of up to several thousand revolutions per minute, and even up to tens of thousands of revolutions per minute. Moreover, the rotating machine is usually operated in a high-temperature, high-corrosion and long-time operating environment, and the bearing is prone to failure under such a severe operating environment. When the bearing failure occurs early, the rotary machine can continue to operate, and it is therefore difficult to find that the bearing has failed. However, if the bearing is in failure for a long time, the bearing is easy to damage the equipment, and the service life of the equipment is influenced. Therefore, early failure diagnosis of the bearing is required.
A typical early fault diagnosis method for a bearing is to perform early fault diagnosis on the bearing by using a machine learning method or a deep learning method. Based on the above manner, the training data is required to be used to train the neural network in advance, and the test data is used to test the trained neural network. The training data and the testing data comprise bearing related working data and fault labels corresponding to the working data, and the fault labels are used for indicating fault types corresponding to the working data.
However, during training and testing, there are often problems as follows: one, the data for the tag with the type of failure is very small and difficult to collect during production. The tagged data sets currently in use are typically experimental data collected by a laboratory. Secondly, the labeled data sets and the data to be diagnosed have different data distributions. When fault diagnosis is carried out by the traditional machine learning and deep learning methods, one part of a fault data set is used as training data, and the other part of the fault data set is used as test data, namely the training data and the test data have the same data distribution. This can result in classifiers trained on labeled datasets using traditional machine learning and deep learning methods that can achieve good classification results on the same distributed test set but not good classification and fault diagnosis results on different distributed datasets or datasets requiring diagnosis.
[ summary of the invention ]
The application provides an early fault diagnosis method of a rotary mechanical bearing based on improved transfer learning, which can solve the problems that the existing machine learning method or deep learning method is used for early fault diagnosis of the bearing, the amount of labeled data in a data set is small, and the data distribution is different from the actual data distribution to be diagnosed. The application provides the following technical scheme: provided is an early failure diagnosis method of a rotary mechanical bearing based on improved transfer learning, the method comprising:
acquiring a labeled data set and a unlabeled data set to be diagnosed, wherein each group of labeled data in the labeled data set comprises bearing related working data and a label of a corresponding fault type; each group of unlabeled data in the data set to be diagnosed comprises bearing related working data;
performing data alignment on the labeled data set and the data set to be diagnosed to obtain an aligned labeled data set and an aligned data set to be diagnosed;
respectively extracting the characteristics of the aligned labeled data sets and the aligned data sets to be diagnosed to obtain characteristic value vector matrixes corresponding to the data sets; the eigenvalue vector matrix corresponding to each data set comprises shallow features and deep features;
clustering the eigenvalue vector matrix corresponding to the aligned labeled data set to obtain an eigenvalue vector matrix corresponding to the label of each fault type;
performing characteristic value vector alignment on the characteristic value vector matrix corresponding to the label of each fault type and the characteristic value vector matrix corresponding to the aligned data set to be diagnosed to obtain a first aligned characteristic value vector corresponding to the aligned labeled data set and a second aligned characteristic value vector corresponding to the aligned data set to be diagnosed;
for each eigenvalue vector to be diagnosed in the second aligned eigenvalue vectors, determining n eigenvalue vectors that are closest to the eigenvalue vector to be diagnosed from the first aligned eigenvalue vector matrix; the n is a positive integer;
and determining the label of the fault type corresponding to the eigenvalue vector to be diagnosed according to the label of the fault type to which the n eigenvalue vectors belong, so as to obtain the fault type of the unlabeled data corresponding to the eigenvalue vector to be diagnosed.
Optionally, the data aligning the labeled data set and the data set to be diagnosed includes:
and performing data alignment on the labeled data set and the data set to be diagnosed based on a correlation alignment distance CORAL algorithm so as to enable the data distribution of the labeled data set to be the same as that of the data set to be diagnosed.
Optionally, the respectively performing feature extraction on the aligned labeled data set and the aligned data set to be diagnosed includes:
respectively carrying out shallow feature extraction on the aligned labeled data sets and the aligned data sets to be diagnosed to obtain shallow features corresponding to each data set;
wherein the shallow features comprise at least one of time-domain features, frequency-domain features, and time-frequency-domain features of the bearing-related operational data;
the time domain features include: at least one of a mean, a variance, a root mean square value, a standard deviation, a peak-to-peak value, a maximum value, and a minimum value of the bearing related operational data;
the frequency domain features include: at least one of a frequency center, a frequency root mean square, a frequency standard deviation and a signal power spectrum of the bearing related working data;
the time-frequency domain features include: and wavelet coefficients are obtained through wavelet packet decomposition calculation.
Optionally, the respectively performing feature extraction on the aligned labeled data set and the aligned data set to be diagnosed includes:
and respectively carrying out deep feature extraction on the aligned labeled data set and the aligned data set to be diagnosed by using a wavelet scattering network to obtain a first deep feature corresponding to each data set.
Optionally, the respectively performing feature extraction on the aligned labeled data set and the aligned data set to be diagnosed includes:
and respectively carrying out deep feature extraction on the aligned labeled data set and the aligned data set to be diagnosed by using a stack type self-encoder to obtain a second deep feature corresponding to each data set.
Optionally, the clustering the eigenvalue vector matrix corresponding to the aligned labeled data set to obtain an eigenvalue vector matrix corresponding to a label of each fault type includes:
and clustering the eigenvalue vector matrix corresponding to the aligned labeled data set based on a K-nearest neighbor classification KNN algorithm by taking the label of the fault type as a central point of clustering respectively to obtain an eigenvalue vector matrix corresponding to the label of each fault type.
Optionally, the performing eigenvalue vector alignment on the eigenvalue vector matrix corresponding to the label of each fault type and the eigenvalue vector matrix corresponding to the aligned dataset to be diagnosed includes:
projecting the characteristic value vectors corresponding to the data sets to a Grassmann manifold space by using a ground wire kernel GFK algorithm;
feature vector alignment is performed on the grassmann manifold space.
Optionally, the determining, according to the label of the fault type to which the n eigenvalue vectors belong, the label of the fault type corresponding to the eigenvalue vector to be diagnosed includes:
and performing label voting on the labels of the fault types to which the n characteristic value vectors belong, and determining a voting result as the label of the fault type corresponding to the characteristic value vector to be diagnosed.
Optionally, the bearing-related operational data comprises vibration data of the bearing.
The beneficial effect of this application lies in: performing data alignment on the labeled data set and the data set to be diagnosed to obtain an aligned labeled data set and an aligned data set to be diagnosed; respectively extracting the characteristics of the aligned labeled data sets and the aligned data sets to be diagnosed to obtain characteristic value vector matrixes corresponding to the data sets; clustering the eigenvalue vector matrix corresponding to the aligned labeled data set to obtain an eigenvalue vector matrix corresponding to the label of each fault type; performing characteristic value vector alignment on the characteristic value vector matrix corresponding to the label of each fault type and the characteristic value vector matrix corresponding to the aligned data set to be diagnosed to obtain a first aligned characteristic value vector corresponding to the aligned labeled data set and a second aligned characteristic value vector corresponding to the aligned data set to be diagnosed; for each eigenvalue vector to be diagnosed in the second aligned eigenvalue vectors, n eigenvalue vectors which are closest to the eigenvalue vector to be diagnosed are determined from the first aligned eigenvalue vector matrix; according to the labels of the fault types to which the n characteristic value vectors belong, determining the label of the fault type corresponding to the characteristic value vector to be diagnosed, and obtaining the fault type of the label-free data corresponding to the characteristic value vector to be diagnosed; the problems that the existing machine learning method or deep learning method is used for early fault diagnosis of the bearing, the amount of labeled data in a data set is small, and the data distribution is different from the actual data distribution to be diagnosed; and (3) classifying the target domain label-free early-stage rotating mechanical bearing fault data by adopting an unsupervised mode only based on a small amount of labeled data which is different from the target domain data in distribution as a source domain.
In addition, a method of combining a shallow layer characteristic value and a depth characteristic value is adopted for extracting the data characteristic value, and the classification accuracy is improved.
In addition, the KNN search tree is used for screening the source domain characteristic value vectors, and the classification accuracy is improved.
In addition, a method of data alignment twice is adopted. First, the original data signal is aligned. Aligning the characteristic value vectors; the classification accuracy can be improved.
The foregoing description is only an overview of the technical solutions of the present application, and in order to make the technical solutions of the present application more clear and clear, and to implement the technical solutions according to the content of the description, the following detailed description is made with reference to the preferred embodiments of the present application and the accompanying drawings.
[ description of the drawings ]
FIG. 1 is a flow chart of a method for early fault diagnosis of a rotating mechanical bearing based on improved transfer learning provided by an embodiment of the present application;
FIG. 2 is a schematic diagram of a feature extraction process provided by one embodiment of the present application;
FIG. 3 is a schematic diagram of an early fault diagnosis process for a rotating mechanical bearing based on improved migration learning provided by an embodiment of the present application;
fig. 4 is a schematic diagram of predicting power consumption of a device based on multi-chain decomposition according to an embodiment of the present application.
[ detailed description ] embodiments
The following detailed description of embodiments of the present application will be described in conjunction with the accompanying drawings and examples. The following examples are intended to illustrate the present application but are not intended to limit the scope of the present application.
Optionally, in the present application, an execution subject of each embodiment is taken as an example of an electronic device with computing capability, the electronic device may be a terminal or a server, the terminal may be a computer, a notebook computer, a tablet computer, and the like, and the embodiment does not limit the type of the terminal and the type of the electronic device.
Fig. 1 is a flowchart of an early failure diagnosis method for a rotating mechanical bearing based on improved transfer learning according to an embodiment of the present application. The method at least comprises the following steps:
step 101, acquiring a labeled data set and an unlabeled data set to be diagnosed.
Each group of labeled data in the labeled data set comprises bearing related working data and a label of a corresponding fault type; each set of unlabeled data in the data set to be diagnosed includes bearing-related working data.
In this embodiment, the bearing related operational data is used to indicate the operational status of the bearing. The bearing-related working data can be data generated by the bearing when the bearing works, such as: the bearing related operational data comprises vibration data of the bearing; alternatively, data may be generated for other components associated with the bearing during operation of the bearing, such as: the vibration data generated by other components under the driving of the bearing, and the data type of the relevant working data of the bearing is not limited in this embodiment.
Optionally, the labeled data set may also be a source domain, and the unlabeled data set to be diagnosed may also be referred to as a target domain, and the name of each data set is not limited in this embodiment.
Optionally, the fault types include, but are not limited to: the inner ring fault and the outer ring fault, in other embodiments, the fault type can also be set as other types of faults according to the diagnosis requirement, and the embodiments are not listed. The labels corresponding to different fault types are different, such as: the inner ring fault is represented by a label "1", and the outer ring fault is represented by a label "2", but of course, the labels may also be represented by other characters, and the embodiment does not limit the arrangement manner of the label corresponding to the fault type.
Such as: referring to table one below, the source domain includes 800 sets of tagged data, and the bearing-related operational data in each set of tagged data includes 500 consecutive vibration signal samples for the corresponding bearing. The target field includes 800 unlabeled data, each comprising 500 consecutive vibration signal sample values for a corresponding bearing.
Table one:
and 102, carrying out data alignment on the labeled data set and the data set to be diagnosed to obtain the aligned labeled data set and the aligned data set to be diagnosed.
Data alignment is used to make the data distribution of the tagged data set the same as the data distribution of the data set to be diagnosed.
In one example, data alignment of a tagged data set and a data set to be diagnosed includes: and performing data Alignment on the labeled data set and the data set to be diagnosed based on a Correlation Alignment distance (CORAL) algorithm so as to enable the data distribution of the labeled data set to be the same as that of the data set to be diagnosed.
The CORAL algorithm is used for decorrelating source domain data, and then reconstructing the distribution of the source domain data according to the distribution of the target domain data. The principle of the CORAL algorithm includes: the variance characterizes the degree of dispersion of the data, and the source domain and the target domain in the ideal state form a distribution in space, which is only related to the class of a given sample and can be described by the variance. After the interference information is added, namely the domain information exists, the transformation is performed on the original ideal distribution, and meanwhile, the variance is correspondingly transformed. The solution variance operation breaks up the distribution, i.e., removes the influence of the domain information. And then re-correlated based on information from the other domain, i.e., its distribution is reconstructed from the other domain.
The CORAL algorithm can be represented by the following formula:
CS＝cov(DS)+eye(size(DS，2))
CT＝cov(DT)+eye(size(DT，2))
DS＝DS*CS -1/2
Ds *＝DS*CT -1/2
wherein D issRepresenting the source domain, DTRepresenting a target domain, Ds *Representing the source domain after alignment, and the size (a, 2) function represents the number of columns of the return matrix a. The eye (N) function represents the unit matrix that generates N × N, and cov represents the function that calculates the covariance.
103, respectively extracting the characteristics of the aligned labeled data sets and the aligned data sets to be diagnosed to obtain characteristic value vector matrixes corresponding to the data sets; the eigenvalue vector matrix corresponding to each data set comprises shallow features and deep features.
In one example, referring to fig. 2, the shallow eigenvalues are empirical eigenvalues and the deep eigenvalues are wavelet scattering network eigenvalues and/or stacked self-encoder network eigenvalues. And the eigenvalue vector matrix corresponding to each data set is an eigenvalue matrix obtained by combining the shallow features and the deep features.
For the extraction of shallow features, the feature extraction is respectively carried out on the aligned labeled data set and the aligned data set to be diagnosed, and the feature extraction comprises the following steps: and respectively carrying out shallow feature extraction on the aligned labeled data sets and the aligned data sets to be diagnosed to obtain the shallow features corresponding to each data set.
Wherein the shallow features include at least one of time domain features, frequency domain features, and time-frequency domain features of the bearing-related operational data.
In one example, the time domain features include: at least one of a mean, a variance, a root mean square value, a standard deviation, a peak-to-peak value, a maximum value, and a minimum value of the bearing related operational data.
The average of the bearing related operational data is calculated by:
wherein the content of the first and second substances,
the average value of the bearing related working data is represented, and n represents the total data amount of the bearing related working data, such as: the total amount of data in table 1 is 500 vibration signal sample values; x is the number ofiIs the ith data in the bearing related working data.
The variance of the bearing related operational data is calculated by:
wherein, XvarRepresenting the variance of the bearing related operational data.
The root mean square value of the bearing related working data is calculated by the following formula:
wherein, XrmsA root mean square value representing bearing related operational data.
The standard deviation of the bearing related operating data is calculated by:
wherein, XstdThe standard deviation of the bearing related operational data is indicated.
The peak-to-peak value of the bearing related operating data is calculated by:
Xppv＝(Xmax-Xmin)
wherein, XppvPeak-to-peak, X, representing bearing related operational datamaxMaximum value, X, representing bearing-related operational dataminRepresents the minimum value of the relevant working data of the bearing.
In other embodiments, the time domain features may also include other features, such as: impulse factor, kurtosis, skewness factor, waveform factor, square root amplitude, peak factor, margin factor, and/or kurtosis factor, etc., and the type of the temporal feature is not limited in this embodiment.
The pulse factor is calculated by:
wherein, CifA pulse factor representing bearing related operational data.
The kurtosis is calculated by:
wherein, CkvIndicating the kurtosis of bearing-related operational data.
Skewness factors are calculated by:
wherein, CsfA skewness factor representing bearing related operational data.
The form factor is calculated by:
Cs＝Xrms/Xmax
Csand the form factor represents relevant work data of the bearing.
The square root amplitude is calculated by:
Xra square root amplitude representing bearing related operational data.
The crest factor is calculated by:
Ccfa crest factor representing bearing related operational data.
The margin coefficient is calculated by:
Clfa margin factor representing bearing related operational data.
The kurtosis factor is calculated by:
Ckfa kurtosis factor representing bearing-related operational data.
The frequency domain features include: at least one of a frequency center, a frequency root mean square, a frequency standard deviation, and a signal power spectrum of the bearing related operational data.
Optionally, the electronic device extracts frequency domain feature values of the bearing related operational data (e.g., vibration signals) using a fast fourier transform.
Wherein the fast fourier transform is an accumulated version of the fourier transform, assuming that the fourier transform of a discrete ordered long sequence x (n) is defined as:
where x (n) can be represented by the form of addition of parity sequences thereof, i.e., x (n) ═ x1(n) +x2(n), therefore, the above formula results in the following conversion steps:
the above formula is an integral form of the fast Fourier transform, where X is1(k)，X2(k) Represents the sequence x1(n)，x2(N) N/2 point discrete Fourier transform.
After the fast Fourier transform is carried out on the original signal, the frequency center value, the frequency root mean square value, the frequency standard deviation value and the signal maximum energy value of the original signal can be calculated, and the four values are extracted to be used as the signal frequency domain characteristic value.
The frequency center is calculated by:
wherein v isiRepresenting the ith line frequency of the frequency domainThe value of the ratio, X (i), represents the harmonic amplitude of the ith line.
The frequency root mean square is calculated by:
the standard deviation of frequency is calculated by:
wherein v iscRepresenting the center value of the frequency. Both the root mean square and standard deviation of the frequency can reflect the degree of fluctuation of the frequency domain signal.
The power spectrum is calculated by:
wherein T is time, FTAnd (ω) represents the signal fourier transform.
The time-frequency domain features include: and wavelet coefficients are obtained through wavelet packet decomposition calculation.
Wherein, calculating wavelet coefficients by wavelet packet decomposition comprises: carrying out multi-Behcet wavelet packet decomposition of N-4 of level 4 on each sample of the source domain and the target domain; and extracting the coefficients of the 0 th to 15 th nodes of the level 4 to obtain the wavelet coefficients.
For deep feature extraction, feature extraction is respectively performed on the aligned labeled data set and the aligned data set to be diagnosed, and the feature extraction comprises the following steps: and/or, respectively carrying out deep feature extraction on the aligned labeled data set and the aligned data set to be diagnosed by using a stacked self-encoder to obtain a second deep feature corresponding to each data set.
Wherein, the wavelet scattering network (invariant scattering convolution network) is a framework for automatically extracting relevant compact features. Each of its layers is also usually composed of three parts, respectively: wavelet convolution, non-linear and averaging operations. The wavelet scattering Network is a Deep Network similar to a Deep Convolutional Neural Network (Deep Convolutional Neural Network). It performs three main tasks that make up the deep network: convolution, non-linearity and pooling. Where the convolution is performed by wavelets, the modulo operator is used as the non-linearization, and the filtering of the wavelet low pass filter is similar to pooling. First, an input signal is averaged using a wavelet low pass filter. This is a layer 0 scattering function and by averaging, high frequency details in the signal are lost. By performing a continuous wavelet transform on the signal to generate a set of scale map coefficients, details lost in the first step can be captured in subsequent layers. The modulus is applied to the scale map coefficients and the output is then filtered using a wavelet low pass filter to generate a set of layer 1 scattering coefficients. The same procedure was repeated to obtain the layer 2 scattering coefficient. That is, the scaling map coefficient output of the previous layer becomes the input for the operation of the next layer. The same modulo operator is then applied and the output filtered using a wavelet low pass function to derive layer 2 scattering coefficients.
And applying wavelet scattering transformation to the source domain and the target domain, wherein the obtained scattering characteristic mean value is the extracted characteristic value.
An autoencoder is a neural network that attempts to replicate its input at its output. The self-encoder consists of an encoder and a decoder. The encoder maps the input into an implicit representation and the decoder attempts an inverse mapping to reconstruct the original input. Thus, the size of its input will be the same as the size of its output. When the number of neurons in the hidden layer is smaller than the size of the input, the self-encoder will learn the feature representation of the input. By using the characteristic, the characteristic value of the input signal learned from the hidden layer of the encoder is extracted as the signal characteristic value of the source domain and the target domain.
The stacked self-encoder is formed by stacking two or more self-encoders. Wherein the feature value learned by the hidden layer of the first self-encoder is used as the input of the second self-encoder. In this embodiment, two self-encoders are used to form a stacked self-encoder, wherein the second self-encoder hides the eigenvalue learned by the neuron in the layer, and extracts the eigenvalue as the sample signal. In other embodiments, the stacked self-encoder may also be three or more self-encoders, and the number of self-encoders in the stacked self-encoder is not limited in this embodiment.
And 104, clustering the eigenvalue vector matrix corresponding to the aligned labeled data set to obtain an eigenvalue vector matrix corresponding to the label of each fault type.
In one example, clustering the eigenvalue vector matrix corresponding to the aligned labeled data set to obtain an eigenvalue vector matrix corresponding to a label of each fault type includes: and based on a K-Nearest Neighbor (KNN) algorithm, clustering the eigenvalue vector matrix corresponding to the aligned labeled data sets by taking the labels of the fault types as the central points of the clustering respectively to obtain the eigenvalue vector matrix corresponding to the label of each fault type. Specifically, the clustering process includes: and 1, establishing a search tree model for the feature value vector matrix corresponding to the aligned labeled data set by using a K-d tree algorithm. And 2, clustering the characteristic value vector matrixes corresponding to the aligned labeled data sets by taking the labels of the fault types as the central points of the clustering respectively based on the KNN algorithm. And 3, taking the central point of each clustered category/group as a search parameter, searching a plurality of nearest (such as 50 eigenvalue vectors) in the search tree model, and forming a new eigenvalue vector matrix by the nearest plurality of eigenvalue vectors corresponding to the central point of each category/group.
Such as: and the electronic equipment performs K-means clustering on the source domain characteristic value matrix, determines a K value according to the fault types existing in the source domain, and assumes that 2 fault types are used, the K value is 2. According to K-means clustering calculation, obtaining central points C1 (corresponding to one fault type) and C2 (corresponding to the other fault type) of corresponding clusters; creating a k-dimensional tree (KD) search model; searching a plurality of (such as 50 or other numerical values, in this embodiment, the selected number of the feature value vectors is not limited) feature value vectors closest to the cluster center point based on the kd tree model by using K neighbor search; and forming a new source domain eigenvalue vector matrix by the new eigenvalue vectors to obtain an eigenvalue vector matrix corresponding to the label of each fault type.
And 105, performing characteristic value vector alignment on the characteristic value vector matrix corresponding to the label of each fault type and the characteristic value vector matrix corresponding to the aligned data set to be diagnosed to obtain a first aligned characteristic value vector corresponding to the aligned labeled data set and a second aligned characteristic value vector corresponding to the aligned data set to be diagnosed.
In one example, the electronic device projects the eigenvalue vectors corresponding to the respective datasets into the grassmann manifold space using a Geodesic Flow Kernel (GFK) algorithm; feature vector alignment is performed on the grassmann manifold space.
When the GFK algorithm is used for aligning the characteristic value vectors, firstly, the optimal subspace dimension needs to be calculated; and then constructing a geodesic line, and finally calculating a geodesic line flow core.
Calculating the optimal subspace dimension includes: for source domain DSAnd a target domain DTPerforming Principal Component Analysis (PCA) to obtain source domain DSCorresponding PSAnd a target domain DTCorresponding PT(ii) a Transforming the source domain and the target domain to respective subspaces; simultaneously combining the source domains DSTarget domain DTAre combined into DS+TAnd performing principal component analysis to obtain PS+T. Respectively calculating the subspace included angle (namely P) of the data set after the source domain and the target domain are combined with the two domainsS，PTAnd P isS+TThe included angle of (a). ) Respectively denoted as alphad，βd(ii) a With D (d) 0.5[ sin alpha ]d+sinβd]Representing the total measure of the two angles. When the two angles are small, the distance between the source domain and the target domain is very close, and when the two angles are maximum, the distance is maximum, namely the source domain is perpendicular to the target domain. Therefore, the value of d is as large as possible, but the situation that two domains are perpendicular to each other cannot be obtained at the same time, so that the situation that the value is more obtained is ensuredA number of subspaces.
The construction of the geodesic line comprises the following steps: p obtained after performing principal component analysis on source domain and target domainSAnd PTBoth vectors are orthogonal. In manifold space, the source domain and the target domain are assumed to be located on two poles, 0 and 1, respectively, after being mapped by a geodesic mapping function phi. I.e., (0) Ps，φ(1)＝PT. For domain definition at [0,1]The point in between t, Φ (t) ∈ G (D, D). G (D, D) is a Grassmann manifold formed by all D vectors in a D-dimensional vector space. At this time, the function value of the geodesic function at t is:
Φ(t)＝PSU1Γ(t)-RSU2∑(t)
wherein R isSIs PSSupplement of (2), satisfy RS T PS＝0,U1，U2Γ, ∑ is obtained by singular value decomposition:
after singular value decomposition, Γ, Σ are both diagonal matrices, and each element of a diagonal matrix is called the principal angle between two matrices, representing the distance between them.
Calculating the geodesic flow kernel includes: for two vectors xi，xjTheir projection on phi (t) is calculated, projected to a vector of infinite dimensions. The inner product of these two infinite dimensional projections defines a geodesic flow kernel:
wherein G is a semi-positive definite matrix.
Wherein the content of the first and second substances,
are all diagonal matrices.
And step 106, for each eigenvalue vector to be diagnosed in the aligned eigenvalue vectors, determining n eigenvalue vectors which are closest to the eigenvalue vector to be diagnosed from the first aligned eigenvalue vector matrix. n is a positive integer.
In one example, the electronic device classifies the feature value vector of the target domain using a 5-neighbor classifier. The principle is that for the eigenvalue vector of each target domain, the eigenvalue vectors of 5 source domains closest to the Euclidean space distance are searched.
And step 107, determining the label of the fault type corresponding to the eigenvalue vector to be diagnosed according to the labels of the fault types to which the n eigenvalue vectors belong, and obtaining the fault type of the unlabeled data corresponding to the eigenvalue vector to be diagnosed.
In one example, determining the label of the fault type corresponding to the eigenvalue vector to be diagnosed according to the labels of the fault types to which the n eigenvalue vectors belong includes: and performing label voting on the labels of the fault types to which the n characteristic value vectors belong, and determining the voting result as the label of the fault type corresponding to the characteristic value vector to be diagnosed.
With reference to the foregoing embodiment and the early failure diagnosis process of the rotating mechanical bearing based on the improved migration learning shown in fig. 3, after the labeled data (source domain) and the unlabeled data to be diagnosed (target domain) are aligned based on CORAL, the aligned labeled data and the aligned unlabeled data to be diagnosed are obtained. And then, respectively extracting data characteristic values to obtain a shallow characteristic based on statistical characteristics and a depth characteristic based on wavelet scattering and stacking type self-toilet stool, wherein the characteristics form a characteristic value matrix of source domain data and a characteristic value matrix of target domain data. And establishing a KNN search tree for the eigenvalue matrix of the source domain data to obtain a representative eigenvalue matrix of the source domain data, aligning the eigenvalue vectors of the representative eigenvalue matrix and the eigenvalue matrix of the target domain data based on GFK, and classifying the target domain data by using a 5-NN classification method to obtain the fault type of each data to be diagnosed.
In summary, according to the early fault diagnosis method for the rotating mechanical bearing based on the improved transfer learning provided by the embodiment, the labeled data set and the data set to be diagnosed are subjected to data alignment, so that the aligned labeled data set and the aligned data set to be diagnosed are obtained; respectively extracting the characteristics of the aligned labeled data sets and the aligned data sets to be diagnosed to obtain characteristic value vector matrixes corresponding to the data sets; clustering the eigenvalue vector matrix corresponding to the aligned labeled data set to obtain an eigenvalue vector matrix corresponding to the label of each fault type; performing characteristic value vector alignment on the characteristic value vector matrix corresponding to the label of each fault type and the characteristic value vector matrix corresponding to the aligned data set to be diagnosed to obtain a first aligned characteristic value vector corresponding to the aligned labeled data set and a second aligned characteristic value vector corresponding to the aligned data set to be diagnosed; for each eigenvalue vector to be diagnosed in the second aligned eigenvalue vectors, n eigenvalue vectors which are closest to the eigenvalue vector to be diagnosed are determined from the first aligned eigenvalue vector matrix; according to the labels of the fault types to which the n characteristic value vectors belong, determining the label of the fault type corresponding to the characteristic value vector to be diagnosed, and obtaining the fault type of the label-free data corresponding to the characteristic value vector to be diagnosed; the problems that the existing machine learning method or deep learning method is used for early fault diagnosis of the bearing, the amount of labeled data in a data set is small, and the data distribution is different from the actual data distribution to be diagnosed; and (3) classifying the target domain label-free early-stage rotating mechanical bearing fault data by adopting an unsupervised mode only based on a small amount of labeled data which is different from the target domain data in distribution as a source domain.
In addition, a method of combining a shallow layer characteristic value and a depth characteristic value is adopted for extracting the data characteristic value, and the classification accuracy is improved.
In addition, the KNN search tree is used for screening the source domain characteristic value vectors, and the classification accuracy is improved.
In addition, a method of data alignment twice is adopted. First, the original data signal is aligned. Aligning the characteristic value vectors; the classification accuracy can be improved.
Fig. 4 is a block diagram of an early failure diagnosis apparatus for a rotating mechanical bearing based on improved migration learning according to an embodiment of the present application. The device at least comprises the following modules: a data acquisition module 410, a first alignment module 420, a feature extraction module 430, a feature clustering module 440, a second alignment module 450, a feature determination module 460, and a fault diagnosis module 470.
A data obtaining module 410, configured to obtain a labeled data set and a unlabeled data set to be diagnosed, where each group of labeled data in the labeled data set includes bearing-related working data and a label of a corresponding fault type; each group of unlabeled data in the data set to be diagnosed comprises bearing related working data;
a first alignment module 420, configured to perform data alignment on the labeled data set and the data set to be diagnosed, so as to obtain an aligned labeled data set and an aligned data set to be diagnosed;
a feature extraction module 430, configured to perform feature extraction on the aligned labeled data set and the aligned to-be-diagnosed data set, respectively, to obtain a feature value vector matrix corresponding to each data set; the eigenvalue vector matrix corresponding to each data set comprises shallow features and deep features;
a feature clustering module 440, configured to cluster the feature value vector matrix corresponding to the aligned labeled data set to obtain a feature value vector matrix corresponding to a label of each fault type;
a second alignment module 450, configured to perform eigenvalue vector alignment on the eigenvalue vector matrix corresponding to the tag of each fault type and the eigenvalue vector matrix corresponding to the aligned to-be-diagnosed data set, so as to obtain a first aligned eigenvalue vector corresponding to the aligned tagged data set and a second aligned eigenvalue vector corresponding to the aligned to-be-diagnosed data set;
a feature determination module 460, configured to, for each to-be-diagnosed feature value vector in the second aligned feature value vectors, determine n feature value vectors that are closest to the to-be-diagnosed feature value vector from the first aligned feature value vector matrix; the n is a positive integer;
and a fault diagnosis module 470, configured to determine, according to the label of the fault type to which the n eigenvalue vectors belong, a label of the fault type corresponding to the eigenvalue vector to be diagnosed, so as to obtain the fault type of the unlabeled data corresponding to the eigenvalue vector to be diagnosed.
For relevant details reference is made to the above-described method embodiments.
It should be noted that: the early failure diagnosis device for a rotating mechanical bearing based on improved migration learning provided in the above embodiments is only illustrated by the above division of the functional modules when performing early failure diagnosis for a rotating mechanical bearing based on improved migration learning, and in practical applications, the above function allocation may be completed by different functional modules according to needs, that is, the internal structure of the early failure diagnosis device for a rotating mechanical bearing based on improved migration learning is divided into different functional modules to complete all or part of the above described functions. In addition, the early fault diagnosis device for the rotating mechanical bearing based on the improved transfer learning and the early fault diagnosis method for the rotating mechanical bearing based on the improved transfer learning provided by the embodiment belong to the same concept, and specific implementation processes thereof are detailed in the method embodiment and are not described herein again.
Optionally, the present application further provides a computer-readable storage medium, in which a program is stored, and the program is loaded and executed by a processor to implement the early fault diagnosis method for a rotating mechanical bearing based on improved migration learning of the above method embodiment.
Optionally, the present application further provides a computer product, which includes a computer-readable storage medium, in which a program is stored, and the program is loaded and executed by a processor to implement the early fault diagnosis method for a rotating mechanical bearing based on improved migration learning of the above method embodiment.
The technical features of the embodiments described above may be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the embodiments described above are not described, but should be considered as being within the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.
The above is only one specific embodiment of the present application, and any other modifications based on the concept of the present application are considered as the protection scope of the present application.
Claims (9)
1. A method for early fault diagnosis of a rotating machine bearing based on improved transfer learning, the method comprising:
acquiring a labeled data set and a unlabeled data set to be diagnosed, wherein each group of labeled data in the labeled data set comprises bearing related working data and a label of a corresponding fault type; each group of unlabeled data in the data set to be diagnosed comprises bearing related working data;
performing data alignment on the labeled data set and the data set to be diagnosed to obtain an aligned labeled data set and an aligned data set to be diagnosed;
respectively extracting the characteristics of the aligned labeled data sets and the aligned data sets to be diagnosed to obtain characteristic value vector matrixes corresponding to the data sets; the eigenvalue vector matrix corresponding to each data set comprises shallow features and deep features;
clustering the eigenvalue vector matrix corresponding to the aligned labeled data set to obtain an eigenvalue vector matrix corresponding to the label of each fault type;
performing characteristic value vector alignment on the characteristic value vector matrix corresponding to the label of each fault type and the characteristic value vector matrix corresponding to the aligned data set to be diagnosed to obtain a first aligned characteristic value vector corresponding to the aligned labeled data set and a second aligned characteristic value vector corresponding to the aligned data set to be diagnosed;
for each eigenvalue vector to be diagnosed in the second aligned eigenvalue vectors, determining n eigenvalue vectors that are closest to the eigenvalue vector to be diagnosed from the first aligned eigenvalue vector matrix; the n is a positive integer;
and determining the label of the fault type corresponding to the eigenvalue vector to be diagnosed according to the label of the fault type to which the n eigenvalue vectors belong, so as to obtain the fault type of the unlabeled data corresponding to the eigenvalue vector to be diagnosed.
2. The method of claim 1, wherein the data aligning the labeled dataset and the dataset to be diagnosed comprises:
and performing data alignment on the labeled data set and the data set to be diagnosed based on a correlation alignment distance CORAL algorithm so as to enable the data distribution of the labeled data set to be the same as that of the data set to be diagnosed.
3. The method of claim 1, wherein the performing feature extraction on the aligned labeled dataset and the aligned dataset to be diagnosed respectively comprises:
respectively carrying out shallow feature extraction on the aligned labeled data sets and the aligned data sets to be diagnosed to obtain shallow features corresponding to each data set;
wherein the shallow features comprise at least one of time-domain features, frequency-domain features, and time-frequency-domain features of the bearing-related operational data;
the time domain features include: at least one of a mean, a variance, a root mean square value, a standard deviation, a peak-to-peak value, a maximum value, and a minimum value of the bearing related operational data;
the frequency domain features include: at least one of a frequency center, a frequency root mean square, a frequency standard deviation and a signal power spectrum of the bearing related working data;
the time-frequency domain features include: and wavelet coefficients are obtained through wavelet packet decomposition calculation.
4. The method of claim 1, wherein the performing feature extraction on the aligned labeled dataset and the aligned dataset to be diagnosed respectively comprises:
and respectively carrying out deep feature extraction on the aligned labeled data set and the aligned data set to be diagnosed by using a wavelet scattering network to obtain a first deep feature corresponding to each data set.
5. The method of claim 1, wherein the performing feature extraction on the aligned labeled dataset and the aligned dataset to be diagnosed respectively comprises:
and respectively carrying out deep feature extraction on the aligned labeled data set and the aligned data set to be diagnosed by using a stack type self-encoder to obtain a second deep feature corresponding to each data set.
6. The method according to claim 1, wherein the clustering the eigenvalue vector matrix corresponding to the aligned labeled data set to obtain an eigenvalue vector matrix corresponding to a label of each fault type comprises:
and clustering the eigenvalue vector matrix corresponding to the aligned labeled data set based on a K-nearest neighbor classification KNN algorithm by taking the label of the fault type as a central point of clustering respectively to obtain an eigenvalue vector matrix corresponding to the label of each fault type.
7. The method according to claim 1, wherein the performing eigenvalue vector alignment on the eigenvalue vector matrix corresponding to the label of each fault type and the eigenvalue vector matrix corresponding to the aligned dataset to be diagnosed comprises:
projecting the characteristic value vectors corresponding to the data sets to a Grassmann manifold space by using a ground wire kernel GFK algorithm;
feature vector alignment is performed on the grassmann manifold space.
8. The method according to claim 1, wherein the determining the label of the fault type corresponding to the eigenvalue vector to be diagnosed according to the label of the fault type to which the n eigenvalue vectors belong comprises:
and performing label voting on the labels of the fault types to which the n characteristic value vectors belong, and determining a voting result as the label of the fault type corresponding to the characteristic value vector to be diagnosed.
9. The method of any one of claims 1 to 8, wherein the bearing-related operational data comprises vibration data of the bearing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110153282.9A CN112990259B (en) | 2021-02-04 | 2021-02-04 | Early fault diagnosis method for rotary mechanical bearing based on improved transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110153282.9A CN112990259B (en) | 2021-02-04 | 2021-02-04 | Early fault diagnosis method for rotary mechanical bearing based on improved transfer learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112990259A true CN112990259A (en) | 2021-06-18 |
| CN112990259B CN112990259B (en) | 2023-12-26 |
Family
ID=76346800
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110153282.9A Active CN112990259B (en) | 2021-02-04 | 2021-02-04 | Early fault diagnosis method for rotary mechanical bearing based on improved transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112990259B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113642433A (en) * | 2021-07-30 | 2021-11-12 | 深圳市中金岭南有色金属股份有限公司凡口铅锌矿 | Method and device for determining working state of elevator, terminal equipment and medium |
| CN113674277A (en) * | 2021-10-22 | 2021-11-19 | 北京矩视智能科技有限公司 | Unsupervised domain adaptive surface defect region segmentation method and device and electronic equipment |
| CN114757305A (en) * | 2022-06-13 | 2022-07-15 | 华中科技大学 | Voltage transformer insulation fault identification method and system based on ensemble learning |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030041041A1 (en) * | 2001-03-01 | 2003-02-27 | Nello Cristianini | Spectral kernels for learning machines |
| US20150170055A1 (en) * | 2013-12-18 | 2015-06-18 | International Business Machines Corporation | Machine learning with incomplete data sets |
| US20150366540A1 (en) * | 2014-06-18 | 2015-12-24 | Kabushiki Kaisha Toshiba | Ultrasonic diagnostic apparatus, image processing apparatus, and image processing method |
| CN110346142A (en) * | 2019-07-25 | 2019-10-18 | 哈尔滨理工大学 | Fault Diagnosis of Roller Bearings under varying load based on the alignment of unsupervised feature |
| CN110443117A (en) * | 2019-06-24 | 2019-11-12 | 江苏师范大学 | A kind of method for fault diagnosis of wind turbines |
| CN111144458A (en) * | 2019-12-13 | 2020-05-12 | 重庆交通大学 | Method for identifying mechanical faults under different working conditions of subspace embedded feature distribution alignment |
| CN111397902A (en) * | 2020-03-22 | 2020-07-10 | 华南理工大学 | Rolling bearing fault diagnosis method based on feature alignment convolutional neural network |
| CN111829782A (en) * | 2020-07-16 | 2020-10-27 | 苏州大学 | Fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment |
| CN111898634A (en) * | 2020-06-22 | 2020-11-06 | 西安交通大学 | Intelligent fault diagnosis method based on depth-to-reactance-domain self-adaption |
| CN112183581A (en) * | 2020-09-07 | 2021-01-05 | 华南理工大学 | Semi-supervised mechanical fault diagnosis method based on self-adaptive migration neural network |
| CN112232252A (en) * | 2020-10-23 | 2021-01-15 | 湖南科技大学 | Optimal transport-based transmission chain unsupervised domain adaptive fault diagnosis method |
-
2021
- 2021-02-04 CN CN202110153282.9A patent/CN112990259B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030041041A1 (en) * | 2001-03-01 | 2003-02-27 | Nello Cristianini | Spectral kernels for learning machines |
| US20150170055A1 (en) * | 2013-12-18 | 2015-06-18 | International Business Machines Corporation | Machine learning with incomplete data sets |
| US20150366540A1 (en) * | 2014-06-18 | 2015-12-24 | Kabushiki Kaisha Toshiba | Ultrasonic diagnostic apparatus, image processing apparatus, and image processing method |
| CN110443117A (en) * | 2019-06-24 | 2019-11-12 | 江苏师范大学 | A kind of method for fault diagnosis of wind turbines |
| CN110346142A (en) * | 2019-07-25 | 2019-10-18 | 哈尔滨理工大学 | Fault Diagnosis of Roller Bearings under varying load based on the alignment of unsupervised feature |
| CN111144458A (en) * | 2019-12-13 | 2020-05-12 | 重庆交通大学 | Method for identifying mechanical faults under different working conditions of subspace embedded feature distribution alignment |
| CN111397902A (en) * | 2020-03-22 | 2020-07-10 | 华南理工大学 | Rolling bearing fault diagnosis method based on feature alignment convolutional neural network |
| CN111898634A (en) * | 2020-06-22 | 2020-11-06 | 西安交通大学 | Intelligent fault diagnosis method based on depth-to-reactance-domain self-adaption |
| CN111829782A (en) * | 2020-07-16 | 2020-10-27 | 苏州大学 | Fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment |
| CN112183581A (en) * | 2020-09-07 | 2021-01-05 | 华南理工大学 | Semi-supervised mechanical fault diagnosis method based on self-adaptive migration neural network |
| CN112232252A (en) * | 2020-10-23 | 2021-01-15 | 湖南科技大学 | Optimal transport-based transmission chain unsupervised domain adaptive fault diagnosis method |
Non-Patent Citations (3)
| Title |
|---|
| RONG GUI 等: "Statistical Scattering Component-Based Subspace Alignment for Unsupervised Cross-Domain PolSAR Image Classification", IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, pages 5449 - 5463 * |
| 康守强 等: "基于无监督特征对齐的变负载下滚动轴承故障诊断方法", 中国电机工程学报, pages 2 * |
| 詹宇斌;殷建平;刘新旺;: "局部切空间对齐算法的核主成分分析解释", 计算机工程与科学, no. 06, pages 158 - 161 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113642433A (en) * | 2021-07-30 | 2021-11-12 | 深圳市中金岭南有色金属股份有限公司凡口铅锌矿 | Method and device for determining working state of elevator, terminal equipment and medium |
| CN113642433B (en) * | 2021-07-30 | 2024-04-02 | 深圳市中金岭南有色金属股份有限公司凡口铅锌矿 | Method, device, terminal equipment and medium for determining working state of elevator |
| CN113674277A (en) * | 2021-10-22 | 2021-11-19 | 北京矩视智能科技有限公司 | Unsupervised domain adaptive surface defect region segmentation method and device and electronic equipment |
| CN113674277B (en) * | 2021-10-22 | 2022-02-22 | 北京矩视智能科技有限公司 | Unsupervised domain adaptive surface defect region segmentation method and device and electronic equipment |
| CN114757305A (en) * | 2022-06-13 | 2022-07-15 | 华中科技大学 | Voltage transformer insulation fault identification method and system based on ensemble learning |
| CN114757305B (en) * | 2022-06-13 | 2022-09-20 | 华中科技大学 | Voltage transformer insulation fault identification method and system based on ensemble learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112990259B (en) | 2023-12-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112990259B (en) | Early fault diagnosis method for rotary mechanical bearing based on improved transfer learning | |
| Jamieson et al. | Active ranking using pairwise comparisons | |
| CN111238807B (en) | Fault diagnosis method for planetary gear box | |
| CN109902399A (en) | Rolling bearing fault recognition methods under a kind of variable working condition based on ATT-CNN | |
| Magri et al. | Robust multiple model fitting with preference analysis and low-rank approximation | |
| CN111858991A (en) | Small sample learning algorithm based on covariance measurement | |
| CN110161388A (en) | A kind of the fault type recognition method and its system of high-tension apparatus | |
| CN110458189A (en) | Compressed sensing and depth convolutional neural networks Power Quality Disturbance Classification Method | |
| CN111310719B (en) | Unknown radiation source individual identification and detection method | |
| CN112418065A (en) | Equipment operation state identification method, device, equipment and storage medium | |
| CN114091504A (en) | Rotary machine small sample fault diagnosis method based on generation countermeasure network | |
| Kong et al. | A high generalizable feature extraction method using ensemble learning and deep auto-encoders for operational reliability assessment of bearings | |
| Ma et al. | Joint-label learning by dual augmentation for time series classification | |
| Zhang et al. | Texture feature extraction and description using Gabor wavelet in content-based medical image retrieval | |
| Wang et al. | Ensemble clustering-based fault diagnosis method incorporating traditional and deep representation features | |
| CN112699921B (en) | Stack denoising self-coding-based power grid transient fault data clustering cleaning method | |
| Wang et al. | A hybrid approach for identification of concurrent control chart patterns | |
| Dong et al. | Rolling bearing fault diagnosis using modified neighborhood preserving embedding and maximal overlap discrete wavelet packet transform with sensitive features selection | |
| CN117113078A (en) | Small sample bearing fault mode identification method and system based on multi-source data integration | |
| EP4285281A1 (en) | Annotation-efficient image anomaly detection | |
| CN115236272A (en) | Gas sensor fault diagnosis method and device under multi-working condition and storage medium | |
| Shah et al. | Analyzing data complexity using metafeatures for classification algorithm selection | |
| Liu et al. | A multisensory time-frequency features fusion method for rotating machinery fault diagnosis under nonstationary case | |
| CN117591813B (en) | Complex equipment fault diagnosis method and system based on multidimensional features | |
| Huo et al. | A Prototypical Networks-based Multi-task Model for Few-shot Fault Diagnosis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |