CN111829782B - Fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment - Google Patents

Fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment Download PDF

Info

Publication number
CN111829782B
CN111829782B CN202010688208.2A CN202010688208A CN111829782B CN 111829782 B CN111829782 B CN 111829782B CN 202010688208 A CN202010688208 A CN 202010688208A CN 111829782 B CN111829782 B CN 111829782B
Authority
CN
China
Prior art keywords
domain
matrix
sample
target domain
fault diagnosis
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202010688208.2A
Other languages
Chinese (zh)
Other versions
CN111829782A (en
Inventor
雷飘
沈长青
谢靖
张爱文
江星星
王俊
石娟娟
黄伟国
朱忠奎
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Suzhou University
Original Assignee
Suzhou University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Suzhou University filed Critical Suzhou University
Priority to CN202010688208.2A priority Critical patent/CN111829782B/en
Priority to PCT/CN2020/106703 priority patent/WO2022011754A1/en
Publication of CN111829782A publication Critical patent/CN111829782A/en
Application granted granted Critical
Publication of CN111829782B publication Critical patent/CN111829782B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/04—Bearings
    • 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
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]

Abstract

The invention discloses a fault diagnosis method based on self-adaptive manifold embedding dynamic distribution alignment, which can effectively avoid the characteristic distortion of data in an original Euclidean space by automatically calculating the optimal subspace dimension and calculating a geodesic flow kernel and the characteristic representation of a manifold after transformation. The method has the advantages of introducing similarity measurement A-distance to define a self-adaptive factor, dynamically adjusting the relative weight of sample data condition distribution and edge distribution, effectively reducing the distribution difference of the source domain samples and the target domain samples, greatly improving the accuracy and effectiveness of rolling bearing fault diagnosis under variable working conditions, along with strong interpretability, low requirement on computer hardware resources, higher execution speed, and excellent diagnosis accuracy, algorithm convergence and parameter robustness. The method is particularly suitable for multi-scene and multi-fault bearing fault diagnosis under variable working conditions, and can be widely applied to fault diagnosis tasks under variable working conditions of complex systems such as machinery, electric power, chemical engineering, aviation and the like.

Description

Fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment
Technical Field
The invention relates to the technical field of mechanical fault diagnosis and machine learning, in particular to a fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment.
Background
The rolling bearing is widely applied to various important fields of machinery, electric power, chemical engineering, aviation and the like as a key part in industrial production equipment, and meanwhile, the rolling bearing also frequently works in severe environments such as high temperature, high speed, heavy load and the like, so that the rolling bearing is very easy to generate faults such as abrasion, crack, fracture and the like.
Most of the existing bearing fault diagnosis technologies are used for analyzing vibration signals acquired by a sensor, such as fault characteristic frequency extraction, short-time Fourier transform, empirical mode decomposition, sparse representation and the like based on the vibration signals. In addition, a number of improved machine learning algorithms are also used to automatically learn bearing fault features, such as Support Vector Machines (SVMs), Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), and Automatic Encoders (AE), among others.
However, most of the current fault diagnosis methods are based on a common assumption that training data and test data must satisfy the same distribution, however, under actual complex working conditions, bearing vibration signal distributions acquired by sensors are usually inconsistent, so that the fault diagnosis effect is reduced sharply. Considering that the working principle and the failure mechanism of the rotary machine are similar, the bearing vibration data under the variable working condition can share similar fault characteristics, and the data distribution difference can be well overcome by the self-adaptive transfer learning. In particular, bearing data collected under varying conditions often have different condition distributions and edge distributions. Furthermore, when adaptive distribution alignment is performed in the original euclidean space, data feature distortion inevitably occurs, resulting in an unsatisfactory diagnostic effect of the model.
Disclosure of Invention
In view of the defects of the prior art, the invention aims to provide a fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment, which has high precision and excellent algorithm convergence and parameter robustness. The technical scheme is as follows:
a fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment comprises the following steps:
s10, collecting vibration signals of the bearing under different working conditions, adding various fault type labels of the bearing, taking the data collected under each working condition as a transferable data set, distributing the data in each transferable domain data set under different conditions and edges, then performing data cutting, and obtaining a sample;
s20, carrying out frequency domain processing on the bearing signals by using fast Fourier transform to obtain a source domain data set and a target domain data set, and inputting the source domain data set and the target domain data set as models;
s30, calculating the optimal subspace dimension through a GFK algorithm, and calculating a geodesic flow type kernel matrix and a transformed manifold feature representation;
s40, training a KNN-based classifier to generate preliminary prediction labels of target domain samples, and respectively calculating a maximum mean difference matrix M0And McRespectively corresponding to the edge distribution and the condition distribution of the sample data, and defining and calculating a self-adaptive dynamic factor through the similarity measurement A-distance;
s50, solving partial derivatives of the target loss function containing the structure risk minimization, the dynamic distribution alignment and the Laplace regularization term to obtain the optimal solution of the coefficient vector of the model, and then solving the secondary prediction label of the target domain sample through a representation theory;
s60, updating the maximum mean difference matrix M by using the secondary prediction label0And McRecalculating the adaptive dynamic factor, and solving a new optimal solution of the objective function to obtain a cubic prediction label of the target domain sample;
and S70, iterating the steps S40-S60 until the algorithm converges, obtaining a final prediction label of the target domain sample, and completing bearing fault diagnosis.
As a further improvement of the present invention, step S20 specifically includes:
converting a bearing time domain vibration signal into a frequency domain signal through Fast Fourier Transform (FFT), reserving unilateral frequency spectrum information of a frequency domain, and giving a marked source domain data set
Figure BDA0002588376500000021
And a target domain data set
Figure BDA0002588376500000022
Wherein x issAnd xtRespectively source domain and target domain samples, n and m are corresponding sample numbers, ysIs a source domain sample label, k is a sample feature number, and
Figure BDA0002588376500000031
as a further improvement of the present invention, the step S30 of calculating the optimal subspace dimension specifically includes:
respectively obtaining subspace data sets P of a source domain and a target domain through principal component analysissAnd PtCalculate PsAnd PtIs combined with the matrix Ps+tAnd separately calculate Ps、PtAnd Ps+tSine included angle alpha therebetweendAnd betadThe subspace inconsistency metric (SDM) c (d) may be expressed as:
C(d)＝0.5[sinαd+sinβd]
here, the more similar the source domain and the target domain, the larger the value of c (d), and under the premise of ensuring that the captured variance of one subspace can be transferred to other subspaces, that is: alpha is alphad≠π/2,βdNot equal to pi/2, the optimal subspace dimension d can be calculated by a greedy algorithm*：
d*＝min{d|C(d)＝1}。
As a further improvement of the present invention, the calculating of the geodesic flow kernel matrix and the transformed manifold feature representation in step S30 specifically includes:
computing PCA subspace PsAnd PtPrincipal angle of (theta)i(0≤θ1≤θ2≤…θdLess than or equal to pi/2), respectively calculating cos thetaiAnd sin θi(i ═ 1,2 ·, d) and as diagonal elements of the diagonal matrices Γ and Σ, PsIs complemented by
Figure BDA0002588376500000032
And
Figure BDA0002588376500000033
represents a pair of orthogonal matrices and can be found by singular value decomposition of:
Figure BDA0002588376500000034
further, the geodesic flow kernel matrix G may be calculated by the following formula:
Figure BDA0002588376500000035
wherein, Λ1To Λ3For a diagonal matrix, the diagonal elements are:
Figure BDA0002588376500000036
the transformed manifold features are further represented as:
Figure BDA0002588376500000037
wherein X ═ Xs,Xt]Source domain sample set W after conversionsAnd a target domain sample set WtCan be calculated by the following formula:
Figure BDA0002588376500000041
as a further improvement of the present invention, the maximum mean difference matrix M is calculated in step S400And McThe method specifically comprises the following steps:
Figure BDA0002588376500000042
Figure BDA0002588376500000043
wherein n iscAnd mcAre respectively a source domain DsAnd a target domain DtThe number of samples of the class c tag in (1).
As a further improvement of the present invention, in step S40, an adaptive dynamic factor is defined and calculated by a similarity measure a-distance, which specifically includes:
using a support vector machine SVM to train a linear two-classifier h on the source domain and the target domain, and using epsilon (h) to represent the loss of the classifier, the similarity measure a-distance across domains can be represented as:
A(Ds,Dt)＝2(1-2ε(h))
the adaptation factor ξ may be defined as:
Figure BDA0002588376500000044
wherein A isMAnd AcRespectively representing an edge similarity measure and a conditional similarity measure.
As a further improvement of the present invention, step S50 specifically includes:
s51, solving an optimal solution of the objective loss function; the method specifically comprises the following steps:
the structural risk minimization function defined on the source domain can be expressed as:
Figure BDA0002588376500000051
wherein HKAnd | · | non-conducting phosphorFRespectively representing the reproducible Hilbert space (RKHS) and the Frobenius norm,
Figure BDA0002588376500000052
is the coefficient vector to be solved, tr (-) is the trace of the solving matrix,
Figure BDA0002588376500000053
is a kernel matrix, and Kij＝K(wi,wj) In the model, an RBF radial basis kernel function is adopted,
Figure BDA0002588376500000054
is a label matrix composed of source domain sample labels and target domain sample prediction labels,
Figure BDA0002588376500000055
is a diagonal indicating matrix and satisfies:
Figure BDA0002588376500000056
s52, in order to fully mine the geometric characteristics among the neighboring samples in the manifold space, a Laplace regularization term Lap (D) is added to the objective functions,Dt) Expressed as follows:
Figure BDA0002588376500000057
wherein L is laplace matrix and L-a-U, a denotes diagonal matrix, and
Figure BDA0002588376500000058
u is a neighbor sample incidence matrix and can be obtained by the following calculation:
Figure BDA0002588376500000061
here, sim (,) is a similarity function used to evaluate the correlation between two samples, Np(wi) Is any one sample wiP neighbor sample sets;
s53, the fault bearing signal collected under the variable working condition often has different condition distribution and edge distribution, so it is necessary to evaluate the relative importance of the two distributions, the maximum mean difference MMD can evaluate the deviation of the two cross-domain distributions, this text defines a self-adaptive factor xi to dynamically adjust the two distributions, and the source domain D expresses the theory and nuclear skillsAnd a target domain DtDIS (D) of data distributions,Dt) Can be expressed as:
Figure BDA0002588376500000062
where E [. cndot. ] represents the mean of the samples mapped to the regenerated Hilbert space, and M is the MMD combination matrix of the conditional distribution and the edge distribution, which can be calculated by:
Figure BDA0002588376500000063
here, when ξ → 0, it indicates that the inter-domain conditional distribution adaptation is more important, and conversely, when ξ → 1, it indicates that the inter-domain edge distribution adaptation is more important;
s54, constructing and solving an objective function to obtain the optimal solution of the coefficient vector
Figure BDA0002588376500000064
The method for obtaining the target domain sample label by using the representation theory specifically comprises the following steps:
the Laplace regularization term Lap (D) described aboves,Dt) And a dynamically distributed alignment term DIS (D)s,Dt) Added to the structure risk minimization objective function, a global objective loss function is obtained:
Figure BDA0002588376500000065
by solving the partial derivatives of the objective function
Figure BDA0002588376500000066
Obtaining an optimal coefficient solution
Figure BDA0002588376500000067
Figure BDA0002588376500000071
Wherein, I is a unit diagonal matrix, λ, ρ, and η are regularization parameters of corresponding terms, and a quadratic prediction label of a target domain sample can be obtained by using a representation theory in machine learning:
Figure BDA0002588376500000072
as a further improvement of the present invention, step S60 specifically includes:
updating the maximum mean difference matrix M with the quadratic prediction label0And McAnd recalculating the self-adaptive dynamic factor, solving a new optimal solution of the objective function, and obtaining a cubic prediction label of the target domain sample by using a representation theory.
As a further improvement of the present invention, the step S10 of collecting vibration signals of the bearing under different working conditions specifically includes: when the bearing rotates and runs under four different working conditions, the vibration signals of the bearing under each working condition are collected through the acceleration sensor.
As a further development of the invention, the four different operating conditions are four different radial loads exerted on the bearing.
The invention has the beneficial effects that:
1. by automatically calculating the optimal subspace dimension and calculating the geodesic flow kernel and the converted manifold feature representation, the feature distortion of data in the original Euclidean space can be effectively avoided.
2. And a self-adaptive factor is defined by introducing similarity measurement A-distance, the relative weight of sample data condition distribution and edge distribution is dynamically adjusted, the distribution difference of the source domain sample and the target domain sample is effectively reduced, and the accuracy and the effectiveness of the fault diagnosis of the rolling bearing under the variable working condition are greatly improved.
3. Compared with the bearing fault diagnosis based on deep learning, the method has the advantages of strong interpretability, lower requirement on computer hardware resources, higher execution speed, and excellent diagnosis accuracy, algorithm convergence and parameter robustness. The method is particularly suitable for multi-scene and multi-fault bearing fault diagnosis under variable working conditions, and can be widely applied to fault diagnosis tasks under variable working conditions of complex systems such as machinery, electric power, chemical engineering, aviation and the like.
The foregoing description is only an overview of the technical solutions of the present invention, and in order to make the technical means of the present invention more clearly understood, the present invention may be implemented in accordance with the content of the description, and in order to make the above and other objects, features, and advantages of the present invention more clearly understood, the following preferred embodiments are described in detail with reference to the accompanying drawings.
Drawings
FIG. 1 is a flow chart of a fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment in a preferred embodiment of the present invention;
FIG. 2 is a frequency domain plot of vibration signals for different health states of a bearing in a preferred embodiment of the present invention;
FIG. 3 is a diagram illustrating conditional distribution and edge distribution of cross-domain data according to a preferred embodiment of the present invention;
FIG. 4 is a schematic diagram of a fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment in a preferred embodiment of the present invention;
FIG. 5 is a schematic diagram of the results of fault diagnosis on experimental data in a preferred embodiment of the present invention;
FIG. 6 is a diagram illustrating the convergence result of the fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment according to the preferred embodiment of the present invention;
FIG. 7 is a parameter sensitivity diagram of the fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment in the preferred embodiment of the present invention.
Detailed Description
The present invention is further described below in conjunction with the following figures and specific examples so that those skilled in the art may better understand the present invention and practice it, but the examples are not intended to limit the present invention.
The invention is explained in detail in connection with the actual experimental data:
experimental data a bearing data set from the university of kasseiki storage (CWRU) was used, and a data acquisition system, as shown in fig. 1, included a 2hp motor (1 hp. 746W), a torque sensor, a dynamometer, and electronic control. Faults are introduced in the test bearing roller, the inner ring and the outer ring through an Electric Discharge Machining (EDM) technology, and different fault sizes are set. And collecting vibration data by using an acceleration sensor.
In this embodiment, a bearing vibration signal with a sampling frequency of 12KHz is selected as original data, the model of the bearing is 6205-2RS JEM, bearing faults are divided into four types, namely normal, inner ring fault, roller fault and outer ring fault, and the fault size in each fault state includes 3 sizes: 0.007 inches, 0.014 inches and 0.021 inches, so there are a total of 3 x 3+ 1-10 health states, sampling 100 signal samples per operating state, each signal being 1024 points in length. The test bearing is subjected to 3 kinds of radial loads, namely 0hp, 1hp, 2hp and 3hp respectively, the rotating speed of the bearing slightly changes under each kind of load, the 10 bearing health state data are collected under any load and are taken as a transferable domain, and specifically, the domains under different loads are represented as: l0 ═ 0hp/1797rpm, L1 ═ 1hp/1772rpm, L2 ═ 2hp/1750rpm, and L3 ═ 3hp/1730 rpm. Specific bearing data can be found in table 1.
Figure BDA0002588376500000091
Table 1 description of 10 failed bearings per migratable domain
As shown in fig. 1, the method for diagnosing a fault based on adaptive manifold embedding dynamic distribution alignment according to the present invention includes the following steps:
s10, collecting vibration signals of the bearing under different working conditions, adding various fault type labels of the bearing, taking the data collected under each working condition as a transferable data set, distributing the data in each transferable domain data set according to different conditions and edges, then performing data cutting, and obtaining samples.
Wherein, gather bearing vibration signal under the different operating modes, specifically include: when the bearing rotates and runs under four different working conditions, the vibration signals of the bearing under each working condition are collected through the acceleration sensor. The four different operating conditions are four different radial loads applied to the bearing.
Specifically, when the test bearing rotates and runs under four different radial loads (0hp, 1hp, 2hp and 3hp), a bearing time domain signal under each load is collected through an acceleration sensor, various fault type labels (10 types) of the bearing are added, as shown in the above table 1, data collected under each load is taken as a transferable domain, data of each domain are subjected to different conditions and edge distribution, data are cut according to the above experimental description, and a sample is obtained, wherein any two of the four transferable domains can define two bidirectional transfer tasks, namely, L0 → L1, L0 → L2, L0 → L3; l1 → L0, L1 → L2, L1 → L3; l2 → L0, L2 → L1, L2 → L3; l3 → L0, L3 → L1, L3 → L2, a total of 12 migration tasks, with the left side of the arrow being the labeled source domain and the right side being the unlabeled target domain, are diagnostic objects to be identified.
And S20, carrying out frequency domain processing on the bearing signals by using fast Fourier transform to obtain a source domain data set and a target domain data set, and taking the source domain data set and the target domain data set as model input.
Specifically, a bearing time domain vibration signal is converted into a frequency domain signal through Fast Fourier Transform (FFT), and single-side frequency spectrum information of a frequency domain is reserved; namely, the number of the input sample feature points is 512, and the bearing frequency domain signals in different health states are shown in fig. 2.
In particular, given a labeled source domain dataset
Figure BDA0002588376500000101
And a target domain data set
Figure BDA0002588376500000102
Wherein x issAnd xtRespectively source domain and target domain samples, n and m are corresponding sample numbers, ysIs a source domain sample label, k is a sample feature number, and
Figure BDA0002588376500000103
the basic assumption is that: feature space xs＝xtAnd a label space ys＝ytConditional distribution Qs(ys(x)|xs)≠Qt(yt(x)|xt) And edge distribution Ps(xs)≠Pt(xt). The method aims at constructing a cross-domain classifier through manifold feature learning and dynamic distribution alignment
Figure BDA0002588376500000104
To predict the target domain label under variable working conditions
Figure BDA0002588376500000105
And S30, calculating the optimal subspace dimension through a GFK algorithm, and calculating a geodesic flow type kernel matrix and a transformed manifold feature representation.
Wherein, calculating the optimal subspace dimension specifically comprises:
respectively obtaining subspace data sets P of a source domain and a target domain through principal component analysissAnd PtCalculate PsAnd PtIs combined with the matrix Ps+tAnd separately calculate Ps、PtAnd Ps+tSine included angle alpha therebetweendAnd betadThe subspace inconsistency metric (SDM) c (d) may be expressed as:
C(d)＝0.5[sinαd+sinβd]
here, the more similar the source domain and the target domain, the larger the value of c (d), and under the premise of ensuring that the captured variance of one subspace can be transferred to other subspaces, that is: alpha is alphad≠π/2,βdNot equal to pi/2, the optimal subspace dimension d can be calculated by a greedy algorithm*：
d*＝min{d|C(d)＝1}。
The method for calculating the geodesic flow type core matrix and the manifold feature representation after transformation specifically comprises the following steps:
computing PCA subspace PsAnd PtPrincipal angle of (theta)i(0≤θ1≤θ2≤…θdLess than or equal to pi/2), respectively calculating cos thetaiAnd sin θi(i ═ 1,2 ·, d) and as diagonal elements of the diagonal matrices Γ and Σ, PsIs complemented by
Figure BDA0002588376500000111
And
Figure BDA0002588376500000112
represents a pair of orthogonal matrices and can be found by singular value decomposition of:
Figure BDA0002588376500000113
further, the geodesic flow kernel matrix G may be calculated by the following formula:
Figure BDA0002588376500000114
wherein, Λ1To Λ3For a diagonal matrix, the diagonal elements are:
Figure BDA0002588376500000115
the transformed manifold features are further represented as:
Figure BDA0002588376500000116
wherein X ═ Xs,Xt]Source domain sample set W after conversionsAnd a target domain sample set WtCan be calculated by the following formula:
Figure BDA0002588376500000117
s40, training a KNN-based classifier to generate preliminary prediction labels of target domain samples, and respectively calculating a maximum mean difference matrix M0And McAnd respectively corresponding to the edge distribution and the condition distribution of the sample data, and defining and calculating the self-adaptive dynamic factor through the similarity measurement A-distance.
Wherein a KNN-based classifier is trained to generate a preliminary prediction label of the target domain sample, wherein the number of nearest neighbor samples is set to 1, KNN is only used in the preliminary prediction, and all the subsequent target domain labels are generated by iteration.
Wherein the maximum mean difference matrix M is calculated separately0And McThe method specifically comprises the following steps:
Figure BDA0002588376500000121
Figure BDA0002588376500000122
wherein n iscAnd mcAre respectively a source domain DsAnd a target domain DtThe number of samples in the class c label, n and m are the source domain DsAnd a target domain DtNumber of samples, n-m-1000, nc＝mc＝100。
The method comprises the following steps of defining and calculating an adaptive dynamic factor through a similarity measurement A-distance, and specifically comprises the following steps:
using a support vector machine SVM to train a linear two-classifier h on the source domain and the target domain, and using epsilon (h) to represent the loss of the classifier, the similarity measure a-distance across domains can be represented as:
A(Ds,Dt)＝2(1-2ε(h))
the adaptation factor ξ may be defined as:
Figure BDA0002588376500000123
wherein A isMAnd AcRespectively representing an edge similarity measure and a conditional similarity measure. Calculation of AMWhen the training is needed, only the data of the source domain and the data of the target domain are respectively regarded as a whole, and two classification training is carried out; calculation of AcThen, the data of c-type labels in the source domain and the target domain are respectively regarded as a whole, and two-classification training is carried out, namely 10 times (10 labels) of two-classification tasks need to be executed, and the average value of 10 average absolute errors epsilon (h), namely A, is calculatedc。
S50, solving partial derivatives of the target loss function containing the structure risk minimization, the dynamic distribution alignment and the Laplace regularization term to obtain the optimal solution of the coefficient vector of the model, and then solving the quadratic prediction label of the target domain sample through a representation theory. The method specifically comprises the following steps:
s51, solving an optimal solution of the objective loss function; the method specifically comprises the following steps:
the structural risk minimization function defined on the source domain can be expressed as:
Figure BDA0002588376500000131
wherein HKAnd | · | non-conducting phosphorFRespectively representing the reproducible Hilbert space (RKHS) and the Frobenius norm,
Figure BDA0002588376500000132
is the coefficient vector to be solved, tr (-) is the trace of the solving matrix,
Figure BDA0002588376500000133
is a kernel matrix, and Kij＝K(wi,wj) In the model, an RBF radial basis kernel function is adopted,
Figure BDA0002588376500000134
is a label matrix composed of source domain sample labels and target domain sample prediction labels,
Figure BDA0002588376500000135
is a diagonal indicating matrix and satisfies:
Figure BDA0002588376500000136
s52, in order to fully mine the geometric characteristics among the neighboring samples in the manifold space, a Laplace regularization term Lap (D) is added to the objective functions,Dt) Expressed as follows:
Figure BDA0002588376500000137
wherein L is laplace matrix and L-a-U, a denotes diagonal matrix, and
Figure BDA0002588376500000138
u is a neighbor sample incidence matrix and can be obtained by the following calculation:
Figure BDA0002588376500000139
here, sim (,) is a similarity function used to evaluate the correlation between two samples, Np(wi) Is any one sample wiP neighbor sample sets;
s53, the fault bearing signals collected under the variable working condition often have different condition distribution and edge distribution, so that the relative importance of the two distributions needs to be evaluated, and the maximum average value differenceThe different MMD can evaluate two kinds of cross-domain distribution deviation, an adaptive factor xi is defined in the text to dynamically adjust the two kinds of distribution, and the source domain D expresses theoretical and nuclear skillssAnd a target domain DtDIS (D) of data distributions,Dt) Can be expressed as:
Figure BDA0002588376500000141
where E [. cndot. ] represents the mean of the samples mapped to the regenerated Hilbert space, and M is the MMD combination matrix of the conditional distribution and the edge distribution, which can be calculated by:
Figure BDA0002588376500000142
here, when ξ → 0, it indicates that the inter-domain conditional distribution adaptation is more important, and conversely, when ξ → 1, it indicates that the inter-domain edge distribution adaptation is more important;
s54, constructing and solving an objective function to obtain the optimal solution of the coefficient vector
Figure BDA0002588376500000143
The method for obtaining the target domain sample label by using the representation theory specifically comprises the following steps:
the Laplace regularization term Lap (D) described aboves,Dt) And a dynamically distributed alignment term DIS (D)s,Dt) Added to the structure risk minimization objective function, a global objective loss function is obtained:
Figure BDA0002588376500000144
by solving the partial derivatives of the objective function
Figure BDA0002588376500000145
Obtaining an optimal coefficient solution
Figure BDA0002588376500000146
Figure BDA0002588376500000147
Wherein, I is a unit diagonal matrix, λ, ρ, and η are regularization parameters of corresponding terms, and a quadratic prediction label of a target domain sample can be obtained by using a representation theory in machine learning:
Figure BDA0002588376500000151
s60, updating the maximum mean difference matrix M by using the secondary prediction label0And McAnd recalculating the self-adaptive dynamic factor, and solving a new optimal solution of the objective function to obtain a cubic prediction label of the target domain sample. The method specifically comprises the following steps:
updating the maximum mean difference matrix M with the quadratic prediction label0And McAnd recalculating the self-adaptive dynamic factor, solving a new optimal solution of the objective function, and obtaining a cubic prediction label of the target domain sample by using a representation theory.
And S70, iterating the steps S40-S60 until the algorithm converges, obtaining a final prediction label of the target domain sample, and completing bearing fault diagnosis. The method specifically comprises the following steps:
iterating the steps S40-S60 until the algorithm converges, namely the prediction result is not changed any more, and obtaining the final prediction label of the target domain sample
Figure BDA0002588376500000152
Based on the above steps, fig. 4 shows a schematic block diagram of a diagnostic method corresponding to the present invention, fig. 5 is a diagnostic result of 12 migration tasks, the diagnostic result indicates that the average precision of the diagnostic tasks under 12 variable conditions is close to 100%, fig. 6 and fig. 7 respectively show detailed displays based on the experimental result, as can be seen from fig. 6, the present invention can realize algorithm convergence within a very short number of iterations when different migration diagnostic tasks are executed, fig. 7 fully illustrates that the present invention has good robustness to algorithm hyper-parameters while ensuring high diagnostic precision, i.e., when the present invention is used for a diagnostic task outside an experiment, excellent classification effect can be obtained without rich prior knowledge.
In summary, the bearing fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment under variable working conditions can effectively avoid the characteristic distortion of data in the original Euclidean space by automatically calculating the optimal subspace dimension and calculating the geodesic flow kernel and the characteristic representation of the manifold after transformation. And a self-adaptive factor is defined by introducing similarity measurement A-distance, the relative weight of sample data condition distribution and edge distribution is dynamically adjusted, the distribution difference of the source domain sample and the target domain sample is effectively reduced, and the accuracy and the effectiveness of the fault diagnosis of the rolling bearing under the variable working condition are greatly improved. Compared with the bearing fault diagnosis based on deep learning, the method has the advantages of strong interpretability, lower requirement on computer hardware resources, higher execution speed, and excellent diagnosis accuracy, algorithm convergence and parameter robustness. The method is particularly suitable for multi-scene and multi-fault bearing fault diagnosis under variable working conditions, and can be widely applied to fault diagnosis tasks under variable working conditions of complex systems such as machinery, electric power, chemical engineering, aviation and the like.
The above embodiments are merely preferred embodiments for fully illustrating the present invention, and the scope of the present invention is not limited thereto. The equivalent substitution or change made by the technical personnel in the technical field on the basis of the invention is all within the protection scope of the invention. The protection scope of the invention is subject to the claims.

Claims (8)

1. A fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment is characterized by comprising the following steps:
s10, collecting vibration signals of the bearing under different working conditions, adding various fault type labels of the bearing, taking the data collected under each working condition as a transferable data set, distributing the data in each transferable domain data set under different conditions and edges, then performing data cutting, and obtaining a sample;
s20, carrying out frequency domain processing on the bearing signals by using fast Fourier transform to obtain a source domain data set and a target domain data set, and inputting the source domain data set and the target domain data set as models;
s30, calculating the optimal subspace dimension through a GFK algorithm, and calculating a geodesic flow type kernel matrix and a transformed manifold feature representation;
s40, training a KNN-based classifier to generate preliminary prediction labels of target domain samples, and respectively calculating a maximum mean difference matrix M0And McRespectively corresponding to the edge distribution and the condition distribution of the sample data, and defining and calculating a self-adaptive dynamic factor through the similarity measurement A-distance;
s50, solving partial derivatives of the target loss function containing the structure risk minimization, the dynamic distribution alignment and the Laplace regularization term to obtain the optimal solution of the coefficient vector of the model, and then solving the secondary prediction label of the target domain sample through a representation theory;
s60, updating the maximum mean difference matrix M by using the secondary prediction label0And McRecalculating the adaptive dynamic factor, and solving a new optimal solution of the objective function to obtain a cubic prediction label of the target domain sample;
s70, iterating the steps S40-S60 until the algorithm converges, obtaining a final prediction label of the target domain sample, and completing bearing fault diagnosis;
in step S30, the calculating the optimal subspace dimension specifically includes:
respectively obtaining subspace data sets P of a source domain and a target domain through principal component analysissAnd PtCalculate PsAnd PtIs combined with the matrix Ps+tAnd separately calculate Ps、PtAnd Ps+tSine included angle alpha therebetweendAnd betadThe subspace inconsistency metric (SDM) c (d) may be expressed as:
C(d)＝0.5[sinαd+sinβd]
here, the more similar the source domain and the target domain, the larger the value of c (d), and under the premise of ensuring that the captured variance of one subspace can be transferred to other subspaces, that is: alpha is alphad≠π/2,βdNot equal to pi/2, the optimal subspace dimension d can be calculated by a greedy algorithm*：
d*＝min{d|C(d)＝1}
In step S30, calculating a geodesic flow kernel matrix and a transformed manifold feature representation, specifically including:
computing PCA subspace PsAnd PtPrincipal angle of (theta)i(0≤θ1≤θ2≤…θdLess than or equal to pi/2), respectively calculating cos thetaiAnd sin θi(i ═ 1,2 ·, d) and as diagonal elements of the diagonal matrices Γ and Σ, PsIs complemented by
Figure FDA0003338007890000021
Figure FDA0003338007890000022
And
Figure FDA0003338007890000023
a pair of orthogonal matrices is represented and,
Figure FDA0003338007890000024
represents another orthogonal matrix obtained by singular value decomposition, and can be found by singular value decomposition as follows:
Figure FDA0003338007890000025
further, the geodesic flow kernel matrix G may be calculated by the following formula:
Figure FDA0003338007890000026
wherein, Λ1To Λ3For a diagonal matrix, the diagonal elements are:
Figure FDA0003338007890000027
the transformed manifold features are further represented as:
Figure FDA0003338007890000028
wherein X ═ Xs,Xt]Source domain sample set W after conversionsAnd a target domain sample set WtCan be calculated by the following formula:
Figure FDA0003338007890000029
2. the fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment as claimed in claim 1, wherein the step S20 specifically includes:
converting a bearing time domain vibration signal into a frequency domain signal through Fast Fourier Transform (FFT), reserving unilateral frequency spectrum information of a frequency domain, and giving a marked source domain data set
Figure FDA00033380078900000210
And a target domain data set
Figure FDA0003338007890000031
Wherein x issAnd xtRespectively source domain and target domain samples, n and m are corresponding sample numbers, ysIs a source domain sample label, k is a sample feature number, and
Figure FDA0003338007890000032
3. the fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment as claimed in claim 1, wherein the maximum values are calculated in step S40Mean difference matrix M0And McThe method specifically comprises the following steps:
Figure FDA0003338007890000033
Figure FDA0003338007890000034
wherein n iscAnd mcAre respectively a source domain DsAnd a target domain DtThe number of samples of the class c tag in (1).
4. The fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment as claimed in claim 3, wherein the step S40 defines and calculates an adaptive dynamic factor by a similarity measure a-distance, which specifically includes:
using a support vector machine SVM to train a linear two-classifier h on the source domain and the target domain, and using epsilon (h) to represent the loss of the classifier, the similarity measure a-distance across domains can be represented as:
A(Ds,Dt)＝2(1-2ε(h))
the adaptation factor ξ may be defined as:
Figure FDA0003338007890000035
wherein A isMAnd AcRespectively representing an edge similarity measure and a conditional similarity measure.
5. The fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment as claimed in claim 1, wherein the step S50 specifically includes:
s51, solving an optimal solution of the objective loss function; the method specifically comprises the following steps:
the structural risk minimization function defined on the source domain can be expressed as:
Figure FDA0003338007890000041
wherein HKAnd | · | non-conducting phosphorFRespectively representing reproducible Hilbert space (RKHS) and Frobenius norm, wherein eta is more than or equal to 0 and is a regularization coefficient used for balancing the experience risk and the model complexity of the model to be learned,
Figure FDA0003338007890000042
is the coefficient vector to be solved, tr (-) is the trace of the solving matrix,
Figure FDA0003338007890000043
is a kernel matrix, and Kij＝K(wi,wj) In the model, an RBF radial basis kernel function is adopted,
Figure FDA0003338007890000044
is a label matrix composed of source domain sample labels and target domain sample prediction labels,
Figure FDA0003338007890000045
is a diagonal indicating matrix and satisfies:
Figure FDA0003338007890000046
s52, in order to fully mine the geometric characteristics among the neighboring samples in the manifold space, a Laplace regularization term Lap (D) is added to the objective functions,Dt) Expressed as follows:
Figure FDA0003338007890000047
wherein L is laplace matrix and L-a-U, a denotes diagonal matrix, and
Figure FDA0003338007890000048
u is a neighbor sample incidence matrix and can be obtained by the following calculation:
Figure FDA0003338007890000051
here, sim (,) is a similarity function used to evaluate the correlation between two samples, Np(wi) Is any one sample wiP neighbor sample sets;
s53, the fault bearing signal collected under the variable working condition often has different condition distribution and edge distribution, so it is necessary to evaluate the relative importance of the two distributions, the maximum mean difference MMD can evaluate the deviation of the two cross-domain distributions, this text defines a self-adaptive factor xi to dynamically adjust the two distributions, and the source domain D expresses the theory and nuclear skillsAnd a target domain DtDIS (D) of data distributions,Dt) Can be expressed as:
Figure FDA0003338007890000052
wherein, E [. C]Means, H, representing the mapping of samples to a regenerated Hilbert spaceKRepresenting the reproducible Hilbert space (RKHS), M is an MMD combinatorial matrix of conditional and marginal distributions, which can be calculated by:
Figure FDA0003338007890000053
here, when ξ → 0, it indicates that the inter-domain conditional distribution adaptation is more important, and conversely, when ξ → 1, it indicates that the inter-domain edge distribution adaptation is more important;
s54, constructing and solving an objective function to obtain the optimal solution of the coefficient vector
Figure FDA0003338007890000054
The method for obtaining the target domain sample label by using the representation theory specifically comprises the following steps:
the Laplace regularization term Lap (D) described aboves,Dt) And a dynamically distributed alignment term DIS (D)s,Dt) Added to the structure risk minimization objective function, a global objective loss function is obtained:
Figure FDA0003338007890000055
by solving the partial derivatives of the objective function
Figure FDA0003338007890000061
Obtaining an optimal coefficient solution
Figure FDA0003338007890000062
Figure FDA0003338007890000063
Wherein, I is a unit diagonal matrix, λ, ρ, and η are regularization parameters of corresponding terms, and a quadratic prediction label of a target domain sample can be obtained by using a representation theory in machine learning:
Figure FDA0003338007890000064
6. the fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment as claimed in claim 1, wherein the step S60 specifically includes:
updating the maximum mean difference matrix M with the quadratic prediction label0And McRecalculating the adaptive dynamic factor, solving the new optimal solution of the objective function, and obtaining the objective domain by using the expression theoryTriple predictive label of the sample.
7. The fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment as claimed in claim 1, wherein the step S10 of collecting vibration signals of the bearing under different operating conditions specifically includes: when the bearing rotates and runs under four different working conditions, the vibration signals of the bearing under each working condition are collected through the acceleration sensor.
8. The method of claim 7, wherein the four different operating conditions are four different radial loads applied to the bearing.
CN202010688208.2A 2020-07-16 2020-07-16 Fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment Active CN111829782B (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202010688208.2A CN111829782B (en) 2020-07-16 2020-07-16 Fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment
PCT/CN2020/106703 WO2022011754A1 (en) 2020-07-16 2020-08-04 Fault diagnosis method based on adaptive manifold embedded dynamic distribution alignment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010688208.2A CN111829782B (en) 2020-07-16 2020-07-16 Fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment

Publications (2)

Publication Number Publication Date
CN111829782A CN111829782A (en) 2020-10-27
CN111829782B true CN111829782B (en) 2021-12-07

Family

ID=72923559

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010688208.2A Active CN111829782B (en) 2020-07-16 2020-07-16 Fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment

Country Status (2)

Country Link
CN (1) CN111829782B (en)
WO (1) WO2022011754A1 (en)

Families Citing this family (34)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112414714A (en) * 2020-11-05 2021-02-26 苏州大学 Bearing fault diagnosis method based on self-adaptive manifold probability distribution
CN112668459B (en) * 2020-12-25 2022-09-13 合肥工业大学 Rolling bearing fault diagnosis method based on supervised contrast learning and convolutional neural network
CN112883994B (en) * 2020-12-28 2022-05-10 重庆邮电大学 Rotating machinery variable working condition fault diagnosis method with balanced distribution adaptation
CN112629863B (en) * 2020-12-31 2022-03-01 苏州大学 Bearing fault diagnosis method for dynamic joint distribution alignment network under variable working conditions
CN112990259B (en) * 2021-02-04 2023-12-26 西交利物浦大学 Early fault diagnosis method for rotary mechanical bearing based on improved transfer learning
CN112883326B (en) * 2021-03-19 2022-07-08 吉林大学 Self-adaptive time-frequency transformation method based on stream algorithm
CN113255432B (en) * 2021-04-02 2023-03-31 中国船舶重工集团公司第七0三研究所 Turbine vibration fault diagnosis method based on deep neural network and manifold alignment
CN113158878B (en) * 2021-04-19 2022-10-28 合肥工业大学 Heterogeneous migration fault diagnosis method, system and model based on subspace
CN113947565B (en) * 2021-09-03 2023-04-18 中国科学院西安光学精密机械研究所 Structured light illumination super-resolution imaging gene detection method based on deep learning
CN114429152A (en) * 2021-12-31 2022-05-03 苏州大学 Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption
CN114371008B (en) * 2022-01-13 2023-12-29 福州大学 Rolling bearing cross-working condition fault diagnosis method based on VMD multi-domain features and MEDA
CN114813105B (en) * 2022-04-11 2023-04-28 西安热工研究院有限公司 Gear box fault early warning method and system based on working condition similarity evaluation
CN114510880B (en) * 2022-04-19 2022-07-12 中国石油大学(华东) Method for diagnosing working condition of sucker-rod pump based on Fourier transform and geometric characteristics
CN114842245A (en) * 2022-04-19 2022-08-02 华南理工大学 Image classification model training method, target classification method, device and medium
CN114997266B (en) * 2022-04-22 2023-07-11 苏州大学 Feature transfer learning method and system for voice recognition
CN114740726B (en) * 2022-04-22 2022-10-21 北京大学 Industrial process small sample fault diagnosis method based on deep optimal feature transmission
CN114970620B (en) * 2022-05-20 2024-03-08 西安交通大学 Cross-equipment rolling bearing fault targeted migration diagnosis method
CN114993710B (en) * 2022-06-02 2023-01-31 浙江加力仓储设备股份有限公司 Remote control type electric-drive folding forklift hub rapid maintenance device and method thereof
CN115062425B (en) * 2022-06-06 2023-08-18 华电电力科学研究院有限公司 Fault early warning method, system and device applied to gas turbine unit
CN115495271A (en) * 2022-06-27 2022-12-20 中国人民解放军海军航空大学 Fault diagnosability quantitative evaluation method based on maximum mean difference
CN115410088B (en) * 2022-10-10 2023-10-31 中国矿业大学 Hyperspectral image field self-adaption method based on virtual classifier
CN115618202A (en) * 2022-10-17 2023-01-17 徐州徐工随车起重机有限公司 Mechanical fault diagnosis method based on manifold embedding and key feature selection
CN115389247B (en) * 2022-11-01 2023-02-03 青岛睿发工程咨询服务合伙企业(有限合伙) Rotating machinery fault monitoring method based on speed self-adaptive encoder
CN115563494B (en) * 2022-12-01 2023-03-10 北京科技大学 Data-driven fault detection method of multi-mode process industrial process based on transfer learning
CN116167288B (en) * 2023-04-19 2023-07-11 中国科学技术大学 UA2FT semi-supervised transfer learning modeling method, medium and equipment
CN116302762B (en) * 2023-05-12 2023-08-18 北京必示科技有限公司 Evaluation method and system for fault location application based on red-blue countermeasure
CN116558828B (en) * 2023-07-10 2023-09-15 昆明理工大学 Rolling bearing health state assessment method based on autocorrelation coefficient sparsity characteristic
CN116720109B (en) * 2023-08-09 2023-10-20 傲拓科技股份有限公司 FPGA-based improved local linear embedded fan bearing fault diagnosis method
CN116774055A (en) * 2023-08-10 2023-09-19 清华四川能源互联网研究院 SVD-MMD strategy-based lithium ion battery SOC cross-domain estimation method, system and medium
CN116878885B (en) * 2023-09-04 2023-12-19 佛山科学技术学院 Bearing fault diagnosis method based on self-adaptive joint domain adaptive network
CN117009794B (en) * 2023-09-27 2023-12-15 山东大学 Machine fault diagnosis method and system based on unsupervised subdomain self-adaption
CN117096070B (en) * 2023-10-19 2024-01-05 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) Semiconductor processing technology abnormality detection method based on field self-adaption
CN117312962A (en) * 2023-11-28 2023-12-29 中国计量大学 Long-service elevator guide rail fault diagnosis method based on transfer learning and data driving
CN117664567B (en) * 2024-01-30 2024-04-02 东北大学 Rolling bearing cross-domain fault diagnosis method for multi-source domain unbalanced data

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106599922A (en) * 2016-12-16 2017-04-26 中国科学院计算技术研究所 Transfer learning method and transfer learning system for large-scale data calibration
JP2017142353A (en) * 2016-02-10 2017-08-17 株式会社Ｗｏｒｌｄ Ｔａｌｋ Ｂｏｘ Language learning device, language learning method, and language learning program
CN108414226A (en) * 2017-12-25 2018-08-17 哈尔滨理工大学 Fault Diagnosis of Roller Bearings under the variable working condition of feature based transfer learning
CN110619342A (en) * 2018-06-20 2019-12-27 鲁东大学 Rotary machine fault diagnosis method based on deep migration learning
CN111060318A (en) * 2020-01-09 2020-04-24 山东科技大学 Bearing fault diagnosis method based on deep countermeasure migration network
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

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101667164B1 (en) * 2015-10-02 2016-10-17 경북대학교 산학협력단 Rotor speed-based bearing fault diagnosis method
JP6695243B2 (en) * 2016-09-09 2020-05-20 Ntn株式会社 Bearing component life diagnosis method, bearing component life diagnosis device, and bearing component life diagnosis program
CN110346142B (en) * 2019-07-25 2021-06-29 哈尔滨理工大学 Rolling bearing fault diagnosis method under variable load based on unsupervised feature alignment
CN110686897A (en) * 2019-10-17 2020-01-14 中国矿业大学 Variable working condition rolling bearing fault diagnosis method based on subspace alignment
CN111337256B (en) * 2020-03-27 2020-12-29 西安交通大学 Method for diagnosing fault depth local migration of rolling bearing weighted by domain asymmetry factor

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2017142353A (en) * 2016-02-10 2017-08-17 株式会社Ｗｏｒｌｄ Ｔａｌｋ Ｂｏｘ Language learning device, language learning method, and language learning program
CN106599922A (en) * 2016-12-16 2017-04-26 中国科学院计算技术研究所 Transfer learning method and transfer learning system for large-scale data calibration
CN108414226A (en) * 2017-12-25 2018-08-17 哈尔滨理工大学 Fault Diagnosis of Roller Bearings under the variable working condition of feature based transfer learning
CN110619342A (en) * 2018-06-20 2019-12-27 鲁东大学 Rotary machine fault diagnosis method based on deep migration learning
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
CN111060318A (en) * 2020-01-09 2020-04-24 山东科技大学 Bearing fault diagnosis method based on deep countermeasure migration network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Geodesic flow kernel for unsupervised domain adaptation;B. Gong,Y. Shi,F. Sha,K. Grauman;《Computer Vision and Pattern Recognitio》;20121231;第2066-2073页 *
源域多样本集成GFK的不同工况下滚动轴承寿命状态识别;陈仁祥 等;《振动工程学报》;20200630;第33卷(第3期);第614-621页 *

Also Published As

Publication number Publication date
CN111829782A (en) 2020-10-27
WO2022011754A1 (en) 2022-01-20

Similar Documents

Publication Publication Date Title
CN111829782B (en) Fault diagnosis method based on adaptive manifold embedding dynamic distribution alignment
Zhao et al. Deep multi-scale convolutional transfer learning network: A novel method for intelligent fault diagnosis of rolling bearings under variable working conditions and domains
Grezmak et al. Interpretable convolutional neural network through layer-wise relevance propagation for machine fault diagnosis
Yang et al. A polynomial kernel induced distance metric to improve deep transfer learning for fault diagnosis of machines
Haidong et al. Intelligent fault diagnosis of rolling bearing using deep wavelet auto-encoder with extreme learning machine
Song et al. Retraining strategy-based domain adaption network for intelligent fault diagnosis
Chen et al. Predicting tool wear with multi-sensor data using deep belief networks
Appana et al. Reliable fault diagnosis of bearings with varying rotational speeds using envelope spectrum and convolution neural networks
Lu et al. Dominant feature selection for the fault diagnosis of rotary machines using modified genetic algorithm and empirical mode decomposition
Jia et al. A deviation based assessment methodology for multiple machine health patterns classification and fault detection
Xiong et al. Application of convolutional neural network and data preprocessing by mutual dimensionless and similar gram matrix in fault diagnosis
Liang et al. Data-driven anomaly diagnosis for machining processes
CN112414714A (en) Bearing fault diagnosis method based on self-adaptive manifold probability distribution
Zhang et al. A novel data-driven method based on sample reliability assessment and improved CNN for machinery fault diagnosis with non-ideal data
Bao et al. An enhanced sparse filtering method for transfer fault diagnosis using maximum classifier discrepancy
Xu et al. Fault diagnosis of rolling bearing based on online transfer convolutional neural network
Liang et al. Semisupervised subdomain adaptation graph convolutional network for fault transfer diagnosis of rotating machinery under time-varying speeds
Xu et al. Deep domain adversarial method with central moment discrepancy for intelligent transfer fault diagnosis
Wang et al. An effective multi-channel fault diagnosis approach for rotating machinery based on multivariate generalized refined composite multi-scale sample entropy
CN114429152A (en) Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption
Addeh et al. Control chart patterns detection using COA based trained MLP neural network and shape features
Chen et al. Adaptation regularization based on transfer learning for fault diagnosis of rotating machinery under multiple operating conditions
Luo et al. Adaptive Fisher-based deep convolutional neural network and its application to recognition of rolling element bearing fault patterns and sizes
Wang et al. Self-adaptive fusion of local-temporal features for tool condition monitoring: A human experience free model
Xiong et al. Combining the multi-genetic algorithm and support vector machine for fault diagnosis of bearings

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant