CN114383846A - Bearing composite fault diagnosis method based on fault label information vector - Google Patents

Bearing composite fault diagnosis method based on fault label information vector Download PDF

Info

Publication number
CN114383846A
CN114383846A CN202210010510.1A CN202210010510A CN114383846A CN 114383846 A CN114383846 A CN 114383846A CN 202210010510 A CN202210010510 A CN 202210010510A CN 114383846 A CN114383846 A CN 114383846A
Authority
CN
China
Prior art keywords
fault
composite
label information
vector
data
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
Application number
CN202210010510.1A
Other languages
Chinese (zh)
Other versions
CN114383846B (en
Inventor
徐娟
李康
吕增威
丁煦
樊玉琦
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.)
Hefei University of Technology
Original Assignee
Hefei University of Technology
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 Hefei University of Technology filed Critical Hefei University of Technology
Priority to CN202210010510.1A priority Critical patent/CN114383846B/en
Publication of CN114383846A publication Critical patent/CN114383846A/en
Application granted granted Critical
Publication of CN114383846B publication Critical patent/CN114383846B/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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation

Landscapes

  • Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • General Physics & Mathematics (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a bearing composite fault diagnosis method based on fault label information vectors, which comprises the steps of constructing a zero sample learning model, and carrying out composite fault diagnosis by using the zero sample learning model; in the zero sample learning model, a data preprocessing module converts original fault data into a fault image; the label information definition module is used for constructing fault label information vectors according to original fault data, wherein the fault label information vectors comprise single fault label information vectors and composite fault label information vectors; the feature extraction module is used for extracting a fault feature vector from the fault image; generating a mapping relation between a countermeasure module learning fault feature space and a fault label information space; and the classification module performs composite fault diagnosis by using the composite fault label information vector according to the mapping relation between the fault feature space and the fault label information space and the incidence relation between the single fault label information vector and the composite fault label information vector.

Description

Bearing composite fault diagnosis method based on fault label information vector
Technical Field
The invention relates to the technical field of composite fault diagnosis of bearings, in particular to a bearing composite fault diagnosis method based on fault label information vectors.
Background
In bearing fault diagnosis, composite fault diagnosis has been a difficult problem due to the coupling of different fault parameters, the diversity of fault characteristics, and the exponential increase in the number of possible fault modes.
The traditional composite fault diagnosis method mainly comprises a qualitative experience-based method, an analysis model-based method and a signal analysis-based method. And describing the functional structure of the system by using incomplete prior knowledge based on a qualitative experience method, and establishing a qualitative model to realize reasoning. Analytical model-based methods generate relevant information between normal and abnormal operation by studying the internal relationships between dynamic parameters and response signals under fault conditions. The method based on signal analysis needs to extract fault features for diagnosis through direct reasoning on the basis of a large amount of original sensing data. However, these methods require expert knowledge and engineering experience, and are difficult to apply to real industrial scenes.
In the field of mechanical fault diagnosis, the development of machine learning, especially deep learning, has become very common. On this basis, a learning model-based approach is proposed for automatically learning representative features and identifying compound faults from raw sensory data, rather than from expert data. However, existing methods of learning models for diagnosing compound faults are generally based on supervised or semi-supervised learning, and sufficient labeled or unlabeled training data is required for learning each compound fault. In an industrial scenario, complex fault training data, whether labeled or unlabeled, is often difficult to collect, sometimes even impossible to obtain, while single fault samples are readily available.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention provides a bearing composite fault diagnosis method based on a fault label information vector, which can identify invisible composite faults by using a single fault sample.
In order to achieve the purpose, the invention adopts the following technical scheme that:
the bearing composite fault diagnosis method based on the fault label information vector is characterized in that a zero sample learning model is constructed, and composite fault diagnosis is carried out by utilizing the zero sample learning model; the zero sample learning model includes: the system comprises a data preprocessing module, a label information definition module, a feature extraction module, a generation countermeasure module and a classification module;
the data preprocessing module is used for converting original fault data phi (t), namely a one-dimensional vibration signal, into a two-dimensional image, namely a fault image q; the original fault data φ (t) includes: single fault data phis(t) and composite fault data φu(t); the failure image q includes a single failure image qsAnd composite fault image qu；
The label information definition module is used for constructing a fault label information vector G according to original fault data, and comprises the following steps: single failure tag information vector GsAnd a composite fault label information vector Gu；
The feature extraction module is used for extracting a fault feature vector from the fault image q converted by the data preprocessing module, and comprises: for single fault image qsExtracting single fault feature vector vsFor the composite failure image quExtracting composite fault feature vector vu；
The generation countermeasure module is used for learning the mapping relation between the fault feature space and the fault label information space;
the classification module is used for classifying the fault according to the mapping relation between the fault feature space and the fault label information space and according to the single fault label information vector GsAnd a composite fault label information vector GuThe incidence relation between the two and utilizes a compound fault label information vector GuAnd carrying out compound fault diagnosis.
Further, the data preprocessing module converts the original fault data phi (t), namely the one-dimensional vibration signal, into a two-dimensional wavelet image, namely a fault image q, through wavelet transformation.
Further, the composite fault is composed of different single faults, and each single fault category corresponds to a single fault label information vector GsEach composite fault category corresponds to a composite fault label information vector Gu(ii) a The label information definition module firstly roots single fault data phis(t) generating a single failure label information vector GsAnd then obtaining a composite fault label information vector G according to the single fault category contained in the composite faultu(ii) a The details are as follows:
s31, the label information definition module utilizes the single fault data phis(t) extracting the single-fault label information vector GsSingle fault label information vector GsAll the dimensions of (A) are C multiplied by 1; wherein, the single fault label information vector corresponding to the kth single fault category
Figure BDA0003458911170000021
Namely, it is
Figure BDA0003458911170000022
Figure BDA0003458911170000023
Figure BDA0003458911170000024
Single fault label information vector corresponding to k-th single fault category
Figure BDA0003458911170000025
The C-th dimensional data of (1), 2, 3 … C; the superscript K represents the single fault category serial number, K is 1, 2, 3 … K, and K represents the total number of single fault categories; subscript s denotes single fault;
selecting single fault data aiming at the kth single fault category
Figure BDA0003458911170000031
Of (3), specifically
Figure BDA0003458911170000032
Namely, it is
Figure BDA0003458911170000033
Figure BDA0003458911170000034
Indicating the kth single fault categoryCorresponding single fault data
Figure BDA0003458911170000035
C-th data point amplitude of (1), 2, 3 … C;
from single fault data corresponding to K single fault categories
Figure BDA0003458911170000036
The maximum value of C data points in (1) is selected as the threshold value mu, i.e.
Figure BDA0003458911170000037
Dividing the amplitude range of the data point into five equal parts according to the threshold value mu, and judging the single fault data of the kth single fault category
Figure BDA0003458911170000038
To obtain the single fault label information vector of the kth single fault category
Figure BDA0003458911170000039
Each dimension data of
Figure BDA00034589111700000310
The values of (A) are as follows:
Figure BDA00034589111700000311
s32, label information definition module according to the incidence relation between the compound fault and the single fault, namely the compound fault is composed of several different single faults, combining the corresponding single fault label information vector GsObtaining a composite fault label information vector Gu(ii) a Composite fault label information vector GuThe dimensions of (A) are also C × 1;
wherein, the composite fault label information vector is composed of single faults with class serial numbers of 1, … and J
Figure BDA00034589111700000312
As follows:
Figure BDA00034589111700000313
Figure BDA00034589111700000314
the superscript 1, …, J refers to the category number of the single fault that constitutes the composite fault, J being 2, 3 …, K.
Further, the training mode of the feature extraction module is as follows: the feature extraction module is used for carrying out single-fault image qsExtracting a single fault feature vector vsAnd predicting the single failure image qsPredicted failure label p ofs(ii) a The characteristic extraction module is used for extracting a single fault image q according to an input single fault imagesActual failure label y ofsAnd predicted failure label psCalculating the classification loss of the feature extraction module, and finishing the training of the feature extraction module if the classification loss meets the requirement;
training completed feature extraction module for single fault image qsExtracting single fault feature vector vsFor the composite failure image quExtracting composite fault feature vector vu。
Further, the generation countermeasure module comprises a generator and an arbiter; the generator is used for generating fault characteristic vectors by taking a fault label information vector G and Gaussian noise z as conditions
Figure BDA0003458911170000041
The discriminator is used for discriminating the generated fault characteristic vector
Figure BDA0003458911170000042
And the difference between the extracted fault feature vectors v;
the training mode of the generation countermeasure module is as follows: generator labels information vector G with single faultsGeneration of single fault feature vector conditioned on sum of Gaussian noise z
Figure BDA0003458911170000043
The discriminator discriminates the generated single fault feature vector
Figure BDA0003458911170000044
And extracted single fault feature vector vsThe difference between them; generating a countermeasure module based on the generated single fault feature vector
Figure BDA0003458911170000045
And extracted single fault feature vector vsCalculating loss according to the difference between the two, and if the loss meets the requirement, finishing the training of the generation countermeasure module;
after the generation countermeasure module finishes training, the generator uses each single fault label information vector GsGenerating each single fault characteristic vector by taking the sum of Gaussian noise z as a condition
Figure BDA0003458911170000046
With each composite fault label information vector GuGenerating each composite fault characteristic vector by taking the sum of Gaussian noise z as a condition
Figure BDA0003458911170000047
Further, the classification module performs a composite fault diagnosis in the following manner:
generating each composite fault feature vector by a generator
Figure BDA0003458911170000048
Each central point is used as each central point in the composite fault feature space, and each central point corresponds to one composite fault category;
composite fault feature vector v extracted by calculation feature extraction moduleuDistance from each central point, selecting and compounding fault characteristic vector vuThe central point with the minimum distance, and the composite fault category corresponding to the central point with the minimum distance is the central point with the minimum distanceComposite fault image vuThe diagnosis result of (1).
Further, the method for constructing the zero-sample learning model comprises the following steps:
s1, the data preprocessing module is used for processing the single-fault original fault data, namely the single-fault data phis(t) converting the two-dimensional wavelet image by wavelet transform to obtain a single failure image qs(ii) a The original fault data of the composite fault, namely the composite fault data phiu(t) converting the two-dimensional wavelet image by wavelet transform to obtain a composite failure image qu；
S2, defining a training set D of zero sample modelstrainAnd test set Dtest：
Training set D of zero sample learning modeltrainFrom single fault class CsBuild, including single fault data phis(t), Single Fault image qsSingle fault label ys(ii) a The single fault label ysThe single fault type is referred to, and the fault type of each single fault corresponds to a single fault label ysAnd each single failure tag ysCorresponding to a single fault label information vector GsSubscript s denotes single fault; the training set DtrainI.e. single fault class CsThe method comprises K single fault classes, and each single fault class comprises N samples, namely N single fault data phis(t) and N single failure images qs；
Test set D of zero sample learning modeltestFrom compound fault class CuConstruction of, including Compound Fault data phiu(t) composite failure image quComposite fault label yu(ii) a The composite fault is composed of a plurality of different single faults, and the label y of the composite faultuRefers to the fault category of the composite fault, i.e. the single fault category included in the composite fault, and each composite fault label yuCorresponding to a compound fault label information vector Gu(ii) a Subscript u represents a compound failure;
s3, the label information definition module first uses the single fault data phis(t) extracting the sheet fault attribute to obtain a sheetFailure tag information vector GsThen, according to the single fault category contained in the composite fault, obtaining the semantic vector g of the composite faultu(ii) a The details are as follows:
s31, the label information definition module utilizes the single fault data phis(t) extracting the single-fault label information vector GsSingle fault label information vector GsAll the dimensions of (A) are C multiplied by 1; wherein, the single fault label information vector corresponding to the kth single fault category
Figure BDA0003458911170000051
Namely, it is
Figure BDA0003458911170000052
Figure BDA0003458911170000053
Figure BDA0003458911170000054
Single fault label information vector corresponding to k-th single fault category
Figure BDA0003458911170000055
C represents a serial number of a dimension, and C is 1, 2, 3 … C; the superscript K indicates a single fault category number, K being 1, 2, 3 … K; subscript s denotes single fault;
selecting single fault data aiming at the kth single fault category
Figure BDA0003458911170000056
Of (3), specifically
Figure BDA0003458911170000057
Namely, it is
Figure BDA0003458911170000058
Figure BDA0003458911170000059
Representing single fault data corresponding to kth single fault category
Figure BDA00034589111700000510
C represents the serial number of the data point, and C is 1, 2, 3 … C;
single fault data corresponding to the Kth single fault category
Figure BDA00034589111700000511
The maximum value of C data points in (1) is selected as the threshold value mu, i.e.
Figure BDA00034589111700000512
Dividing the amplitude range of the data point into five equal parts according to the threshold value mu, and judging the single fault data of the kth single fault category
Figure BDA00034589111700000513
To obtain the single fault label information vector of the kth single fault category
Figure BDA00034589111700000514
Each dimension data of
Figure BDA00034589111700000515
The values of (A) are as follows:
Figure BDA0003458911170000061
s32, label information definition module according to the incidence relation between the compound fault and the single fault, namely the compound fault is composed of several different single faults, combining the corresponding single fault label information vector GsObtaining a composite fault label information vector Gu(ii) a Composite fault label information vector GuThe dimensions of (A) are also C × 1;
wherein, the composite fault label information vector is composed of single faults with class serial numbers of 1, … and J
Figure BDA0003458911170000062
As follows:
Figure BDA0003458911170000063
Figure BDA0003458911170000064
superscript 1, …, J refers to the category number of the single fault that constitutes the composite fault, J being 2, 3 …, K;
s4, the feature extraction module is a feature extractor established by a convolutional neural network,
using training set DtrainTraining the feature extractor: feature extractor for single fault image qsExtracting a single fault feature vector vsAnd predicting the single failure image qsPredicted failure label p ofs(ii) a The feature extractor is based on the input single fault image qsActual failure label y ofsAnd predicted failure label psCalculating the classification loss of the feature extractor, and finishing the training of the feature extractor if the classification loss meets the requirement;
after the training of the feature extractor is completed, the feature extractor pairs the test set DtestTo composite failure image q inuExtracting composite fault feature vector vu；
S5, the generation countermeasure module comprises a generator and a discriminator, wherein the generator is used for generating the fault feature vector by taking the fault label information vector G and the Gaussian noise z as conditions
Figure BDA0003458911170000065
The discriminator is used for discriminating the generated fault characteristic vector
Figure BDA0003458911170000066
And the difference between the extracted fault feature vectors v;
using training set DtrainPair generationTraining the anti-module, and generating a belief vector G with a single fault labelsGeneration of single fault feature vector conditioned on sum of Gaussian noise z
Figure BDA0003458911170000067
The discriminator discriminates the generated single fault feature vector
Figure BDA0003458911170000068
And extracted single fault feature vector vsThe difference between them; generating a countermeasure module based on the generated single fault feature vector
Figure BDA0003458911170000069
And extracted single fault feature vector vsCalculating loss according to the difference between the two, and if the loss meets the requirement, finishing the training of the generation countermeasure module;
after the generation countermeasure module finishes training, the generator respectively uses each composite fault label information vector GuGenerating each composite fault characteristic vector by taking the sum of Gaussian noise z as a condition
Figure BDA0003458911170000071
S6, classifying module pair test set DtestComposite failure image q of (1)uCarrying out a composite fault diagnosis test;
generating each composite fault feature vector by a generator
Figure BDA0003458911170000072
Each central point is used as each central point in the composite fault feature space, and each central point corresponds to one composite fault category;
computational feature extraction module for test set DtestComposite failure image q of (1)uExtracted composite fault feature vector vuDistance from each central point, selecting and compounding fault characteristic vector vuThe central point with the minimum distance, and the composite fault category corresponding to the central point with the minimum distance is the composite fault image vuThe diagnosis result of (1);
if test set DtestComposite failure image q of (1)uComposite fault label y ofuWith the composite fault image vuIf the diagnosis results are consistent, the composite fault diagnosis is correct; otherwise, a composite fault diagnosis error is indicated.
Further, in step S2,
training set DtrainAnd test set DtestSatisfying the condition in the following formula:
Figure BDA0003458911170000073
wherein, p (phi)s(t))、p(φu(t)) are respectively single-fault data phis(t), composite fault data φu(t) data distribution; i (-) is used to calculate mutual information between two data distributions, I (p (φ)s(t))、p(φu(t))) is the data distribution p (phi)s(t))、p(φu(t)) mutual information between; single fault label ysAnd a composite fault label yuAre disjoint.
Further, in step S4, the classification loss L1 of the feature extractor is as follows:
Figure BDA0003458911170000074
wherein, ys(i) Is the ith single failure image q in the single failure categorys(i) The single failure label of (1), i.e., single failure category; i represents the serial number of the single fault image, and i is 1, 2, 3 … N; p is a radical ofs(i) Is that the feature extraction module aims at the ith single fault image q in the single fault categorys(i) A predicted category of predicted faults.
Further, in step S5, the loss function L2 of the countermeasure module is generated as follows:
Figure BDA0003458911170000075
wherein p isrFor extracted single fault feature vector vsData distribution of pgFor the generated single fault feature vector
Figure BDA0003458911170000081
The data distribution of (2); dw(. to) is a model representation of the discriminator for discriminating between true and false scores of the fault feature vector, and subscript w is discriminator Dw(ii) model parameters of (g);
Figure BDA0003458911170000082
is generator with single fault label information vector GsAnd gaussian noise z,
Figure BDA0003458911170000083
Figure BDA0003458911170000084
for the model representation of the generator, for generating a fault feature vector, the subscript θ2Is other generator
Figure BDA0003458911170000085
The model parameters of (1); v. ofsIs a single fault feature vector extracted by a feature extractor;
Figure BDA0003458911170000086
in order to be a gradient penalty term,
Figure BDA0003458911170000087
is from vsAnd
Figure BDA0003458911170000088
the medium-level uniform sampling obtains a sampling single-fault feature vector,
Figure BDA0003458911170000089
representing sampled single-fault feature vectors
Figure BDA00034589111700000810
The distribution of the data of (a) is,
Figure BDA00034589111700000811
representing a sampled single fault feature vector
Figure BDA00034589111700000812
Derivation is carried out, wherein lambda is a gradient penalty coefficient;
Figure BDA00034589111700000813
is v issAnd
Figure BDA00034589111700000814
mean square error between.
The invention has the advantages that:
(1) the invention provides a zero sample model constructed based on a label information vector, the model can identify invisible compound faults by using vibration data of single faults through training, and the model generates high-dimensional compound fault characteristics for classification, so that the pivot point problem is further alleviated, and the performance of the model is improved.
(2) The invention designs a new method for defining the fault label information vector to solve the problem that the fault label information is annotated and defined, and the single fault label information vector comes from the peak value of the vibration signal. The composite fault signature information vector is a combination of single fault category information vectors based on theoretical correlations. Thus, without a composite fault instance, a composite fault label information vector set may be derived. The method does not need the prior knowledge of experts, is simple to calculate, and the experimental result verifies the feasibility of the method.
(3) The invention provides a zero sample model for compound fault diagnosis, which is constructed based on a fault label information vector and aims at solving the problem that compound fault training data with or without a label is difficult to acquire under the actual condition.
Drawings
FIG. 1 is a block diagram of a zero sample learning model of the present invention.
FIG. 2 is a flow chart of the construction of the zero sample learning model of the present invention.
Fig. 3 is a schematic diagram of vibration signals of the bearing of the present embodiment in seven fault states.
Fig. 4 is a comparison graph of the classification results of the composite fault based on different numbers of training samples in this embodiment.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1, according to the bearing composite fault diagnosis method based on the fault label information vector, a zero sample model is constructed by using the fault label information vector, and then composite fault diagnosis is performed by using the zero sample model.
The zero sample model consists of five parts: the system comprises a data preprocessing module, a label information definition module, a feature extraction module, a generation countermeasure module and a classification module.
The data preprocessing module is used for converting original fault data, namely a one-dimensional vibration signal, into a two-dimensional wavelet image, namely a fault image, through wavelet transformation;
the label information definition module is used for constructing a fault label information vector according to original fault data;
the feature extraction module is used for extracting a fault feature vector from the fault image converted by the data preprocessing module;
the generation countermeasure module is used for learning the mapping relation between the fault feature space and the fault label information space;
the classification module is used for performing fault classification on the fault feature vector according to the mapping relation between the fault feature space and the fault label information space;
the construction process of the zero sample model is as follows:
s1, constructing a data preprocessing module, which is configured to convert the original fault data Φ (t), i.e., the one-dimensional vibration data, into a two-dimensional wavelet image, i.e., a fault image q, through wavelet transform, as shown in detail below:
extracting useful information from original fault data, namely a one-dimensional vibration signal, and analyzing the details of the vibration data by using a Wavelet Transform (WT) mode, wherein the wavelet transform mode is shown as the following formula:
Figure BDA0003458911170000101
wherein a represents a scale factor for the scale transformation of the wavelet function ψ (·); b denotes a translation factor for the translation of the wavelet function ψ (·); phi (t) is a time domain vibration signal sequence, namely original fault data;
the definition of the wavelet function ψ (·) is shown below:
Figure BDA0003458911170000102
the invention utilizes wavelet transformation to convert every 256 time domain data points into 64 two-dimensional black-and-white time-frequency domain images, namely fault images, reflecting the characteristics of vibration signals on the time domain and the frequency domain, wherein the intermediate interest rate of the wavelet transformation is 0.8125;
wherein the original fault data of a single fault, i.e. the single fault data phis(t) converting the two-dimensional wavelet image by wavelet transform to obtain a single failure image qs(ii) a Original fault data of compound fault, i.e. compound fault data phiu(t) two-dimensional wavelet image converted by wavelet transformI.e. the composite fault image qu。
S2, defining a training set D of zero sample modelstrainAnd test set Dtest：
Training set D of zero sample learning modeltrainIs composed of a visible single fault class CsConstructed to include single fault data phis(t), Single Fault image qsSingle fault label ys(ii) a The single fault label ysThe single fault type is referred to, and the fault type of each single fault corresponds to a single fault label ysAnd each single failure tag ysCorresponding to a single fault label information vector GsSubscript s denotes single fault; the training set DtrainI.e. single fault class CsThe method comprises K single fault classes, and each single fault class comprises N samples, namely N single fault data phis(t) and N single failure images qs；
Test set D of zero sample learning modeltestIs composed of invisible compound fault class CuConstructed, including composite fault data phiu(t) composite failure image quComposite fault label yu(ii) a The composite fault is composed of a plurality of different single faults, and the label y of the composite faultuRefers to the fault category of the composite fault, i.e. the single fault category included in the composite fault, and each composite fault label yuCorresponding to a compound fault label information vector Gu(ii) a Subscript u represents a compound failure;
training set DtrainAnd test set DtestSatisfying the condition in the following formula:
Figure BDA0003458911170000111
wherein, p (phi)s(t))、p(φu(t)) are respectively single-fault data phis(t), composite fault data φu(t) data distribution; i (-) is used to calculate mutual information between two data distributions, I (p (φ)s(t))、p(φu(t))) is the dataDistribution p (phi)s(t))、p(φu(t)) mutual information between; single fault label ysAnd a composite fault label yuAre disjoint; guFor a composite fault signature information vector, GsFor single fault label information vector, compound fault label information vector GuPassable function
Figure BDA0003458911170000112
From a single fault label information vector GsThus obtaining the product.
S3, constructing a label information definition module, wherein for the fault label information vector in fault diagnosis, the existing definition method of the fault label information vector can not be directly useds(t) extracting single fault attribute as single fault label information vector GsBecause the composite fault is composed of different single faults, the label information definition module combines the corresponding single fault label information vector G according to the single fault category contained in the composite faultsObtaining a composite fault label information vector Gu(ii) a The details are as follows:
s31, the label information definition module utilizes the single fault data phis(t) extracting the single-fault label information vector GsSingle fault label information vector GsThe dimension of (a) is C × 1;
wherein, the single fault label information vector corresponding to the kth single fault category
Figure BDA0003458911170000113
Figure BDA0003458911170000114
Namely, it is
Figure BDA0003458911170000115
Figure BDA0003458911170000116
Represents the k < th >Single fault label information vector corresponding to single fault category
Figure BDA0003458911170000117
C represents a serial number of a dimension, and C is 1, 2, 3 … C;
selecting single fault data aiming at the kth single fault category
Figure BDA0003458911170000118
C data points in (1), specifically
Figure BDA0003458911170000119
Namely, it is
Figure BDA00034589111700001110
Figure BDA00034589111700001111
Representing single fault data corresponding to kth single fault category
Figure BDA00034589111700001112
The superscript K represents the single fault category number, and K is 1, 2, 3 … K; c represents the serial number of the data points, wherein C is 1, 2, 3 … C, and is larger than the period of the vibration signal; subscript s denotes single fault;
single fault data corresponding to the Kth single fault category
Figure BDA00034589111700001113
The maximum value of C data points in (1) is selected as the threshold value mu, i.e.
Figure BDA0003458911170000121
Dividing the amplitude range of the data point into five equal parts according to the threshold value mu, and judging the single fault data of the kth single fault category
Figure BDA0003458911170000122
To obtain the kth single eventSingle fault label information vector for fault classes
Figure BDA0003458911170000123
Each dimension data of
Figure BDA0003458911170000124
Is taken as
Figure BDA0003458911170000125
By the method, the single fault label information vector corresponding to the kth single fault category can be obtained
Figure BDA0003458911170000126
S32, label information definition module according to the incidence relation between the compound fault and the single fault, namely the compound fault is composed of several different single faults, combining the corresponding single fault label information vector GsObtaining a composite fault label information vector Gu(ii) a Composite fault label information vector GuThe dimension of (a) is C × 1;
wherein the composite fault label information vector
Figure BDA0003458911170000127
Figure BDA0003458911170000128
Figure BDA0003458911170000129
Superscript 1, …, J refers to the category number of the single fault that constitutes the composite fault, J being 2, 3 …, K;
in this embodiment, all single fault categories of the bearing are known, including an inner ring fault (IF), a rolling element fault (BF), and an outer ring fault (of), and all composite fault categories of the bearing can be obtained by arbitrarily combining the three single fault categories, and four composite fault categories are obtained: composite failure OF inner ring and rolling element (IF & BF), composite failure OF outer ring and rolling element (OF & BF), composite failure OF inner ring and outer ring (IF & OF), and composite failure OF inner ring, outer ring, and rolling element (IF & OF & BF).
S4, constructing a feature extraction module, where the feature extraction module is configured to extract a fault feature vector v from a fault image q, and the details are as follows:
establishing a feature extractor by using a Convolutional Neural Network (CNN), extracting fault feature vectors from the fault images, wherein the feature extractor, namely a convolutional layer, a pooling layer, a flattening layer and a full-link layer of the convolutional neural network, are sequentially represented as C, P, F, FC, the Input layer of the feature extractor, namely Input, is a two-dimensional wavelet image, namely a fault image q, and the output of the last full-link layer is a fault feature vector v;
in the present invention, the structure and parameters of the feature extractor are shown in table 1 below:
layer name Operation of Nucleus size Step size Number of nuclei
Input Input signal - - -
C1 Convolution with a bit line 5×5 1×1 32
P1 Pooling 2×2 2×2 32
C2 Convolution with a bit line 5×5 1×1 64
P2 Pooling 2×2 2×2 64
F Flattening 16384×1 - 1
FC1 Full connection 4096×1 - 1
FC2 Full connection 2048×1 - 1
FC3 Full connection 2048×1 - 1
TABLE 1
Using training set DtrainTraining the feature extraction module, and inputting a single fault image q into an input layer of the feature extractor in the training stage of the feature extraction modulesOutput single fault feature vector v of the last full link layersAnd the output layer of the feature extractor outputs the single fault image qsPredicted failure label p ofs(ii) a According to the single fault image qsActual failure label y ofsAnd predicted failure label psCalculating the classification loss of the feature extraction module, and finishing the training of the feature extraction module if the classification loss meets the requirement;
for a single fault category, the feature extraction module classifies a loss function L by optimizing1So that the prediction effect of the feature extraction module is better, and the classification loss function L1As follows:
Figure BDA0003458911170000131
wherein, ys(i) Is the ith single failure image q in the single failure categorys(i) The single failure label of (1), i.e., single failure category; i represents the serial number of the single fault image, and i is 1, 2, 3 … N; p is a radical ofs(i) Is that the feature extraction module aims at the ith single fault image q in the single fault categorys(i) Predicted predictive failure classRespectively;
and in the feature extraction module training stage, a back propagation algorithm is used for training the model parameters, and specifically an Adam algorithm is used.
Training completed feature extraction module for single fault image qsExtracting single fault feature vector vsFor the composite failure image quExtracting composite fault feature vector vu；
S5, constructing a generation countermeasure module, wherein the generation countermeasure module comprises a generator and a discriminator; the generator is used for generating fault characteristic vectors by taking a fault label information vector G and Gaussian noise z as conditions
Figure BDA0003458911170000132
The discriminator is used for discriminating the generated fault characteristic vector
Figure BDA0003458911170000141
And the difference between the extracted fault feature vectors v, thereby directing the generator to generate high quality fault feature vectors
Figure BDA0003458911170000142
The generator and the discriminator are mutually game, and the generator generates high-quality fault feature vectors
Figure BDA0003458911170000143
The discriminator can not discriminate the difference, and then a stable generator is obtained;
the details are as follows:
the discriminator tries to let the fault feature vector generated
Figure BDA0003458911170000144
And the difference L between the extracted single-fault feature vectors vDMaximization:
Figure BDA0003458911170000145
wherein p isrFor extracted faultsData distribution of feature vectors v, pgFor the generated fault feature vector
Figure BDA0003458911170000146
The data distribution of (2);
Figure BDA0003458911170000147
a model representation for a generator for generating a fault feature vector; dw(. to) is a model representation of a discriminator for discriminating between true and false scores, subscripts w and θ, of a fault feature vector2Is a generator
Figure BDA0003458911170000148
And a discriminator Dw(ii) model parameters of (g);
Figure BDA0003458911170000149
Figure BDA00034589111700001410
is a generator
Figure BDA00034589111700001411
Generating a fault characteristic vector by taking a fault label information vector G and Gaussian noise z as conditions; v is the true fault feature vector extracted by the feature extractor.
Figure BDA00034589111700001412
The method is characterized in that a discrimination feature extractor extracts the average value of scores obtained by various fault feature vectors,
Figure BDA00034589111700001413
the mean value of the fractions obtained by generating various fault feature vectors by a discrimination generator is used for calculating the difference between the two, so that the extracted real fault feature vector v and the generated fault feature vector are combined
Figure BDA00034589111700001414
As close as possible, and, in addition, to ensure
Figure BDA00034589111700001415
As close as possible to Dw(v) But not exceeding Dw(v) The invention also adds a penalty term of stable gradient, namely:
Figure BDA00034589111700001416
Figure BDA00034589111700001417
are from v and
Figure BDA00034589111700001418
uniformly sampling to obtain a sampling single fault feature vector;
Figure BDA00034589111700001419
representing a straight-line uniform sampling between pairs of points sampled from the data profile Pr and the generator profile Pg, i.e. sampling a single fault feature vector
Figure BDA00034589111700001420
The data distribution of (2);
Figure BDA00034589111700001421
representing sampled fault feature vectors
Figure BDA00034589111700001422
And (5) derivation, wherein lambda is a gradient penalty coefficient.
In the present invention, the structure of the resulting countermeasure module is shown in table 2 below:
Figure BDA0003458911170000151
TABLE 2
Using training set DtrainGenerating training reactance module, generator using single fault label information vector GsGeneration of single fault feature vector conditioned on sum of Gaussian noise z
Figure BDA0003458911170000152
The discriminator is used for discriminating the generated single fault characteristic vector
Figure BDA0003458911170000153
And extracted single fault feature vector vsDifference between them, thereby directing the generator to generate high quality single fault feature vectors
Figure BDA0003458911170000154
The generator and the discriminator are mutually game, and the generator generates high-quality single fault feature vectors
Figure BDA0003458911170000155
The discriminator cannot discriminate the difference, so that a stable generator is obtained, and the training of generating the countermeasure module is completed;
generating a countermeasure module by constantly optimizing a loss function L2Obtaining feature vectors capable of generating high quality single fault
Figure BDA0003458911170000156
Of the loss function L2As follows:
Figure BDA0003458911170000157
wherein p isrFor extracted single fault feature vector vsData distribution of pgFor the generated single fault feature vector
Figure BDA0003458911170000158
The data distribution of (2); dw(. to) is a model representation of the discriminator for discriminating between true and false scores of the fault feature vector, and subscript w is discriminator Dw(ii) model parameters of (g);
Figure BDA0003458911170000159
is generator with single fault label information vector GsAnd gaussian noise z,
Figure BDA00034589111700001510
Figure BDA00034589111700001511
for the model representation of the generator, for generating a fault feature vector, the subscript θ2Is other generator
Figure BDA00034589111700001512
The model parameters of (1); v. ofsIs a single fault feature vector extracted by a feature extractor; .
Figure BDA0003458911170000161
In order to be a gradient penalty term,
Figure BDA0003458911170000162
is from vsAnd
Figure BDA0003458911170000163
the medium-level uniform sampling obtains a sampling single-fault feature vector,
Figure BDA0003458911170000164
representing sampled single-fault feature vectors
Figure BDA0003458911170000165
The distribution of the data of (a) is,
Figure BDA0003458911170000166
representing a sampled single fault feature vector
Figure BDA0003458911170000167
Derivation is carried out, wherein lambda is a gradient penalty coefficient;
Figure BDA0003458911170000168
is v issAnd
Figure BDA0003458911170000169
mean square error between;
Figure BDA00034589111700001610
is v issAnd
Figure BDA00034589111700001611
norm of (d); | | non-woven hair2Is a notation of norm.
Loss function L2Item 1 and item 2 in the above are used for solving the problems of high training difficulty and unstable gradient and mean square error, and are used for further reducing the real single-fault feature vector vsAnd the generated single fault feature vector
Figure BDA00034589111700001612
The distance between them.
And a generation countermeasure module training stage, wherein a back propagation algorithm is used for training the model parameters, and specifically, an RMSprop algorithm is used.
After the generation countermeasure module finishes training, the generator uses each composite fault label information vector GuGenerating each composite fault characteristic vector by taking the sum of Gaussian noise z as a condition
Figure BDA00034589111700001613
S6, constructing a classification module, wherein the classification module is used for carrying out composite fault classification according to the mapping relation between the fault feature space and the fault label information space, and utilizing the test set DtestComposite failure image q of (1)uAnd (3) carrying out a composite fault diagnosis test:
generating each composite fault feature vector by a generator
Figure BDA00034589111700001614
Each central point is used as each central point in the composite fault feature space, and each central point corresponds to one composite fault category;
the classification module is used for calculating the composite fault feature vector v extracted by the feature extraction moduleuDistance from each central point, selecting and compounding fault characteristic vector vuThe central point with the minimum distance, and the composite fault category corresponding to the central point with the minimum distance is the composite fault image vuThe predicted result of (2); the distance refers to the Euclidean distance;
if test set DtestComposite failure image q of (1)uComposite fault label y ofuWith the composite fault image vuIf the diagnosis results of the two are consistent, the composite fault diagnosis is correct.
In the invention, the zero sample model is trained by using the training set, namely the single fault set, and the zero sample model is tested by using the test set, namely the composite fault set, so that the accuracy of the composite fault classification of the zero sample model is verified.
In the embodiment, a test bed is used for collecting vibration signals of a fault bearing to verify the effectiveness of the method, in the test bed, the rotating speed of the bearing is controlled by a three-phase motor through a flexible coupling, an acceleration sensor is installed on a bearing seat to collect the vibration signals, and the sampling frequency is 51200 Hz.
The vibration signals for the seven fault conditions of the bearing are shown in figure 3. The vibration signals under seven fault conditions include: three single fault vibration signals and four compound fault vibration signals. The three single faults are respectively an inner ring fault (IF), a rolling Body Fault (BF) and an outer ring fault (OF); the four composite faults are respectively: composite failure OF inner ring and outer ring (IF & OF), composite failure OF inner ring and rolling element (IF & BF), composite failure OF outer ring and rolling element (OF & BF), and composite failure OF inner ring, outer ring and rolling element (IF & OF & BF).
The effectiveness of the model is verified through experiments on two groups of composite fault diagnosis tasks, the specific conditions are shown in table 3, the fault classification results of different training sample numbers are shown in fig. 4, the classification accuracy of the task A and the task B is obviously improved along with the increase of the training sample number, and when the training sample number of each single fault category is 2000, the average classification accuracy of the task A and the task B respectively reaches 77.03% and 65.80%. In addition, it can be observed that the classification accuracy of task a is higher than that of task B because task B is more complex and difficult to classify than task a, and task B mostly includes a composite fault coupled by three single faults, i.e., mostly includes composite fault data of inner and outer rings and rolling elements.
Figure BDA0003458911170000171
Figure BDA0003458911170000181
TABLE 3
The invention provides a zero sample model for compound fault diagnosis based on fault label information vector generation, aiming at the problem that compound fault training data with or without marks are difficult to collect under the actual condition. The invention designs a unified fault label information vector definition method to represent fault label information vectors of single faults and compound faults. The generation countermeasure module of the invention learns the mapping relation between the fault feature space and the fault label information space, and then the generator generates the fault feature vector through the fault label information vector. The model classifies the composite fault according to the similarity measurement between the real fault label information vector and the generated fault label information vector, and the experimental result on the experimental data set of the embodiment shows that the classification precision of the model is obvious under the condition of no composite fault sample.
The invention utilizes the zero sample learning model to carry out bearing composite fault diagnosis, and the method comprises the following steps:
s201, knowing all single fault categories of the bearing, wherein the composite fault is composed of a plurality of different single faults, and combining according to the single fault categories of the bearing to obtain all composite fault categories;
s202, the label vector definition module firstly utilizes a vibration signal of a single fault, namely single fault data phis(t) extracting single fault attribute as single fault label information vector GsBecause the composite fault is composed of different single faults, the label information definition module combines the corresponding single fault label information vector G according to the single fault category contained in the composite faultsObtaining a composite fault label information vector Gu(ii) a Specifically refer to step S3;
s203, the generator in the generation countermeasure module uses each composite fault label information vector G obtained in the step S202uRespectively mapped to the fault feature space, i.e. the generator respectively obtains each composite fault label information vector G in step S202uGenerating each composite fault feature vector by taking Gaussian noise z as condition
Figure BDA0003458911170000182
Generating each composite fault feature vector by a generator
Figure BDA0003458911170000183
As each central point in the composite fault feature space, each central point corresponds to a composite fault label information vector GuAnd each composite fault label information vector GuThe composite fault categories of the central points are known, namely, each central point corresponds to one composite fault category;
s204, the feature extraction module treats the predicted composite fault image quExtracting composite fault feature vector vu；
S205, calculating the extracted composite fault feature vector vuDistance from each central point in the composite fault feature space, selecting a composite fault feature vector vuThe central point with the minimum distance, and the composite fault category corresponding to the central point with the minimum distance is the composite fault image quThe predicted result of (1).
The invention is not to be considered as limited to the specific embodiments shown and described, but is to be understood to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the invention as defined by the appended claims.
The various flow charts in the present document do not necessarily have a sequential order of execution unless specifically stated otherwise. Although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (10)

1. The bearing composite fault diagnosis method based on the fault label information vector is characterized in that a zero sample learning model is constructed, and composite fault diagnosis is carried out by utilizing the zero sample learning model; the zero sample learning model includes: the system comprises a data preprocessing module, a label information definition module, a feature extraction module, a generation countermeasure module and a classification module;
the data preprocessing module is used for converting original fault data phi (t), namely a one-dimensional vibration signal, into a two-dimensional image, namely a fault image q; the original fault data φ (t) includes: single fault data phis(t) and composite fault data φu(t); the failure image q includes a single failure image qsAnd composite fault image qu；
The label information definition module is used for constructing a fault label information vector G according to original fault data, and comprises the following steps: single failure tag information vector GsAnd a composite fault label information vector Gu；
The feature extraction module is used for extracting a fault feature vector from the fault image q converted by the data preprocessing module, and comprises: for single fault image qsExtracting single fault feature vector vsFor the composite failure image quExtracting composite fault feature vector vu；
The generation countermeasure module is used for learning the mapping relation between the fault feature space and the fault label information space;
the classification module is used for classifying the fault according to the mapping relation between the fault feature space and the fault label information space and according to the single fault label information vector GsAnd a composite fault label information vector GuThe incidence relation between the two and utilizes a compound fault label information vector GuAnd carrying out compound fault diagnosis.
2. The method for diagnosing composite fault of bearing based on information vector of fault label as claimed in claim 1, wherein said data preprocessing module converts original fault data phi (t), i.e. one-dimensional vibration signal, into two-dimensional wavelet image, i.e. fault image q, by wavelet transform.
3. The method according to claim 1, wherein the composite fault is composed of different single faults, and each single fault category corresponds to a single fault label information vector GsEach composite fault category corresponds to a composite fault label information vector Gu(ii) a The label information definition module firstly roots single fault data phis(t) generating a single failure label information vector GsAnd then obtaining a composite fault label information vector G according to the single fault category contained in the composite faultu(ii) a The details are as follows:
s31, the label information definition module utilizes the single fault data phis(t) extracting the single-fault label information vector GsSingle fault label information vector GsAll the dimensions of (A) are C multiplied by 1; wherein, the single fault label information vector corresponding to the kth single fault category
Figure FDA0003458911160000021
Namely, it is
Figure FDA0003458911160000022
Figure FDA0003458911160000023
Figure FDA0003458911160000024
Single fault label information vector corresponding to k-th single fault category
Figure FDA0003458911160000025
The C-th dimensional data of (1), 2, 3 … C; the superscript K represents the single fault category serial number, K is 1, 2, 3 … K, and K represents the total number of single fault categories; subscript s denotes single fault;
selecting single fault data aiming at the kth single fault category
Figure FDA0003458911160000026
Of (3), specifically
Figure FDA0003458911160000027
Namely, it is
Figure FDA0003458911160000028
Figure FDA0003458911160000029
Representing single fault data corresponding to kth single fault category
Figure FDA00034589111600000210
C-th data point amplitude of (1), 2, 3 … C;
from single fault data corresponding to K single fault categories
Figure FDA00034589111600000211
The maximum value of C data points in (1) is selected as the threshold value mu, i.e.
Figure FDA00034589111600000212
Dividing the amplitude range of the data point into five equal parts according to the threshold value mu, and judging the kth single fault classOther single fault data
Figure FDA00034589111600000213
To obtain the single fault label information vector of the kth single fault category
Figure FDA00034589111600000214
Each dimension data of
Figure FDA00034589111600000215
The values of (A) are as follows:
Figure FDA00034589111600000216
s32, label information definition module according to the incidence relation between the compound fault and the single fault, namely the compound fault is composed of several different single faults, combining the corresponding single fault label information vector GsObtaining a composite fault label information vector Gu(ii) a Composite fault label information vector GuThe dimensions of (A) are also C × 1;
wherein, the composite fault label information vector is composed of single faults with class serial numbers of 1, … and J
Figure FDA00034589111600000217
As follows:
Figure FDA00034589111600000218
Figure FDA00034589111600000219
the superscript 1, …, J refers to the category number of the single fault that constitutes the composite fault, J being 2, 3 …, K.
4. The method for diagnosing the composite fault of the bearing based on the fault label information vector as claimed in claim 1, wherein the training mode of the feature extraction module is as follows: the feature extraction module is used for carrying out single-fault image qsExtracting a single fault feature vector vsAnd predicting the single failure image qsPredicted failure label p ofs(ii) a The characteristic extraction module is used for extracting a single fault image q according to an input single fault imagesActual failure label y ofsAnd predicted failure label psCalculating the classification loss of the feature extraction module, and finishing the training of the feature extraction module if the classification loss meets the requirement;
training completed feature extraction module for single fault image qsExtracting single fault feature vector vsFor the composite failure image quExtracting composite fault feature vector vu。
5. The method according to claim 1, wherein the generation countermeasure module comprises a generator and a discriminator; the generator is used for generating fault characteristic vectors by taking a fault label information vector G and Gaussian noise z as conditions
Figure FDA0003458911160000031
The discriminator is used for discriminating the generated fault characteristic vector
Figure FDA0003458911160000032
And the difference between the extracted fault feature vectors v;
the training mode of the generation countermeasure module is as follows: generator labels information vector G with single faultsGeneration of single fault feature vector conditioned on sum of Gaussian noise z
Figure FDA0003458911160000033
The discriminator discriminates the generated single fault feature vector
Figure FDA0003458911160000034
And extracted single fault feature vector vsThe difference between them; generating a countermeasure module based on the generated single fault feature vector
Figure FDA0003458911160000035
And extracted single fault feature vector vsCalculating loss according to the difference between the two, and if the loss meets the requirement, finishing the training of the generation countermeasure module;
after the generation countermeasure module finishes training, the generator uses each single fault label information vector GsGenerating each single fault characteristic vector by taking the sum of Gaussian noise z as a condition
Figure FDA0003458911160000036
With each composite fault label information vector GuGenerating each composite fault characteristic vector by taking the sum of Gaussian noise z as a condition
Figure FDA0003458911160000037
6. The method for diagnosing the composite fault of the bearing based on the fault label information vector as claimed in claim 5, wherein the classification module performs the composite fault diagnosis in the following way:
generating each composite fault feature vector by a generator
Figure FDA0003458911160000038
Each central point is used as each central point in the composite fault feature space, and each central point corresponds to one composite fault category;
composite fault feature vector v extracted by calculation feature extraction moduleuDistance from each central point, selecting and compounding fault characteristic vector vuThe central point with the minimum distance, and the composite fault category corresponding to the central point with the minimum distance is the composite fault image vuThe diagnosis result of (1).
7. The method for diagnosing the composite fault of the bearing based on the fault label information vector is characterized in that the method for constructing the zero sample learning model comprises the following steps:
s1, the data preprocessing module is used for processing the single-fault original fault data, namely the single-fault data phis(t) converting the two-dimensional wavelet image by wavelet transform to obtain a single failure image qs(ii) a The original fault data of the composite fault, namely the composite fault data phiu(t) converting the two-dimensional wavelet image by wavelet transform to obtain a composite failure image qu；
S2, defining a training set D of zero sample modelstrainAnd test set Dtest：
Training set D of zero sample learning modeltrainFrom single fault class CsBuild, including single fault data phis(t), Single Fault image qsSingle fault label ys(ii) a The single fault label ysThe single fault type is referred to, and the fault type of each single fault corresponds to a single fault label ysAnd each single failure tag ysCorresponding to a single fault label information vector GsSubscript s denotes single fault; the training set DtrainI.e. single fault class CsThe method comprises K single fault classes, and each single fault class comprises N samples, namely N single fault data phis(t) and N single failure images qs；
Test set D of zero sample learning modeltestFrom compound fault class CuConstruction of, including Compound Fault data phiu(t) composite failure image quComposite fault label yu(ii) a The composite fault is composed of a plurality of different single faults, and the label y of the composite faultuRefers to the fault category of the composite fault, i.e. the single fault category included in the composite fault, and each composite fault label yuCorresponding to a compound fault label information vector Gu(ii) a Subscript u represents a compound failure;
s3, the label information definition module first uses the single fault data phis(t) extracting single fault attribute to obtain single fault label information vector GsThen, according to the single fault category contained in the composite fault, obtaining the semantic vector g of the composite faultu(ii) a The details are as follows:
s31, the label information definition module utilizes the single fault data phis(t) extracting the single-fault label information vector GsSingle fault label information vector GsAll the dimensions of (A) are C multiplied by 1; wherein, the single fault label information vector corresponding to the kth single fault category
Figure FDA0003458911160000041
Namely, it is
Figure FDA0003458911160000042
Figure FDA0003458911160000043
Figure FDA0003458911160000044
Single fault label information vector corresponding to k-th single fault category
Figure FDA0003458911160000045
C represents a serial number of a dimension, and C is 1, 2, 3 … C; the superscript K indicates a single fault category number, K being 1, 2, 3 … K; subscript s denotes single fault;
selecting single fault data aiming at the kth single fault category
Figure FDA0003458911160000046
Of (3), specifically
Figure FDA0003458911160000047
Namely, it is
Figure FDA0003458911160000048
Figure FDA0003458911160000049
Representing single fault data corresponding to kth single fault category
Figure FDA00034589111600000410
C represents the serial number of the data point, and C is 1, 2, 3 … C;
single fault data corresponding to the Kth single fault category
Figure FDA0003458911160000051
The maximum value of C data points in (1) is selected as the threshold value mu, i.e.
Figure FDA0003458911160000052
Dividing the amplitude range of the data point into five equal parts according to the threshold value mu, and judging the single fault data of the kth single fault category
Figure FDA0003458911160000053
To obtain the single fault label information vector of the kth single fault category
Figure FDA0003458911160000054
Each dimension data of
Figure FDA0003458911160000055
The values of (A) are as follows:
Figure FDA0003458911160000056
s32, label information definition module according to the incidence relation between the compound fault and the single fault, namely the compound fault is composed of several different single faults, combining the corresponding single fault label information vector GsObtaining a composite fault label information vector Gu(ii) a Composite fault signature informationQuantity GuThe dimensions of (A) are also C × 1;
wherein, the composite fault label information vector is composed of single faults with class serial numbers of 1, … and J
Figure FDA0003458911160000057
As follows:
Figure FDA0003458911160000058
Figure FDA0003458911160000059
superscript 1, …, J refers to the category number of the single fault that constitutes the composite fault, J being 2, 3 …, K;
s4, the feature extraction module is a feature extractor established by a convolutional neural network,
using training set DtrainTraining the feature extractor: feature extractor for single fault image qsExtracting a single fault feature vector vsAnd predicting the single failure image qsPredicted failure label p ofs(ii) a The feature extractor is based on the input single fault image qsActual failure label y ofsAnd predicted failure label psCalculating the classification loss of the feature extractor, and finishing the training of the feature extractor if the classification loss meets the requirement;
after the training of the feature extractor is completed, the feature extractor pairs the test set DtestTo composite failure image q inuExtracting composite fault feature vector vu；
S5, the generation countermeasure module comprises a generator and a discriminator, wherein the generator is used for generating the fault feature vector by taking the fault label information vector G and the Gaussian noise z as conditions
Figure FDA00034589111600000510
The discriminator is used for discriminatingGenerated fault feature vector
Figure FDA0003458911160000061
And the difference between the extracted fault feature vectors v;
using training set DtrainGenerating training reactance module, generator using single fault label information vector GsGeneration of single fault feature vector conditioned on sum of Gaussian noise z
Figure FDA0003458911160000062
The discriminator discriminates the generated single fault feature vector
Figure FDA0003458911160000063
And extracted single fault feature vector vsThe difference between them; generating a countermeasure module based on the generated single fault feature vector
Figure FDA0003458911160000064
And extracted single fault feature vector vsCalculating loss according to the difference between the two, and if the loss meets the requirement, finishing the training of the generation countermeasure module;
after the generation countermeasure module finishes training, the generator respectively uses each composite fault label information vector GuGenerating each composite fault characteristic vector by taking the sum of Gaussian noise z as a condition
Figure FDA0003458911160000065
S6, classifying module pair test set DtestComposite failure image q of (1)uCarrying out a composite fault diagnosis test;
generating each composite fault feature vector by a generator
Figure FDA0003458911160000066
Each central point is used as each central point in the composite fault feature space, and each central point corresponds to one composite fault category;
computing feature extraction modelBlock for test set DtestComposite failure image q of (1)uExtracted composite fault feature vector vuDistance from each central point, selecting and compounding fault characteristic vector vuThe central point with the minimum distance, and the composite fault category corresponding to the central point with the minimum distance is the composite fault image vuThe diagnosis result of (1);
if test set DtestComposite failure image q of (1)uComposite fault label y ofuWith the composite fault image vuIf the diagnosis results are consistent, the composite fault diagnosis is correct; otherwise, a composite fault diagnosis error is indicated.
8. The method for diagnosing composite fault of bearing based on fault label information vector as claimed in claim 7, wherein in step S2,
training set DtrainAnd test set DtestSatisfying the condition in the following formula:
Figure FDA0003458911160000067
wherein, p (phi)s(t))、p(φu(t)) are respectively single-fault data phis(t), composite fault data φu(t) data distribution; i (-) is used to calculate mutual information between two data distributions, I (p (φ)s(t))、p(φu(t))) is the data distribution p (phi)s(t))、p(φu(t)) mutual information between; single fault label ysAnd a composite fault label yuAre disjoint.
9. The method for diagnosing composite fault of bearing based on fault label information vector as claimed in claim 7, wherein in step S4, classification loss L of feature extractor1As follows:
Figure FDA0003458911160000071
wherein, ys(i) Is the ith single failure image q in the single failure categorys(i) The single failure label of (1), i.e., single failure category; i represents the serial number of the single fault image, and i is 1, 2, 3 … N; p is a radical ofs(i) Is that the feature extraction module aims at the ith single fault image q in the single fault categorys(i) A predicted category of predicted faults.
10. The method for diagnosing composite fault of bearing based on fault label information vector as claimed in claim 7, wherein in step S5, a loss function L of countermeasure module is generated2As follows:
Figure FDA0003458911160000072
wherein p isrFor extracted single fault feature vector vsData distribution of pgFor the generated single fault feature vector
Figure FDA0003458911160000073
The data distribution of (2); dw(. to) is a model representation of the discriminator for discriminating between true and false scores of the fault feature vector, and subscript w is discriminator Dw(ii) model parameters of (g);
Figure FDA0003458911160000074
is generator with single fault label information vector GsAnd gaussian noise z,
Figure FDA0003458911160000075
Figure FDA0003458911160000076
for the model representation of the generator, for generating a fault feature vector, the subscript θ2Is other generator
Figure FDA0003458911160000077
The model parameters of (1); v. ofsIs a single fault feature vector extracted by a feature extractor;
Figure FDA0003458911160000078
in order to be a gradient penalty term,
Figure FDA0003458911160000079
is from vsAnd
Figure FDA00034589111600000710
the medium-level uniform sampling obtains a sampling single-fault feature vector,
Figure FDA00034589111600000711
representing sampled single-fault feature vectors
Figure FDA00034589111600000712
The distribution of the data of (a) is,
Figure FDA00034589111600000713
representing a sampled single fault feature vector
Figure FDA00034589111600000714
Derivation is carried out, wherein lambda is a gradient penalty coefficient;
Figure FDA00034589111600000715
is v issAnd
Figure FDA00034589111600000716
mean square error between.
CN202210010510.1A 2022-01-06 2022-01-06 Bearing composite fault diagnosis method based on fault label information vector Active CN114383846B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210010510.1A CN114383846B (en) 2022-01-06 2022-01-06 Bearing composite fault diagnosis method based on fault label information vector

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210010510.1A CN114383846B (en) 2022-01-06 2022-01-06 Bearing composite fault diagnosis method based on fault label information vector

Publications (2)

Publication Number Publication Date
CN114383846A true CN114383846A (en) 2022-04-22
CN114383846B CN114383846B (en) 2023-06-30

Family

ID=81199358

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210010510.1A Active CN114383846B (en) 2022-01-06 2022-01-06 Bearing composite fault diagnosis method based on fault label information vector

Country Status (1)

Country Link
CN (1) CN114383846B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109655259A (en) * 2018-11-23 2019-04-19 华南理工大学 Combined failure diagnostic method and device based on depth decoupling convolutional neural networks
CN110823574A (en) * 2019-09-30 2020-02-21 安徽富煌科技股份有限公司 Fault diagnosis method based on semi-supervised learning deep countermeasure network
WO2021135630A1 (en) * 2019-12-31 2021-07-08 福州大学 Rolling bearing fault diagnosis method based on grcmse and manifold learning
CN113177584A (en) * 2021-04-19 2021-07-27 合肥工业大学 Zero sample learning-based composite fault diagnosis method

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109655259A (en) * 2018-11-23 2019-04-19 华南理工大学 Combined failure diagnostic method and device based on depth decoupling convolutional neural networks
CN110823574A (en) * 2019-09-30 2020-02-21 安徽富煌科技股份有限公司 Fault diagnosis method based on semi-supervised learning deep countermeasure network
WO2021135630A1 (en) * 2019-12-31 2021-07-08 福州大学 Rolling bearing fault diagnosis method based on grcmse and manifold learning
CN113177584A (en) * 2021-04-19 2021-07-27 合肥工业大学 Zero sample learning-based composite fault diagnosis method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
赵晓平;吴家新;钱承山;张永宏;王丽华;: "基于多任务深度学习的齿轮箱多故障诊断方法", 振动与冲击, no. 23 *

Also Published As

Publication number Publication date
CN114383846B (en) 2023-06-30

Similar Documents

Publication Publication Date Title
CN110823574B (en) Fault diagnosis method based on semi-supervised learning deep countermeasure network
Chen et al. Intelligent fault diagnosis for rotary machinery using transferable convolutional neural network
CN107657279B (en) Remote sensing target detection method based on small amount of samples
CN112200244B (en) Intelligent detection method for anomaly of aerospace engine based on hierarchical countermeasure training
CN114048568B (en) Rotary machine fault diagnosis method based on multisource migration fusion shrinkage framework
CN111751133B (en) Intelligent fault diagnosis method of deep convolutional neural network model based on non-local mean embedding
CN113869208B (en) Rolling bearing fault diagnosis method based on SA-ACWGAN-GP
CN114358123B (en) Generalized open set fault diagnosis method based on deep countermeasure migration network
CN116894187A (en) Gear box fault diagnosis method based on deep migration learning
CN114048769A (en) Multi-source multi-domain information entropy fusion and model self-optimization method for bearing fault diagnosis
CN115563553B (en) Aviation bearing fault diagnosis method of deep migration map convolution network under fluctuation working condition
CN113375941A (en) Open set fault diagnosis method for high-speed motor train unit bearing
CN114155397B (en) Small sample image classification method and system
CN114429152A (en) Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption
CN113609569B (en) Distinguishing type generalized zero sample learning fault diagnosis method
CN111310719B (en) Unknown radiation source individual identification and detection method
CN114118138A (en) Bearing composite fault diagnosis method based on multi-label field self-adaptive model
Fu et al. Rolling bearing fault diagnosis in limited data scenarios using feature enhanced generative adversarial networks
CN114548199A (en) Multi-sensor data fusion method based on deep migration network
Chou et al. SHM data anomaly classification using machine learning strategies: A comparative study
CN115221973A (en) Aviation bearing fault diagnosis method based on enhanced weighted heterogeneous ensemble learning
CN114357372A (en) Aircraft fault diagnosis model generation method based on multi-sensor data driving
CN114037001A (en) Mechanical pump small sample fault diagnosis method based on WGAN-GP-C and metric learning
CN111428772A (en) Photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting
CN114383846B (en) Bearing composite fault diagnosis method based on fault label information vector

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