CN114722879A - Bearing cross-working-condition fault prediction method based on anti-migration learning - Google Patents

Bearing cross-working-condition fault prediction method based on anti-migration learning Download PDF

Info

Publication number
CN114722879A
CN114722879A CN202210435858.5A CN202210435858A CN114722879A CN 114722879 A CN114722879 A CN 114722879A CN 202210435858 A CN202210435858 A CN 202210435858A CN 114722879 A CN114722879 A CN 114722879A
Authority
CN
China
Prior art keywords
fault
domain
bearing
migration learning
module
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210435858.5A
Other languages
Chinese (zh)
Inventor
黄承赓
韩瑜
温建棋
门昌昊
刘梓阳
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.)
Sun Yat Sen University
Original Assignee
Sun Yat Sen 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 Sun Yat Sen University filed Critical Sun Yat Sen University
Priority to CN202210435858.5A priority Critical patent/CN114722879A/en
Publication of CN114722879A publication Critical patent/CN114722879A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • 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
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E10/00—Energy generation through renewable energy sources
    • Y02E10/70—Wind energy
    • Y02E10/72—Wind turbines with rotation axis in wind direction

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Signal Processing (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The invention discloses a bearing cross-working condition fault prediction method based on anti-migration learning, which comprises the following steps: setting different working conditions and carrying out signal acquisition on the bearing to obtain a vibration signal; identifying the vibration signal by a continuous abnormal point detection method based on peak value measurement to obtain a health stage signal and a degradation stage signal; preprocessing the signals in the degradation stage and extracting fault characteristics to obtain a fault characteristic set; inputting the fault feature set into a domain confrontation migration learning double-branch neural network and carrying out updating training to obtain a prediction model; and acquiring data to be detected and inputting the data to be detected into the prediction model to obtain a fault prediction result. By using the method, the service life of the ball bearing of the high-speed shaft of the large fan gearbox, which is oriented to the characteristics of actual working conditions, can be accurately predicted under different working conditions. The method for predicting the fault of the bearing across working conditions based on the transfer-resistant learning can be widely applied to the field of engineering component service life prediction.

Description

Bearing cross-working-condition fault prediction method based on anti-migration learning
Technical Field
The invention relates to the field of engineering component service life prediction, in particular to a bearing cross-working condition fault prediction method based on anti-migration learning.
Background
The wind power development prompts the vigorous development of the wind power industry. The gear box is an important mechanical part in a transmission chain of a large-scale wind generating set, has the characteristics of complex structure, low speed, heavy load, long service cycle and the like, and has the frequent failure of key parts (such as an input end front bearing, a high-speed shaft rear bearing, a planet wheel, a sun wheel, a gear ring and the like) under the action of complex variable load and super-strong instantaneous impact. Because the severity of the failure result is obvious, the gear box is always a key part and a weak link of the large-scale fan. The gearbox failure has the typical chain characteristic that a slight failure is easily expanded into a catastrophic failure in a short time, which generally directly causes the shutdown of a large-scale fan, and secondly, the difficulty in repairing the gearbox failure causes longer shutdown time, which causes huge economic loss, thereby remarkably reducing the availability and the economy of the whole life cycle of the large-scale fan.
Due to the influence of factors such as a complex structural form, a severe working condition environment, a fine assembly process and the like of the large-scale wind power gear box, the large-scale wind power gear box is different from a common rotating mechanical system, so that the residual life prediction method facing engineering practice is challenged in multiple aspects: fault characteristics are difficult to effectively extract, residual service life is difficult to stably predict due to strong randomness of a degradation process, implementation steps are limited by expert experience, and the intelligent level of service life prediction research is not high.
Disclosure of Invention
In order to solve the technical problems, the invention aims to provide a bearing cross-working-condition fault prediction method based on countermeasure transfer learning, which has an unsupervised deep transfer learning network with good high precision and domain countermeasures and realizes the cross-working-condition fault prediction of a high-speed shaft ball bearing of a large fan gearbox.
The first technical scheme adopted by the invention is as follows: a bearing cross-working condition fault prediction method based on anti-migration learning comprises the following steps:
setting different working conditions and carrying out signal acquisition on the bearing to obtain a vibration signal;
identifying the vibration signal by a continuous abnormal point detection method based on peak value measurement to obtain a health stage signal and a degradation stage signal;
preprocessing the signals in the degradation stage and extracting fault characteristics to obtain a fault characteristic set;
inputting the fault feature set into a domain anti-migration learning double-branch neural network and performing updating training to obtain a prediction model;
and acquiring data to be detected and inputting the data to be detected into the prediction model to obtain a fault prediction result.
Further, set for different operating modes and carry out signal acquisition to the bearing, obtain this step of vibration signal, it specifically includes:
acquiring a bearing vibration signal when equipment runs to an interception point to obtain target domain fault characteristic data;
the same equipment runs to a fault from a starting point in a controllable environment, and a bearing vibration signal is collected to obtain source domain fault characteristic data;
and integrating the target domain fault characteristic data and the source domain fault characteristic data to obtain a vibration signal.
Further, the continuous abnormal point detection method based on peak value measurement identifies the vibration signal to obtain a healthy stage signal and a degraded stage signal, and specifically includes:
identifying the vibration signal by a continuous abnormal point detection method based on peak value measurement to obtain a state inflection point;
and dividing the vibration signal into a healthy stage and a degraded stage according to the state inflection point to obtain a healthy stage signal and a degraded stage signal.
Further, the step of preprocessing the degradation stage signal and extracting the fault feature to obtain a fault feature set specifically includes:
performing feature extraction on the degradation stage signal based on the one-dimensional time sequence to obtain a first fault feature;
performing feature extraction on the signals in the degradation stage based on the two-dimensional time spectrum to obtain second fault features;
converting the first fault characteristic into a normalized scale and reducing the dimension by adopting bilinear interpolation to obtain a preprocessed first fault characteristic;
and integrating the preprocessed first fault feature and the preprocessed second fault feature to obtain a fault feature set.
Further, the predictive model includes a feature extractor module, a remaining life predictor module, and a domain confrontation module.
Further, the step of inputting the failure feature set into the domain anti-migration learning double-branch neural network and performing update training to obtain the prediction model specifically includes:
respectively inputting the first fault feature and the second fault feature after the fault feature centralized preprocessing into a first channel and a second channel of a domain confrontation migration learning double-branch neural network;
extracting high-level fusion characteristics based on a characteristic extractor and inputting the high-level fusion characteristics to a residual life prediction module and a domain confrontation module;
predicting a remaining life of the bearing based on a remaining life predictor module;
performing domain classification judgment on the high-level fusion features based on a domain confrontation module;
and based on a self-adaptive optimization algorithm, updating training network parameters in combination with real label back propagation iteration to obtain a prediction model.
Further, the remaining life predictor module has the following expression:
Figure BDA0003612766890000021
in the above formula, the first and second carbon atoms are,
Figure BDA0003612766890000031
represents the predicted percentage of remaining useful life, Gp(·；θp) Denotes the module of the predictor of the residual life consisting of the fully connected layers, σ (-) denotes the sigmoid activation function, θpA parameter set representing a remaining life prediction module.
Further, the optimization objective of the iteratively updated training network parameters includes a regression loss of the remaining life predictor module and a domain classification loss of the domain confrontation module.
The method has the beneficial effects that: according to the invention, the dual-channel heterogeneous input of the domain confrontation migration learning double-branch neural network is utilized to improve the transmission fault prediction precision, the problem of obvious distribution difference can be effectively solved through the domain confrontation training in the unsupervised domain adaptation frame, the accurate service life prediction of the large fan gearbox high-speed shaft ball bearing oriented to the actual working condition characteristics under different working conditions is realized, and the method has excellent migration prediction performance and strong robustness and generalization capability.
Drawings
FIG. 1 is a flow chart illustrating the steps of a cross-working-condition fault prediction method for a bearing based on anti-migration learning according to the present invention;
FIG. 2 is a flow chart illustrating a prediction method according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a predictive model according to an embodiment of the invention.
Detailed Description
The invention is described in further detail below with reference to the figures and the specific embodiments. The step numbers in the following embodiments are provided only for convenience of illustration, the order between the steps is not limited at all, and the execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art.
Referring to fig. 1 and 2, the invention provides a bearing cross-working condition fault prediction method based on anti-migration learning, which comprises the following steps:
s1, setting different working conditions and carrying out signal acquisition on the bearing to obtain a vibration signal;
s1.1, collecting bearing vibration signals when equipment runs to an interception point to obtain target domain fault characteristic data;
s1.2, operating the same equipment from a starting point to a fault in a controllable environment, and acquiring a bearing vibration signal to obtain source domain fault characteristic data;
and S1.3, integrating the target domain fault characteristic data and the source domain fault characteristic data to obtain a vibration signal.
Specifically, the data in the source domain is the bearing vibration signal collected when similar equipment is operated from a starting point to a fault in a controlled laboratory environment, and the data in the target domain is the vibration signal collected when the equipment is operated to an interception point, and is small sample monitoring data with few labels or no labels.
S2, identifying the vibration signal based on a continuous abnormal point detection method of peak value measurement to obtain a health stage signal and a degradation stage signal;
s2.1, identifying the vibration signal by a continuous abnormal point detection method based on peak value measurement to obtain a state inflection point;
and S2.2, dividing the vibration signal into a health stage and a degradation stage according to the state inflection point to obtain a health stage signal and a degradation stage signal.
Specifically, a First Prediction Time (FPT) of the transferable fault prediction method is further determined, an FPT post-trigger fault prediction algorithm is determined to predict the remaining life (RUL) of the bearing in the target domain, and the FPT discriminant function is expressed as:
{|kt-j-μ|＞3σ}i＝0,1,2,3,4
wherein k ist-jThe sequence of raw peak metrics denoted (t-j), μ and σ denote the mean and variance of the peak metrics, respectively, i ═ 0,1,2,3,4 denotes 5 consecutive monitoring points, and the time parameter t is defined as the FPT triggering the fault prediction algorithm.
S3, preprocessing the signals at the degradation stage and extracting fault characteristics to obtain a fault characteristic set;
s3.1, extracting the characteristics of the signals in the degradation stage based on the one-dimensional time sequence to obtain first fault characteristics;
s3.2, extracting the characteristics of the signals in the degradation stage based on the two-dimensional time spectrum to obtain second fault characteristics;
specifically, the first input channel is a fault feature based on a one-dimensional time series extracted from the time and frequency domains of the original vibration signal. The second input channel is based on the failure characteristics of two-dimensional Time Spectrums (TFRs), which is a two-dimensional coefficient matrix obtained by converting a one-dimensional time sequence through a continuous wavelet transform algorithm. These two-channel heterogeneous fault signatures are extracted from the original vibrator sequence between the FPT point and the end point. In the source domain is the fault signature with the RUL label whose extraction end point is the end-of-life (EoL), while in the target domain is the fault signature with a small number of RUL labels or no labels whose extraction end point is the truncation point of the bearing.
S3.3, converting the first fault characteristic into a normalized scale and reducing dimensions by adopting bilinear interpolation to obtain a preprocessed first fault characteristic;
specifically, Max-Min normalization is utilized to convert the fault characteristics of the first input channel into a normalized scale, bilinear interpolation is adopted to reduce the high dimension of the original TFRs, and the adverse effect of signal difference is eliminated.
And S3.4, integrating the preprocessed first fault feature and the preprocessed second fault feature to obtain a fault feature set.
In particular, the amount of the solvent to be used,
Figure BDA0003612766890000041
representation containing training samples NSSource field of (x)SIs the feature space of the source domain,
Figure BDA0003612766890000042
Figure BDA0003612766890000043
representing a RUL label value corresponding to an ith training sample in the source domain;
Figure BDA0003612766890000044
indicates that the test sample N is containedTThe target domain of (a) is selected,
Figure BDA0003612766890000045
is expressed in degree XTThe jth test sample of the target feature space of (1), but the corresponding RUL tag value is not available.
Specifically, the first fault feature and the second fault feature are associated with the source domain fault feature and the target fault feature, the source domain fault feature is divided into one-dimensional and two-dimensional, that is, the source domain fault feature includes fault features of a dimensional time sequence and fault features of a two-dimensional time frequency spectrum, and the target fault feature is the same.
S4, inputting the fault feature set into a domain confrontation migration learning double-branch neural network and performing update training to obtain a prediction model;
s4.1, inputting the first fault feature and the second fault feature subjected to the centralized preprocessing of the fault features into a first channel and a second channel of a domain antagonistic migration learning double-branch neural network respectively;
specifically, two-channel heterogeneous original fault characteristics, namely source domain fault characteristics
Figure BDA0003612766890000051
And target Domain Fault features
Figure BDA0003612766890000052
Input to the domain antagonistic migration learning dual-branch neural network (DBNN-DA), referring to FIG. 3, the DBNN-DA network model comprises three modules: a feature extractor module, a RUL predictor module, and a domain confrontation module.
S4.2, extracting high-level fusion characteristics based on the characteristic extractor and inputting the high-level fusion characteristics to the residual life prediction module and the domain confrontation module;
specifically, a feature extractor module of the DBNN-DA network model automatically extracts high-level fusion features from two-channel heterogeneous original fault features in a marked source domain and an unmarked target domain, and outputs:
hFL＝Gf(xS_1D,xS_2D,xT_1D,xT_2D；θf)
in the above formula, hFLRepresenting extracted high-level fusion features, Gf(·；θf) Representing feature extractor modules, i.e. a two-branch network, thetafIs a set of parameters for the feature extractor.
S4.3, predicting the residual life of the bearing based on a residual life predictor module;
specifically, the expression of the remaining life predictor module is as follows:
Figure BDA0003612766890000053
in the above formula, the first and second carbon atoms are,
Figure BDA0003612766890000054
represents the predicted percentage of remaining useful life, Gp(·；θp) Represents a residual life predictor module consisting of a fully connected layer, sigma (-) represents a sigmoid activation function, thetapA parameter set representing a remaining life prediction module.
S4.4, performing domain classification judgment on the high-grade fusion features based on the domain confrontation module;
in particular, advanced fusion features are simultaneously input into the domain confrontation module to enable unsupervised domain adaptation. The domain confrontation module is composed of a domain classifier, and can judge whether a certain sample is from a source domain or a target domain, wherein the domain classifier is defined as:
Figure BDA0003612766890000055
in the above formula, Gd(·；θd) Is a domain classifier composed of several fully-connected layers, the corresponding last layer is a binary classification layer, thetadIs a set of parameters for the domain classifier, γ (-) represents the logarithmic softmax activation function,
Figure BDA0003612766890000056
is the predicted domain label (source domain: 0, target domain: 1).
And S4.5, based on a self-adaptive optimization algorithm, iteratively updating the training network parameters by combining with real label back propagation to obtain a prediction model.
Specifically, the optimization objective of the iterative update training network parameters comprises the regression loss L of the residual life predictor modulepAnd domain antagonismDomain classification penalty L for modulesd。
Minimizing the first loss term LpFeature extractor Gf(·；θf) And RUL predictor Gp(·；θp) Is jointly optimized to minimize the empirical regression loss of the source domain samples and ensure the high-level fusion characteristics hFLAnd the feature extractor and the RUL predictor have overall good RUL prediction accuracy in the source domain. Loss function L defined as mean square errorpComprises the following steps:
Figure BDA0003612766890000061
in the above formula, NbatchIs the batch size of the marked samples in the source domain, yiAnd
Figure BDA0003612766890000062
actual RULP labels and predicted RULP values, respectively.
Maximizing the second loss term LdBecause it is desirable that the high-level fused feature domain is invariant, thereby making the domain classifier unable to distinguish samples between the source domain and the target domain, this process is called domain confrontation training, whose basic idea is to align distributions in a confrontational manner using the domain classifier and the feature extractor. The domain classifier is used to distinguish domain labels of the high-level fused features generated by the feature extractor, which is trained to fool the domain classifier. Loss function L defined as cross entropy lossdComprises the following steps:
Figure BDA0003612766890000063
in the above formula, /)iIs the actual domain label that is used,
Figure BDA0003612766890000064
is a predicted domain label generated by the domain classifier.
The final optimization objective function of the DBNN-DA model is summarized as follows:
Figure BDA0003612766890000065
in the above formula, λ is control LpAnd LdA non-negative over-parameter of the trade-off between,
Figure BDA0003612766890000066
and
Figure BDA0003612766890000067
is the corresponding loss function at the ith training sample.
Parameter set by finding saddle points of an objective function
Figure BDA0003612766890000068
And (5) realizing optimization.
Figure BDA0003612766890000069
Figure BDA00036127668900000610
At the saddle point, the parameter set θ of the domain classifierdMinimizing domain classification loss (with a minus sign in the optimization objective function), parameter set θ for RUL predictorpMinimizing the RUL regression loss, parameter set θ for feature extractorfThe RUL regression loss is minimized while the domain classification loss is maximized.
With Adam optimizer, the network parameter set Θ ═ θ (θ)f,θp,θd) The updating is as follows:
Figure BDA0003612766890000071
Figure BDA0003612766890000072
Figure BDA0003612766890000073
where epsilon represents the learning rate.
And S5, acquiring the data to be detected and inputting the data to be detected into the prediction model to obtain a fault prediction result.
Specifically, real-time detection data of the ball bearing of the high-speed shaft of the large fan gearbox under different working conditions in the target domain are input into a trained DBNN-DA network model, and accurate online fault prediction is carried out.
A bearing cross-working condition fault prediction device based on antagonistic transfer learning comprises:
at least one processor;
at least one memory for storing at least one program;
when executed by the at least one processor, cause the at least one processor to implement a method of cross-regime bearing fault prediction based on anti-migration learning as described above.
The contents in the above method embodiments are all applicable to the present apparatus embodiment, the functions specifically implemented by the present apparatus embodiment are the same as those in the above method embodiments, and the advantageous effects achieved by the present apparatus embodiment are also the same as those achieved by the above method embodiments.
A storage medium having stored therein instructions executable by a processor, the storage medium comprising: the processor-executable instructions, when executed by the processor, are for implementing a cross-regime bearing fault prediction method based on anti-migration learning, as described above.
The contents in the above method embodiments are all applicable to the present storage medium embodiment, the functions specifically implemented by the present storage medium embodiment are the same as those in the above method embodiments, and the advantageous effects achieved by the present storage medium embodiment are also the same as those achieved by the above method embodiments.
While the preferred embodiments of the present invention have been illustrated and described, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims (8)

1. A bearing cross-working condition fault prediction method based on anti-migration learning is characterized by comprising the following steps:
setting different working conditions and carrying out signal acquisition on the bearing to obtain a vibration signal;
identifying the vibration signal by a continuous abnormal point detection method based on peak value measurement to obtain a health stage signal and a degradation stage signal;
preprocessing the signals in the degradation stage and extracting fault characteristics to obtain a fault characteristic set;
inputting the fault feature set into a domain confrontation migration learning double-branch neural network and carrying out updating training to obtain a prediction model;
and acquiring data to be detected and inputting the data to be detected into the prediction model to obtain a fault prediction result.
2. The method for predicting the fault of the bearing across the working conditions based on the anti-migration learning as claimed in claim 1, wherein the step of setting different working conditions and carrying out signal acquisition on the bearing to obtain a vibration signal specifically comprises:
acquiring a bearing vibration signal when equipment runs to an interception point to obtain target domain fault characteristic data;
the same equipment is operated from a starting point to a fault in a controllable environment, and a bearing vibration signal is collected to obtain source domain fault characteristic data;
and integrating the fault characteristic data of the target domain and the fault characteristic data of the source domain to obtain a vibration signal.
3. The method for predicting the fault of the bearing across the working conditions based on the anti-migration learning as claimed in claim 2, wherein the method for detecting the continuous abnormal point based on the peak value measurement is used for identifying the vibration signal to obtain the healthy stage signal and the degraded stage signal, and specifically comprises the following steps:
identifying the vibration signal by a continuous abnormal point detection method based on peak value measurement to obtain a state inflection point;
and dividing the vibration signal into a healthy stage and a degraded stage according to the state inflection point to obtain a healthy stage signal and a degraded stage signal.
4. The method for predicting the fault of the bearing across the working conditions based on the anti-migration learning as claimed in claim 3, wherein the step of preprocessing the degradation stage signal and extracting the fault feature to obtain the fault feature set specifically comprises:
performing feature extraction on the degradation stage signal based on the one-dimensional time sequence to obtain a first fault feature;
performing feature extraction on the signals at the degradation stage based on the two-dimensional time spectrum to obtain second fault features;
converting the first fault characteristic into a normalized scale and reducing dimensions by adopting bilinear interpolation to obtain a preprocessed first fault characteristic;
and integrating the preprocessed first fault feature and the preprocessed second fault feature to obtain a fault feature set.
5. The method for predicting the fault of the bearing across the working conditions based on the migration learning resisting method is characterized in that the prediction model comprises a feature extractor module, a residual life predictor module and a domain resisting module.
6. The method for predicting the bearing cross-working-condition fault based on the anti-migration learning as claimed in claim 5, wherein the step of inputting the fault feature set into the domain anti-migration learning double-branch neural network and performing update training to obtain the prediction model specifically comprises:
respectively inputting the first fault feature and the second fault feature after the fault feature centralized preprocessing into a first channel and a second channel of a domain confrontation migration learning double-branch neural network;
extracting high-grade fusion characteristics based on the characteristic extractor and inputting the high-grade fusion characteristics to a residual life prediction module and a domain confrontation module;
predicting the residual life of the bearing based on a residual life predictor module;
performing domain classification judgment on the high-level fusion features based on a domain confrontation module;
and based on a self-adaptive optimization algorithm, updating training network parameters in combination with real label back propagation iteration to obtain a prediction model.
7. The method for predicting the fault of the bearing across working conditions based on the migration learning resistance as claimed in claim 6, wherein the expression of the residual life predictor module is as follows:
Figure FDA0003612766880000021
in the above formula, the first and second carbon atoms are,
Figure FDA0003612766880000022
represents the predicted percentage of remaining useful life, Gp(·；θp) Represents a residual life predictor module consisting of a fully connected layer, sigma (-) represents a sigmoid activation function, thetapA parameter set representing a remaining life prediction module.
8. The method of claim 7, wherein the optimization objectives of the iteratively updated training network parameters include a regression loss of the remaining life predictor module and a domain classification loss of the domain countermeasure module.
CN202210435858.5A 2022-04-24 2022-04-24 Bearing cross-working-condition fault prediction method based on anti-migration learning Pending CN114722879A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210435858.5A CN114722879A (en) 2022-04-24 2022-04-24 Bearing cross-working-condition fault prediction method based on anti-migration learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210435858.5A CN114722879A (en) 2022-04-24 2022-04-24 Bearing cross-working-condition fault prediction method based on anti-migration learning

Publications (1)

Publication Number Publication Date
CN114722879A true CN114722879A (en) 2022-07-08

Family

ID=82245267

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210435858.5A Pending CN114722879A (en) 2022-04-24 2022-04-24 Bearing cross-working-condition fault prediction method based on anti-migration learning

Country Status (1)

Country Link
CN (1) CN114722879A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116821694A (en) * 2023-08-30 2023-09-29 中国石油大学(华东) Soil humidity inversion method based on multi-branch neural network and segmented model
GB2622124A (en) * 2022-09-01 2024-03-06 Univ Chongqing Intelligent mechanical fault diagnosis method based on relationship transfer domain generalization network (RTDGN)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB2622124A (en) * 2022-09-01 2024-03-06 Univ Chongqing Intelligent mechanical fault diagnosis method based on relationship transfer domain generalization network (RTDGN)
CN116821694A (en) * 2023-08-30 2023-09-29 中国石油大学(华东) Soil humidity inversion method based on multi-branch neural network and segmented model
CN116821694B (en) * 2023-08-30 2023-12-01 中国石油大学(华东) Soil humidity inversion method based on multi-branch neural network and segmented model

Similar Documents

Publication Publication Date Title
Zan et al. Control chart pattern recognition using the convolutional neural network
Kumar et al. Novel convolutional neural network (NCNN) for the diagnosis of bearing defects in rotary machinery
CN109635677B (en) Compound fault diagnosis method and device based on multi-label classification convolutional neural network
Pandhare et al. Convolutional neural network based rolling-element bearing fault diagnosis for naturally occurring and progressing defects using time-frequency domain features
Zhang et al. Instance-based ensemble deep transfer learning network: A new intelligent degradation recognition method and its application on ball screw
CN114722879A (en) Bearing cross-working-condition fault prediction method based on anti-migration learning
CN114048600A (en) Digital twin-driven multi-model fusion industrial system anomaly detection method
Thoppil et al. Deep learning algorithms for machinery health prognostics using time-series data: A review
CN113033309A (en) Fault diagnosis method based on signal downsampling and one-dimensional convolution neural network
CN112883994A (en) Rotating machinery variable working condition fault diagnosis method with balanced distribution adaptation
CN115127814A (en) Unsupervised bearing fault diagnosis method based on self-adaptive residual error countermeasure network
Chang et al. One-dimensional fully decoupled networks for fault diagnosis of planetary gearboxes
CN116010900A (en) Multi-scale feature fusion gearbox fault diagnosis method based on self-attention mechanism
Zhang et al. Intelligent machine fault diagnosis using convolutional neural networks and transfer learning
CN114118138A (en) Bearing composite fault diagnosis method based on multi-label field self-adaptive model
CN117009916A (en) Actuator fault diagnosis method based on multi-sensor information fusion and transfer learning
CN113835964B (en) Cloud data center server energy consumption prediction method based on small sample learning
CN115217534A (en) Method and system for monitoring service quality state of steam turbine
CN115452376A (en) Bearing fault diagnosis method based on improved lightweight deep convolution neural network
CN114358075A (en) Data-driven rotor system typical fault automatic identification method
Zhou et al. Research on federated learning method for fault diagnosis in multiple working conditions
Li et al. Comparison of data-driven prognostics models: A process perspective
CN113158769A (en) CEEMDAN and FastICA-based electromechanical device bearing vibration signal denoising method
CN113723592A (en) Fault diagnosis method based on wind power gear box monitoring system
Chakrapani et al. Health Monitoring of Dry Clutch System Using Deep Learning Approach.

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