CN110942101A - Rolling bearing residual life prediction method based on depth generation type countermeasure network - Google Patents

Rolling bearing residual life prediction method based on depth generation type countermeasure network Download PDF

Info

Publication number
CN110942101A
CN110942101A CN201911201619.8A CN201911201619A CN110942101A CN 110942101 A CN110942101 A CN 110942101A CN 201911201619 A CN201911201619 A CN 201911201619A CN 110942101 A CN110942101 A CN 110942101A
Authority
CN
China
Prior art keywords
rolling bearing
prediction
predicting
sequence
network
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
CN201911201619.8A
Other languages
Chinese (zh)
Other versions
CN110942101B (en
Inventor
刘朝华
陆碧良
王畅通
张红强
吴亮红
陈磊
李小花
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.)
Hunan University of Science and Technology
Original Assignee
Hunan University of Science and 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 Hunan University of Science and Technology filed Critical Hunan University of Science and Technology
Priority to CN201911201619.8A priority Critical patent/CN110942101B/en
Publication of CN110942101A publication Critical patent/CN110942101A/en
Application granted granted Critical
Publication of CN110942101B publication Critical patent/CN110942101B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a method for predicting the residual life of a rolling bearing based on a deep generation type countermeasure network, which comprises the following steps: collecting an original vibration signal of a rolling bearing; acquiring characteristic parameters of an original vibration signal; dividing the characteristic parameters into a training set and a prediction set; sending the training set into a generator long and short term memory network for training; predicting the degradation process of the rolling bearing and generating a prediction result; an automatic encoder model is built to serve as a discriminator, and whether a prediction result is from real historical data or not is judged; a long-term and short-term memory network and an automatic encoder of a discriminator are generated to carry out countermeasure training and seek an optimal solution; and outputting the prediction result of the residual service life of the rolling bearing. According to the method, the degradation process of the rolling bearing is predicted through the long-term and short-term memory network learning, the prediction result of the long-term and short-term memory network is judged by using the automatic encoder, and the long-term and short-term memory network and the automatic encoder resist learning until the precision requirement is met, so that the problem of prediction error superposition of the traditional method is reduced, and the prediction accuracy is improved.

Description

Rolling bearing residual life prediction method based on depth generation type countermeasure network
Technical Field
The invention relates to the field of fault diagnosis and bearing life prediction, in particular to a rolling bearing residual life prediction method based on a deep generation type countermeasure network.
Background
Rolling bearings play an important role in rotating machines such as high-speed railways, induction motors, and wind turbine transmission systems. The harsh working environment of the rotary machine enables the rolling bearing to be exposed to the high-low temperature, high-pressure and humid working environment, and the rolling bearing can be damaged quickly. Any failure of the rolling bearing leads to a failure of the entire machine, which leads to high maintenance costs. If a failure of the rolling bearing can be predicted in advance, a stoppage of the entire rotary machine due to the failure of the bearing can be avoided. Therefore, prediction of the remaining service life (RUL) of the rolling bearing has been receiving increasing attention in recent years.
The remaining service life prediction methods of rolling bearings can be classified into two categories: model-based methods and data-driven based methods. For model-based methods, describing the degradation process of a machine requires the creation of a mathematical or physical model and the determination of model parameters in conjunction with measured data. In addition, the model-based RUL prediction method can combine expert knowledge with actual operation data. Therefore, this method has good performance in predicting RUL of the rolling bearing. However, the conventional method for predicting the RUL of the rolling bearing based on the model mainly has two defects: 1) if the physical or mathematical model of the rolling bearing drive train is not sufficiently simplified, and even the mechanical principles of the rotating machine itself are an extremely complex system, then the model-based RUL prediction method may fail in such cases. 2) The RUL prediction method based on the model has better prediction performance; different rolling bearings need to be modeled separately. For rolling bearings where existing models exist, the cost is acceptable, but for rolling bearings where no model is available, the cost of reconstructing a mathematical or physical model is very expensive.
Data-driven methods tend to infer the Rolling Bearing Degradation Process (RBDP) from historical data through deep learning or machine learning techniques. Therefore, the data-driven RUL prediction method is mainly affected by two aspects: 1) the quantity and quality of the historical actual data. 2) The predictive performance of deep learning or machine learning models. In recent years, with the rapid development of sensor technologies, especially deep learning technologies, accurate and reliable sensors can acquire high-quality and sufficient amount of data from complex practical applications, and deep learning methods such as Deep Belief Networks (DBNs) and long-term memory networks (LSTM) can acquire RBDP by using the acquired data. The data-driven approach may incorporate both measurement information and intelligent deep learning models. Therefore, data-driven RUL prediction methods are gradually becoming a trend.
Data-driven RUL prediction methods typically employ a single-step time-series prediction or a direct remaining life prediction. The former requires that the entire RBDP be known in advance and each predicted point be obtained from the measurement data. The latter only requires knowledge of the first half of the RBDP, which is inferred from the predictive power of the deep learning method. Obviously, the data-driven RUL prediction method based on single-step time series prediction is simpler and more accurate. When the entire RBDP is available, a single step time series prediction method may be beneficial. However, this depends on whether the entire RBDP is available or not, which is challenging for practical applications. On the contrary, the direct remaining life prediction method is more suitable for actual working conditions.
The direct RUL prediction method refers to predicting one or several points from the historical data, then adding the predicted points to the historical data to predict the next part, and then cycling to get the whole RBDP. Due to this feature, the direct RUL prediction method is more available and challenging than the single-step time-series prediction. However, this method has a disadvantage that is still not solved-the problem of superposition of prediction errors means that the previous prediction errors will be accumulated in the next prediction. This is caused by the method itself, and it also limits the application of the data driven method.
Recently, deep learning techniques, particularly the generation of countermeasure networks (GANs), have attracted considerable attention and have been successfully applied in many fields. The GAN is usually composed of a generator and a discriminator, and the discriminator is trained to achieve nash equalization. The generator attempts to capture the data distribution and the arbiter estimates the probability. The two resist each other and utilize this mechanism to improve the performance of GAN. The core idea of a generator and discriminator based GAN network can be used to solve the prediction error superposition problem described above.
Disclosure of Invention
In order to solve the technical problem, the invention provides a rolling bearing residual life prediction method based on a deep-generation countermeasure network, which can reduce superposition errors and improve prediction accuracy.
The technical scheme for solving the problems is as follows: a method for predicting the residual life of a rolling bearing based on a deep generation type countermeasure network comprises the following steps:
1) collecting an original vibration signal of a rolling bearing;
2) acquiring characteristic parameters of an original vibration signal;
3) dividing the characteristic parameters obtained in the step 2) into a training set and a prediction set;
4) sending the training set into a generator long-short term memory network LSTM for training;
5) predicting the rolling bearing degradation process RBDP in a direct prediction mode and generating a prediction result;
6) building an automatic encoder model AE as a discriminator, and judging whether the prediction result obtained in the step 5) is from real historical data;
7) a long-term and short-term memory network and a discriminator AE are generated to carry out countermeasure training, and an optimal solution is sought by adopting a random gradient descent method;
8) and outputting the prediction results of the residual life of the rolling bearing, namely the results after LSTM and AE antagonistic training is finished.
In the method for predicting the remaining life of the rolling bearing based on the depth generating type countermeasure network, in the step 1), the original vibration signals of the rolling bearing comprise a horizontal vibration signal and a vertical vibration signal, and the horizontal vibration signal and the vertical vibration signal are respectively measured by an accelerometer.
In the method for predicting the residual life of the rolling bearing based on the depth generating countermeasure network, in the step 2), the original vibration signal is preprocessed to obtain characteristic parameters, wherein the characteristic parameters comprise a root mean square value, a standard deviation, a peak value and an average value.
In the method for predicting the remaining life of the rolling bearing based on the depth-generating countermeasure network, in the step 3), the statistical characteristic parameter is considered to be a time series X ═ { X } containing n samples1,x2,...,xm,...,xnThe first half is used for training the model, called training set
Figure BDA0002296020510000043
The second half is a prediction part, called prediction set
Figure BDA0002296020510000044
In the method for predicting the remaining life of the rolling bearing based on the deep generation type countermeasure network, in the step 4), the LSTM controls the transmission of signals through the storage unit and the cell state of the combination of the input gate, the forgetting gate and the output gate, so as to learn the long-term dependence in the time sequence;
the forward propagation algorithm for LSTM is as follows:
it＝σ(Wi[ht-1,xt]+bi)
ft＝σ(Wf[ht-1,xt]+bf)
ot＝σ(Wo[ht-1,xt]+bo)
gt＝σ(Wg[ht-1,xt]+bg)
Figure BDA0002296020510000041
Figure BDA0002296020510000042
wherein it、ftAnd otRespectively showing an input gate, a forgetting gate and an output gate, gtRepresents new candidate cell information, ht-1And htRespectively representing the output, x, of the previous and current celltRepresenting the input of the current cell, Wi、Wf、WoAnd WgThe weight matrix of the input gate, the forgetting gate, the output gate and the new candidate cell information of LSTM, respectively, bi、bf、boAnd bgOffsets for input gate, forget gate, output gate and new candidate cell information, respectively, ctAnd ct-1Respectively representing the current state and the last state of the cell, sigma and tanh are sigmoid and hyperbolic activation functions respectively,
Figure BDA0002296020510000051
the dot product is represented.
In the method for predicting the remaining life of the rolling bearing based on the depth-generating countermeasure network, in the step 5), the direct prediction mode is to add the predicted point into the original sequence to predict the next point, and the specific mode is as follows:
Figure BDA0002296020510000052
wherein, x is the history data,
Figure BDA0002296020510000057
is the prediction data.
In the method for predicting the remaining life of the rolling bearing based on the depth-generating countermeasure network, in step 6), the automatic encoder AE is composed of a three-layer neural network, which jointly form an encoding layer and a decoding layer, the encoding layer compresses an original input sequence into typical features by reducing the number of neurons, then the typical features are converted into a reconstructed sequence with the same dimension as the input sequence by a decoder, and the purpose of the AE is to reconstruct the input sequence by minimizing the error between the reconstructed sequence and the input sequence;
AE is mapped h by a certaintyθInput sequence of dimension n
Figure BDA0002296020510000053
Typical features mapping to dimension m
Figure BDA0002296020510000054
(m<n), the expression is as follows:
y＝hθ(x)＝f(Wx+b)
wherein the content of the first and second substances,
Figure BDA0002296020510000055
theta is the overall parameter of the coding layer,
Figure BDA0002296020510000056
the expression 'is defined as', f is the activation function of the coding layer, W represents the weight matrix of m multiplied by n, and b represents the offset vector of the coding part;
accordingly, the decoding part restores the characteristic features to a sequence of the same dimension as the input sequence x', as follows:
x′＝gθ′(y)＝f(W′y+b′)
wherein the content of the first and second substances,
Figure BDA0002296020510000061
theta ' is all parameters of the decoding layer, g is the activation function of the decoding layer, W ' is a weight matrix of n multiplied by m, and b ' is a bias vector of the decoding part;
the entire AE network then updates the weight matrix W, W 'and the offsets b, b' by the following loss functions:
Figure BDA0002296020510000062
wherein the content of the first and second substances,
Figure BDA0002296020510000069
is a function of the loss as a function of,
Figure BDA00022960205100000610
θ*and theta'*For the final optimization parameter, x, of the AE networks theta and theta(i)And x'(i)The input sequence of the coding layer and the decoding layer are respectively.
In the method for predicting the remaining life of the rolling bearing based on the deep-generation countermeasure network, in the step 6), in the AE network, the softmax regression is placed at the end of the structure, and typical features extracted by AE are classified:
given a sample x comprising q samples(i)Input sequence of
Figure BDA0002296020510000063
Wherein
Figure BDA0002296020510000064
Including a set of k-class labels
Figure BDA0002296020510000065
y(i)E {1, 2.... k }, the role of Softmax is to estimate the probability that each sample belongs to each class, and to take the class with the highest probability as the class of the sample, which is given by:
Figure BDA0002296020510000066
wherein the content of the first and second substances,
Figure BDA0002296020510000067
θ∈{θ1,θ2,......,θkthe parameters of the model are determined,
Figure BDA0002296020510000068
is a hypothesis function that adjusts the probability distribution;
the cost function of the Softmax regression model is:
Figure BDA0002296020510000071
where 1{ } is an indicative function where 1{ an expression whose value is true } ═ 1, and 1{ an expression whose value is false } -, 0;
softmax regression by minimization
Figure BDA0002296020510000072
And classifying the typical features extracted by the AE.
In the method for predicting the remaining life of the rolling bearing based on the deep-generation countermeasure network, in the step 7), the LSTM passes through a training set
Figure BDA00022960205100000713
Generating a predicted sequence
Figure BDA0002296020510000073
Arbiter AE compressed prediction sequence
Figure BDA0002296020510000074
And generates a corresponding label
Figure BDA0002296020510000075
Means that
Figure BDA0002296020510000076
The prediction requirements are met;
while training set
Figure BDA00022960205100000715
When known, the prediction set
Figure BDA00022960205100000714
The conditional probability of (c) is defined as follows:
Figure BDA0002296020510000077
in order to minimize the prediction error superposition problem, the above conditional probability is used to obtain the optimal parameters in consideration of the internal parameter θ, and the following formula is adopted:
Figure BDA0002296020510000078
wherein θ means all parameters of the model, including weight, bias;
for the discriminator AE, the aim is to train a classifier for encoding the input sequence and mapping the input features to the output
Figure BDA0002296020510000079
To represent
Figure BDA00022960205100000710
From the real RBDP, not from the generative model LSTM, then the objective of the arbiter is functionalized as a penalty as follows:
Figure BDA00022960205100000711
where m represents the number of samples in the input sequence and y(i)And
Figure BDA00022960205100000712
respectively representing a real label and a discrimination label;
the goal of the training process is to minimize the log-likelihood, as follows:
Figure BDA0002296020510000081
wherein the content of the first and second substances,
Figure BDA0002296020510000082
the mathematical expectation is represented by the mathematical expectation,
Figure BDA00022960205100000810
respectively, the generation sequence of the generator and the discrimination probability of the discriminator.
The rolling bearing based on the deep generation type countermeasure network has the residual lifeA life prediction method, said step 7) of an AE feature vector with a dimension of n-m
Figure BDA0002296020510000083
SGD optimization model generation by random gradient descent method and AE feature vector
Figure BDA0002296020510000084
And rolling bearing degradation process X ═ X1,x2,...,xm,...,xnThe SGD optimization discriminant model is used, and the random gradients of the two optimization processes are as follows:
Figure BDA0002296020510000085
Figure BDA0002296020510000086
wherein the content of the first and second substances,
Figure BDA0002296020510000087
meaning that the parameters of discriminator D are derived, and, similarly,
Figure BDA0002296020510000088
meaning that the parameters of the generator G are differentiated,
Figure BDA00022960205100000812
representing a training set of discriminator pairs
Figure BDA00022960205100000811
The discrimination probability of (1).
This process is repeated until the AE input feature vector
Figure BDA0002296020510000089
Until the predicted performance requirements are met.
The invention has the beneficial effects that:
1) the invention provides a depth generation type countermeasure network for predicting the residual service life of a rolling bearing on the basis of analyzing the superposition problem of prediction errors. The method uses an LSTM network as a generator to predict the degradation process of a rolling bearing. The LSTM network has the capability of mining long-term historical data rules, and is therefore suitable for the degradation process of the rolling bearing. Then, the automatic encoder discriminates whether the input RBDP belongs to the measurement data or the prediction data, and both of them compete against each other in the course of the cycle. The mechanism can correct the prediction curve, reduce the superposition problem of prediction errors and obviously improve the prediction precision of the LSTM.
2) The method for predicting the residual life of the rolling bearing based on the deep generation type countermeasure network, which is provided by the invention based on a direct prediction mode, can train the network by using limited bearing fault historical data. This method does not need to know all the historical data, but only a part of the previous historical data, and can deduce the following data through the prediction ability of the LSTM and correct the prediction error through AE, and can complete the prediction even in the case of data shortage.
Drawings
FIG. 1 is a flow chart of the present invention.
Fig. 2 is a schematic diagram of a prediction error superposition process.
Fig. 3 is a schematic diagram of the basic storage unit of the LSTM network.
FIG. 4 is a diagram illustrating an automatic encoding process of an input sequence in a discriminant model.
FIG. 5 is a graph showing the comparison of the prediction method of the present invention with both the SVR and LSTM methods.
Detailed Description
The invention is further described below with reference to the figures and examples.
As shown in fig. 1, a method for predicting the remaining life of a rolling bearing based on a deep generative countermeasure network includes the following steps:
1) and collecting an original vibration signal of the rolling bearing.
The original vibration signals of the rolling bearing comprise a horizontal vibration signal and a vertical vibration signal which are respectively measured by an accelerometer.
2) And acquiring characteristic parameters of the original vibration signal.
Preprocessing an original vibration signal, and solving characteristic parameters, wherein the characteristic parameters comprise a root mean square value, a standard deviation, a peak value and an average value.
3) Dividing the characteristic parameters obtained in the step 2) into a training set and a prediction set.
Considering that the statistical characteristic parameter is a time series X ═ X containing n samples1,x2,...,xm,...,xnThe first half is used for training the model, called training set
Figure BDA0002296020510000104
The second half is a prediction part, called prediction set
Figure BDA0002296020510000105
4) The training set is sent to a generator long-short term memory network LSTM for training.
The LSTM controls the transfer of signals through the combined memory cell and cell states of the input gate, the forgetting gate and the output gate, thereby learning the long-term dependence in the time series, as shown in fig. 2;
the forward propagation algorithm for LSTM is as follows:
it＝σ(Wi[ht-1,xt]+bi)
ft＝σ(Wf[ht-1,xt]+bf)
ot＝σ(Wo[ht-1,xt]+bo)
gt＝σ(Wg[ht-1,xt]+bg)
Figure BDA0002296020510000101
Figure BDA0002296020510000102
wherein it、ftAnd otRespectively showing an input gate, a forgetting gate and an output gate, gtRepresents new candidate cell information, ht-1And htRespectively representing the output, x, of the previous and current celltRepresenting the input of the current cell, Wi、Wf、WoAnd WgThe weight matrix of the input gate, the forgetting gate, the output gate and the new candidate cell information of LSTM, respectively, bi、bf、boAnd bgOffsets for input gate, forget gate, output gate and new candidate cell information, respectively, ctAnd ct-1Respectively representing the current state and the last state of the cell, sigma and tanh are sigmoid and hyperbolic activation functions respectively,
Figure BDA0002296020510000103
the dot product is represented.
5) And predicting the degradation process RBDP of the rolling bearing by adopting a direct prediction mode and generating a prediction result.
The direct prediction method is to add a predicted point to the original sequence to predict the next point. The specific mode is as follows:
Figure BDA0002296020510000111
wherein, x is the history data,
Figure BDA00022960205100001110
is the prediction data.
It is obvious that
Figure BDA0002296020510000112
There will be prediction errors in the process, so
Figure BDA0002296020510000113
However, it is possible to use a single-layer,
Figure BDA0002296020510000114
is used to predict the next point, the result will be
Figure BDA0002296020510000115
Resulting in a larger error as shown in fig. 3. This is a prediction error superposition problem.
The present invention takes into account the problem of superposition of prediction errors. In this architecture, when historical data is given, a generator is used to predict RBDP. The arbiter then determines whether the RBDP is derived from real historical data or a predicted RBDP, thereby looping the process to compensate for the prediction error overlap problem.
6) And (3) building an automatic encoder model AE as a discriminator to judge whether the prediction result obtained in the step 5) is from real historical data.
The automatic encoder AE consists of a three-layer neural network, which together form an encoding layer and a decoding layer, as shown in fig. 4, the encoding layer compresses the original input sequence into typical features by reducing the number of neurons, and these typical features are then converted by the decoder into a reconstructed sequence of the same dimensions as the input sequence, the purpose of the AE being to reconstruct the input sequence by minimizing the error between the reconstructed sequence and the input sequence;
AE is mapped h by a certaintyθInput sequence of dimension n
Figure BDA0002296020510000116
Typical features mapping to dimension m
Figure BDA0002296020510000117
(m<n), the expression is as follows:
y＝hθ(x)＝f(Wx+b)
wherein the content of the first and second substances,
Figure BDA0002296020510000118
theta is the overall parameter of the coding layer,
Figure BDA0002296020510000119
the expression 'is defined as', f is the activation function of the coding layer, W represents the weight matrix of m multiplied by n, and b represents the offset vector of the coding part;
accordingly, the decoding part restores the characteristic features to a sequence of the same dimension as the input sequence x', as follows:
x′＝gθ′(y)＝f(W′y+b′)
wherein the content of the first and second substances,
Figure BDA0002296020510000121
theta ' is all parameters of the decoding layer, g is the activation function of the decoding layer, W ' is a weight matrix of n multiplied by m, and b ' is a bias vector of the decoding part;
the entire AE network then updates the weight matrix W, W 'and the offsets b, b' by the following loss functions:
Figure BDA0002296020510000122
wherein the content of the first and second substances,
Figure BDA00022960205100001210
is a function of the loss as a function of,
Figure BDA0002296020510000129
θ*and theta'*For the final optimization parameter, x, of the AE networks theta and theta(i)And x'(i)The input sequence of the coding layer and the decoding layer are respectively.
Further, in the AE network, softmax regression was placed at the end of the structure, classifying the characteristic features extracted by AE. The Softmax regression can be viewed as a generalized form of logistic regression.
Given a sample x comprising q samples(i)Input sequence of
Figure BDA0002296020510000123
Wherein
Figure BDA0002296020510000124
Including a set of k-class labels
Figure BDA0002296020510000125
y(i)∈{1,2,....., k, Softmax, the role of Softmax is to estimate the probability that each sample belongs to each class, and to take the class with the highest probability as the class of the sample, which is given by:
Figure BDA0002296020510000126
wherein the content of the first and second substances,
Figure BDA0002296020510000127
θ∈{θ1,θ2,......,θkthe parameters of the model are determined,
Figure BDA0002296020510000128
is a hypothesis function that adjusts the probability distribution;
the cost function of the Softmax regression model is:
Figure BDA0002296020510000131
where 1{ } is an indicative function where 1{ an expression whose value is true } ═ 1, and 1{ an expression whose value is false } -, 0.
Softmax regression by minimization
Figure BDA0002296020510000132
And classifying the typical features extracted by the AE.
7) The generator LSTM network and the discriminator AE carry out countermeasure training and seek the optimal solution by adopting a random gradient descent method.
LSTM pass training set
Figure BDA00022960205100001312
Generating a predicted sequence
Figure BDA0002296020510000133
Arbiter AE compressed prediction sequence
Figure BDA0002296020510000134
And generates a corresponding label
Figure BDA0002296020510000135
Means that
Figure BDA0002296020510000136
The prediction requirements are met;
while training set
Figure BDA00022960205100001313
When known, the prediction set
Figure BDA00022960205100001314
The conditional probability of (c) is defined as follows:
Figure BDA0002296020510000137
in order to minimize the prediction error superposition problem, the above conditional probability is used to obtain the optimal parameters in consideration of the internal parameter θ, and the following formula is adopted:
Figure BDA0002296020510000138
wherein θ means all parameters of the model, including weight, bias;
for the discriminator AE, the aim is to train a classifier for encoding the input sequence and mapping the input features to the output
Figure BDA0002296020510000139
To represent
Figure BDA00022960205100001310
From the real RBDP, not from the generative model LSTM, then the objective of the arbiter is functionalized as a penalty as follows:
Figure BDA00022960205100001311
where m represents the number of samples in the input sequence and y(i)And
Figure BDA0002296020510000141
respectively representing a real label and a discrimination label;
the goal of the training process is to minimize the log-likelihood, as follows:
Figure BDA0002296020510000142
wherein the content of the first and second substances,
Figure BDA0002296020510000143
the mathematical expectation is represented by the mathematical expectation,
Figure BDA00022960205100001411
respectively, the generation sequence of the generator and the discrimination probability of the discriminator.
Further, in step 7), AE feature vector is only in dimension n-m
Figure BDA0002296020510000144
SGD optimization model generation by random gradient descent method and AE feature vector
Figure BDA0002296020510000145
And rolling bearing degradation process X ═ X1,x2,...,xm,...,xnThe SGD optimization discriminant model is used, and the random gradients of the two optimization processes are as follows:
Figure BDA0002296020510000146
Figure BDA0002296020510000147
wherein the content of the first and second substances,
Figure BDA0002296020510000148
meaning that the parameters of discriminator D are derived, and, similarly,
Figure BDA0002296020510000149
meaning that the parameters of the generator G are differentiated,
Figure BDA00022960205100001414
representing a training set of discriminator pairs
Figure BDA00022960205100001413
The discrimination probability of (1).
This process is repeated until the AE input feature vector
Figure BDA00022960205100001410
Until the predicted performance requirements are met.
8) And outputting the prediction results of the residual life of the rolling bearing, namely the results after LSTM and AE antagonistic training is finished.
As shown in FIG. 5, the present invention sets the support vector regression SVR and LSTM as a control, and performs a comparative experiment with the present invention. The results show that the LSTM or SVR predictor can obtain better prediction results in early time periods. However, after half the lifetime, the LSTM or SVR gradually deviates from the predicted trajectory. Both of these methods fail in RUL prediction, particularly late in the degeneration process. The prediction method provided by the invention can obviously improve the prediction precision of the service life of the rolling bearing and reduce the problem of prediction error superposition.
The invention discloses a rolling bearing residual life prediction method based on a deep generation type countermeasure network, and three strategies of RUL prediction, countermeasure learning and RUL prediction updating are designed to realize an LSTM-GAN network. The LSTM-GAN network adopting the antagonism learning strategy can learn more typical characteristics in statistical characteristics, is favorable for solving the problem of prediction error superposition, and improves the precision of the residual life prediction of the rolling bearing. Through detailed experimental research, the LSTM-GAN framework provided by the invention also has great potential in the aspect of improving the quality and diversity of RUL prediction, and training is carried out under the condition that the historical data of a rolling bearing is limited, so that an effective method is provided for carrying out RUL prediction under the limited historical fault data.

Claims (10)

1. A method for predicting the residual life of a rolling bearing based on a deep generation type countermeasure network comprises the following steps:
1) collecting an original vibration signal of a rolling bearing;
2) acquiring characteristic parameters of an original vibration signal;
3) dividing the characteristic parameters obtained in the step 2) into a training set and a prediction set;
4) sending the training set into a generator long-short term memory network LSTM for training;
5) predicting the rolling bearing degradation process RBDP in a direct prediction mode and generating a prediction result;
6) building an automatic encoder model AE as a discriminator, and judging whether the prediction result obtained in the step 5) is from real historical data;
7) a long-term and short-term memory network and a discriminator AE are generated to carry out countermeasure training, and an optimal solution is sought by adopting a random gradient descent method;
8) and outputting the prediction results of the residual life of the rolling bearing, namely the results after LSTM and AE antagonistic training is finished.
2. The method for predicting the residual life of a rolling bearing based on a depth-generating countermeasure network as claimed in claim 1, wherein: in the step 1), the original vibration signals of the rolling bearing comprise a horizontal vibration signal and a vertical vibration signal, and are respectively measured by an accelerometer.
3. The method for predicting the residual life of a rolling bearing based on a depth-generating countermeasure network as claimed in claim 1, wherein: in the step 2), the original vibration signal is preprocessed to obtain statistical characteristic parameters, wherein the statistical characteristic parameters comprise a root mean square value, a standard deviation, a peak value and an average value.
4. The method for predicting the residual life of a rolling bearing based on a depth-generating countermeasure network as claimed in claim 1, wherein: in the step 3), the parameter considering the statistical characteristic is an inclusionTime series of n samples X ═ X1,x2,...,xm,...,xnThe first half is used for training the model, called training set
Figure FDA0002296020500000021
The second half is a prediction part, called prediction set
Figure FDA0002296020500000022
5. The method for predicting the residual life of a rolling bearing based on a depth-generating countermeasure network of claim 4, wherein: in the step 4), the LSTM controls the transmission of signals through the storage unit and the cell state of the combination of the input gate, the forgetting gate and the output gate, so as to learn the long-term dependence in the time sequence;
the forward propagation algorithm for LSTM is as follows:
it＝σ(Wi[ht-1,xt]+bi)
ft＝σ(Wf[ht-1,xt]+bf)
ot＝σ(Wo[ht-1,xt]+bo)
gt＝σ(Wg[ht-1,xt]+bg)
Figure FDA0002296020500000023
Figure FDA0002296020500000024
wherein it、ftAnd otRespectively showing an input gate, a forgetting gate and an output gate, gtRepresents new candidate cell information, ht-1And htRespectively representing the output, x, of the previous and current celltRepresenting the input of the current cell, Wi、Wf、WoAnd WgThe weight matrix of the input gate, the forgetting gate, the output gate and the new candidate cell information of LSTM, respectively, bi、bf、boAnd bgOffsets for input gate, forget gate, output gate and new candidate cell information, respectively, ctAnd ct-1Respectively representing the current state and the last state of the cell, sigma and tanh are sigmoid and hyperbolic activation functions respectively,
Figure FDA0002296020500000025
the dot product is represented.
6. The method for predicting the residual life of a rolling bearing based on a depth-generating countermeasure network of claim 5, wherein: in the step 5), the direct prediction mode refers to that the predicted point is added into the original sequence to predict the next point, and the specific mode is as follows:
Figure FDA0002296020500000031
wherein, x is the history data,
Figure FDA0002296020500000032
is the prediction data.
7. The method for predicting the residual life of a rolling bearing based on a depth-generating countermeasure network of claim 6, wherein: in said step 6), the automatic encoder AE consists of a three-layer neural network, which together form an encoding layer and a decoding layer, the encoding layer compresses the original input sequence into typical features by reducing the number of neurons, and these typical features are then converted by the decoder into a reconstructed sequence of the same dimensions as the input sequence, the purpose of the AE being to reconstruct the input sequence by minimizing the error between the reconstructed sequence and the input sequence;
AE is mapped h by a certaintyθInput sequence of dimension n
Figure FDA0002296020500000033
Typical features mapping to dimension m
Figure FDA0002296020500000034
The expression is as follows:
y＝hθ(x)＝f(Wx+b)
wherein the content of the first and second substances,
Figure FDA0002296020500000035
theta is the overall parameter of the coding layer,
Figure FDA0002296020500000036
the expression 'is defined as', f is the activation function of the coding layer, W represents the weight matrix of m multiplied by n, and b represents the offset vector of the coding part;
accordingly, the decoding part restores the characteristic features to a sequence of the same dimension as the input sequence x', as follows:
x′＝gθ′(y)＝f(W′y+b′)
wherein the content of the first and second substances,
Figure FDA0002296020500000037
theta ' is all parameters of the decoding layer, g is the activation function of the decoding layer, W ' is a weight matrix of n multiplied by m, and b ' is a bias vector of the decoding part;
the entire AE network then updates the weight matrix W, W 'and the offsets b, b' by the following loss functions:
Figure FDA0002296020500000041
wherein the content of the first and second substances,
Figure FDA0002296020500000042
is a function of the loss as a function of,
Figure FDA0002296020500000043
θ*and theta'*For the final optimization parameter, x, of the AE networks theta and theta(i)And x'(i)The input sequence of the coding layer and the decoding layer are respectively.
8. The method for predicting the remaining life of a rolling bearing based on a depth-generating countermeasure network of claim 7, wherein: in the step 6), in the AE network, the softmax regression is placed at the end of the structure, and typical features extracted by the AE are classified:
given a sample x comprising q samples(i)Input sequence of
Figure FDA0002296020500000044
Wherein
Figure FDA0002296020500000045
Including a set of k-class labels
Figure FDA0002296020500000046
The effect of Softmax is to estimate the probability that each sample belongs to each class, and to take the class with the highest probability as the class of the sample, which is given by:
Figure FDA0002296020500000047
wherein the content of the first and second substances,
Figure FDA0002296020500000048
θ∈{θ1,θ2,......,θkthe parameters of the model are determined,
Figure FDA0002296020500000049
is a hypothesis function that adjusts the probability distribution;
the cost function of the Softmax regression model is:
Figure FDA00022960205000000410
where 1{ } is an indicative function where 1{ an expression whose value is true } ═ 1, and 1{ an expression whose value is false } -, 0;
softmax regression by minimization
Figure FDA00022960205000000411
And classifying the typical features extracted by the AE.
9. The method for predicting the residual life of a rolling bearing based on a depth-generating countermeasure network of claim 8, wherein: in the step 7), the LSTM passes through a training set
Figure FDA0002296020500000051
Generating a predicted sequence
Figure FDA0002296020500000052
Arbiter AE compressed prediction sequence
Figure FDA0002296020500000053
And generates a corresponding label
Figure FDA0002296020500000054
Figure FDA0002296020500000055
Means that
Figure FDA0002296020500000056
The prediction requirements are met;
while training set
Figure FDA0002296020500000057
When known, the prediction set
Figure FDA0002296020500000058
The conditional probability of (c) is defined as follows:
Figure FDA0002296020500000059
in order to minimize the prediction error superposition problem, the above conditional probability is used to obtain the optimal parameters in consideration of the internal parameter θ, and the following formula is adopted:
Figure FDA00022960205000000510
wherein θ means all parameters of the model, including weight, bias;
for the discriminator AE, the aim is to train a classifier for encoding the input sequence and mapping the input features to the output
Figure FDA00022960205000000511
To represent
Figure FDA00022960205000000512
From the real RBDP, not from the generative model LSTM, then the objective of the arbiter is functionalized as a penalty as follows:
Figure FDA00022960205000000513
where m represents the number of samples in the input sequence and y(i)And
Figure FDA00022960205000000514
respectively representing a real label and a discrimination label;
the goal of the training process is to minimize the log-likelihood, as follows:
Figure FDA00022960205000000515
wherein the content of the first and second substances,
Figure FDA00022960205000000516
the mathematical expectation is represented by the mathematical expectation,
Figure FDA00022960205000000517
respectively, the generation sequence of the generator and the discrimination probability of the discriminator.
10. The method for predicting the residual life of a rolling bearing based on a depth-generating countermeasure network of claim 9, wherein: in the step 7), only the AE feature vector with the dimension of n-m
Figure FDA0002296020500000061
SGD optimization model generation by random gradient descent method and AE feature vector
Figure FDA0002296020500000062
And rolling bearing degradation process X ═ X1,x2,...,xm,...,xnThe SGD optimization discriminant model is used, and the random gradients of the two optimization processes are as follows:
Figure FDA0002296020500000063
Figure FDA0002296020500000064
wherein the content of the first and second substances,
Figure FDA0002296020500000065
meaning that the parameters of discriminator D are derived, and, similarly,
Figure FDA0002296020500000066
meaning that the parameters of the generator G are differentiated,
Figure FDA0002296020500000067
representing a training set of discriminator pairs
Figure FDA0002296020500000068
The discrimination probability of (2);
this process is repeated until the AE input feature vector
Figure FDA0002296020500000069
Until the predicted performance requirements are met.
CN201911201619.8A 2019-11-29 2019-11-29 Rolling bearing residual life prediction method based on depth generation type countermeasure network Active CN110942101B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911201619.8A CN110942101B (en) 2019-11-29 2019-11-29 Rolling bearing residual life prediction method based on depth generation type countermeasure network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911201619.8A CN110942101B (en) 2019-11-29 2019-11-29 Rolling bearing residual life prediction method based on depth generation type countermeasure network

Publications (2)

Publication Number Publication Date
CN110942101A true CN110942101A (en) 2020-03-31
CN110942101B CN110942101B (en) 2023-06-27

Family

ID=69908824

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911201619.8A Active CN110942101B (en) 2019-11-29 2019-11-29 Rolling bearing residual life prediction method based on depth generation type countermeasure network

Country Status (1)

Country Link
CN (1) CN110942101B (en)

Cited By (31)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111458142A (en) * 2020-04-02 2020-07-28 苏州智传新自动化科技有限公司 Sliding bearing fault diagnosis method based on generation of countermeasure network and convolutional neural network
CN111507046A (en) * 2020-04-16 2020-08-07 哈尔滨工程大学 Method and system for predicting remaining service life of electric gate valve
CN111581888A (en) * 2020-05-18 2020-08-25 中车永济电机有限公司 Construction method of residual service life prediction model of wind turbine bearing
CN111639467A (en) * 2020-06-08 2020-09-08 长安大学 Aero-engine service life prediction method based on long-term and short-term memory network
CN111709577A (en) * 2020-06-17 2020-09-25 上海海事大学 RUL prediction method based on long-range correlation GAN-LSTM
CN111723851A (en) * 2020-05-30 2020-09-29 同济大学 Production line fault detection method
CN111832624A (en) * 2020-06-12 2020-10-27 上海交通大学 Tool remaining life prediction method based on anti-migration learning
CN111882825A (en) * 2020-06-18 2020-11-03 闽江学院 Fatigue prediction method and device based on electroencephalogram-like wave data
CN112001482A (en) * 2020-08-14 2020-11-27 佳都新太科技股份有限公司 Vibration prediction and model training method and device, computer equipment and storage medium
CN112084717A (en) * 2020-09-15 2020-12-15 复旦大学 Ultraviolet light emitting diode performance degradation prediction model construction and service life prediction method
CN112179691A (en) * 2020-09-04 2021-01-05 西安交通大学 Mechanical equipment running state abnormity detection system and method based on counterstudy strategy
CN112215421A (en) * 2020-10-13 2021-01-12 北京工业大学 Deep learning water quality index prediction method based on generation countermeasure network
CN112214852A (en) * 2020-10-09 2021-01-12 电子科技大学 Degradation rate-considered turbine mechanical performance degradation prediction method
CN112434390A (en) * 2020-12-01 2021-03-02 江苏科技大学 PCA-LSTM bearing residual life prediction method based on multi-layer grid search
CN112446415A (en) * 2020-10-09 2021-03-05 山东中医药大学 Fusion subtraction automatic encoder algorithm for image feature extraction
CN112990439A (en) * 2021-03-30 2021-06-18 太原理工大学 Method for enhancing correlation of time series data under mine
CN113158348A (en) * 2021-05-21 2021-07-23 上海交通大学 Aircraft engine residual life prediction method based on deep learning coupling modeling
CN113269356A (en) * 2021-05-18 2021-08-17 中国人民解放军火箭军工程大学 Missing data-oriented equipment residual life prediction method and system
CN113376516A (en) * 2021-06-07 2021-09-10 科润智能控制股份有限公司 Medium-voltage vacuum circuit breaker operation fault self-diagnosis and early-warning method based on deep learning
CN113449465A (en) * 2021-06-15 2021-09-28 南京航空航天大学 Service life prediction method for rolling bearing
CN113779859A (en) * 2021-02-02 2021-12-10 北京瑞莱智慧科技有限公司 Interpretable time sequence prediction model training method and device and computing equipment
CN113807005A (en) * 2021-08-12 2021-12-17 北京工业大学 Bearing residual life prediction method based on improved FPA-DBN
CN113947186A (en) * 2021-10-13 2022-01-18 广东工业大学 Heat supply energy consumption circulation prediction method based on generation of countermeasure network
CN114584230A (en) * 2022-03-07 2022-06-03 东南大学 Predictive channel modeling method based on countermeasure network and long-short term memory network
CN114662653A (en) * 2022-02-24 2022-06-24 苏州恒美电子科技股份有限公司 Double-LSTM battery capacity estimation method based on generative countermeasure network
CN115329682A (en) * 2022-10-14 2022-11-11 南京国电南自轨道交通工程有限公司 LSTM-SVR subway station temperature prediction method based on multi-cycle characteristics
CN115456073A (en) * 2022-09-14 2022-12-09 杭州电子科技大学 Generation type confrontation network model modeling analysis method based on long-term and short-term memory
CN116010609A (en) * 2023-03-23 2023-04-25 山东中翰软件有限公司 Material data classifying method and device, electronic equipment and storage medium
CN116680554A (en) * 2023-06-19 2023-09-01 扬州大学 Rotary machine life prediction method based on probabilistic element learning model
CN117706360A (en) * 2024-02-02 2024-03-15 深圳市昱森机电有限公司 Method, device, equipment and storage medium for monitoring running state of motor
CN114662653B (en) * 2022-02-24 2024-06-04 苏州恒美电子科技股份有限公司 Double-LSTM battery capacity estimation method based on generation type countermeasure network

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109271705A (en) * 2018-09-14 2019-01-25 湘潭大学 A kind of machine prediction maintaining method based on deep learning
CN109726524A (en) * 2019-03-01 2019-05-07 哈尔滨理工大学 A kind of rolling bearing remaining life prediction technique based on CNN and LSTM
CN109766583A (en) * 2018-12-14 2019-05-17 南京航空航天大学 Based on no label, unbalanced, initial value uncertain data aero-engine service life prediction technique
CN109883699A (en) * 2018-12-20 2019-06-14 上海理工大学 A kind of rolling bearing method for predicting residual useful life based on long memory network in short-term
US20190235484A1 (en) * 2018-01-31 2019-08-01 Hitachi, Ltd. Deep learning architecture for maintenance predictions with multiple modes
CN110472280A (en) * 2019-07-10 2019-11-19 广东工业大学 A kind of power amplifier behavior modeling method based on generation confrontation neural network
WO2019221654A1 (en) * 2018-05-17 2019-11-21 Tobii Ab Autoencoding generative adversarial network for augmenting training data usable to train predictive models

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190235484A1 (en) * 2018-01-31 2019-08-01 Hitachi, Ltd. Deep learning architecture for maintenance predictions with multiple modes
WO2019221654A1 (en) * 2018-05-17 2019-11-21 Tobii Ab Autoencoding generative adversarial network for augmenting training data usable to train predictive models
CN109271705A (en) * 2018-09-14 2019-01-25 湘潭大学 A kind of machine prediction maintaining method based on deep learning
CN109766583A (en) * 2018-12-14 2019-05-17 南京航空航天大学 Based on no label, unbalanced, initial value uncertain data aero-engine service life prediction technique
CN109883699A (en) * 2018-12-20 2019-06-14 上海理工大学 A kind of rolling bearing method for predicting residual useful life based on long memory network in short-term
CN109726524A (en) * 2019-03-01 2019-05-07 哈尔滨理工大学 A kind of rolling bearing remaining life prediction technique based on CNN and LSTM
CN110472280A (en) * 2019-07-10 2019-11-19 广东工业大学 A kind of power amplifier behavior modeling method based on generation confrontation neural network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
YOON J，ET AL，: "Time-series Generative Adversarial Networks", pages 2 *
唐赛;何荇兮;张家悦;尹爱军;: "基于长短期记忆网络的轴承故障识别", 汽车工程学报, no. 04, pages 67 - 73 *

Cited By (44)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111458142A (en) * 2020-04-02 2020-07-28 苏州智传新自动化科技有限公司 Sliding bearing fault diagnosis method based on generation of countermeasure network and convolutional neural network
CN111507046A (en) * 2020-04-16 2020-08-07 哈尔滨工程大学 Method and system for predicting remaining service life of electric gate valve
CN111507046B (en) * 2020-04-16 2022-09-06 哈尔滨工程大学 Method and system for predicting remaining service life of electric gate valve
CN111581888A (en) * 2020-05-18 2020-08-25 中车永济电机有限公司 Construction method of residual service life prediction model of wind turbine bearing
CN111723851A (en) * 2020-05-30 2020-09-29 同济大学 Production line fault detection method
CN111639467A (en) * 2020-06-08 2020-09-08 长安大学 Aero-engine service life prediction method based on long-term and short-term memory network
CN111639467B (en) * 2020-06-08 2024-04-16 长安大学 Aero-engine service life prediction method based on long-term and short-term memory network
CN111832624A (en) * 2020-06-12 2020-10-27 上海交通大学 Tool remaining life prediction method based on anti-migration learning
CN111709577A (en) * 2020-06-17 2020-09-25 上海海事大学 RUL prediction method based on long-range correlation GAN-LSTM
CN111709577B (en) * 2020-06-17 2023-08-15 上海海事大学 RUL prediction method based on long-range correlation GAN-LSTM
CN111882825B (en) * 2020-06-18 2021-05-28 闽江学院 Fatigue prediction method and device based on electroencephalogram-like wave data
CN111882825A (en) * 2020-06-18 2020-11-03 闽江学院 Fatigue prediction method and device based on electroencephalogram-like wave data
CN112001482B (en) * 2020-08-14 2024-05-24 佳都科技集团股份有限公司 Vibration prediction and model training method, device, computer equipment and storage medium
CN112001482A (en) * 2020-08-14 2020-11-27 佳都新太科技股份有限公司 Vibration prediction and model training method and device, computer equipment and storage medium
CN112179691A (en) * 2020-09-04 2021-01-05 西安交通大学 Mechanical equipment running state abnormity detection system and method based on counterstudy strategy
CN112179691B (en) * 2020-09-04 2021-07-13 西安交通大学 Mechanical equipment running state abnormity detection system and method based on counterstudy strategy
CN112084717A (en) * 2020-09-15 2020-12-15 复旦大学 Ultraviolet light emitting diode performance degradation prediction model construction and service life prediction method
CN112214852A (en) * 2020-10-09 2021-01-12 电子科技大学 Degradation rate-considered turbine mechanical performance degradation prediction method
CN112446415A (en) * 2020-10-09 2021-03-05 山东中医药大学 Fusion subtraction automatic encoder algorithm for image feature extraction
CN112446415B (en) * 2020-10-09 2024-04-30 山东中医药大学 Method for fusion-subtracting automatic encoder for image feature extraction
CN112215421A (en) * 2020-10-13 2021-01-12 北京工业大学 Deep learning water quality index prediction method based on generation countermeasure network
CN112434390B (en) * 2020-12-01 2024-03-29 江苏科技大学 PCA-LSTM bearing residual life prediction method based on multi-layer grid search
CN112434390A (en) * 2020-12-01 2021-03-02 江苏科技大学 PCA-LSTM bearing residual life prediction method based on multi-layer grid search
CN113779859A (en) * 2021-02-02 2021-12-10 北京瑞莱智慧科技有限公司 Interpretable time sequence prediction model training method and device and computing equipment
CN113779859B (en) * 2021-02-02 2022-04-05 北京瑞莱智慧科技有限公司 Interpretable time sequence prediction model training method and device and computing equipment
CN112990439A (en) * 2021-03-30 2021-06-18 太原理工大学 Method for enhancing correlation of time series data under mine
CN113269356A (en) * 2021-05-18 2021-08-17 中国人民解放军火箭军工程大学 Missing data-oriented equipment residual life prediction method and system
CN113269356B (en) * 2021-05-18 2024-03-15 中国人民解放军火箭军工程大学 Missing data-oriented equipment residual life prediction method and system
CN113158348B (en) * 2021-05-21 2023-10-03 上海交通大学 Aircraft engine residual life prediction method based on deep learning coupling modeling
CN113158348A (en) * 2021-05-21 2021-07-23 上海交通大学 Aircraft engine residual life prediction method based on deep learning coupling modeling
CN113376516A (en) * 2021-06-07 2021-09-10 科润智能控制股份有限公司 Medium-voltage vacuum circuit breaker operation fault self-diagnosis and early-warning method based on deep learning
CN113449465A (en) * 2021-06-15 2021-09-28 南京航空航天大学 Service life prediction method for rolling bearing
CN113807005A (en) * 2021-08-12 2021-12-17 北京工业大学 Bearing residual life prediction method based on improved FPA-DBN
CN113947186A (en) * 2021-10-13 2022-01-18 广东工业大学 Heat supply energy consumption circulation prediction method based on generation of countermeasure network
CN114662653B (en) * 2022-02-24 2024-06-04 苏州恒美电子科技股份有限公司 Double-LSTM battery capacity estimation method based on generation type countermeasure network
CN114662653A (en) * 2022-02-24 2022-06-24 苏州恒美电子科技股份有限公司 Double-LSTM battery capacity estimation method based on generative countermeasure network
CN114584230A (en) * 2022-03-07 2022-06-03 东南大学 Predictive channel modeling method based on countermeasure network and long-short term memory network
CN115456073A (en) * 2022-09-14 2022-12-09 杭州电子科技大学 Generation type confrontation network model modeling analysis method based on long-term and short-term memory
CN115329682A (en) * 2022-10-14 2022-11-11 南京国电南自轨道交通工程有限公司 LSTM-SVR subway station temperature prediction method based on multi-cycle characteristics
CN116010609B (en) * 2023-03-23 2023-06-09 山东中翰软件有限公司 Material data classifying method and device, electronic equipment and storage medium
CN116010609A (en) * 2023-03-23 2023-04-25 山东中翰软件有限公司 Material data classifying method and device, electronic equipment and storage medium
CN116680554A (en) * 2023-06-19 2023-09-01 扬州大学 Rotary machine life prediction method based on probabilistic element learning model
CN116680554B (en) * 2023-06-19 2024-04-19 扬州大学 Rotary machine life prediction method based on probabilistic element learning model
CN117706360A (en) * 2024-02-02 2024-03-15 深圳市昱森机电有限公司 Method, device, equipment and storage medium for monitoring running state of motor

Also Published As

Publication number Publication date
CN110942101B (en) 2023-06-27

Similar Documents

Publication Publication Date Title
CN110942101A (en) Rolling bearing residual life prediction method based on depth generation type countermeasure network
CN109102005B (en) Small sample deep learning method based on shallow model knowledge migration
CN109029975B (en) Fault diagnosis method for wind power gear box
Zhang et al. DeepHealth: A self-attention based method for instant intelligent predictive maintenance in industrial Internet of Things
CN111178553A (en) Industrial equipment health trend analysis method and system based on ARIMA and LSTM algorithms
CN112130086B (en) Method and system for predicting remaining life of power battery
CN112132394B (en) Power plant circulating water pump predictive state evaluation method and system
CN114676742A (en) Power grid abnormal electricity utilization detection method based on attention mechanism and residual error network
CN112784920B (en) Yun Bianduan coordinated rotating component reactance domain self-adaptive fault diagnosis method
CN111079926B (en) Equipment fault diagnosis method with self-adaptive learning rate based on deep learning
Berghout et al. A semi-supervised deep transfer learning approach for rolling-element bearing remaining useful life prediction
CN113449919B (en) Power consumption prediction method and system based on feature and trend perception
CN113420509A (en) Wind turbine state evaluation method and device and storage medium
CN114218872B (en) DBN-LSTM semi-supervised joint model-based residual service life prediction method
CN115508073B (en) Prototype adaptive mechanical equipment fault diagnosis method based on multi-scale attention
CN112763967A (en) BiGRU-based intelligent electric meter metering module fault prediction and diagnosis method
CN113947186A (en) Heat supply energy consumption circulation prediction method based on generation of countermeasure network
CN111709577A (en) RUL prediction method based on long-range correlation GAN-LSTM
CN113887789B (en) Improved ship track prediction method and device based on cyclic neural network
CN114692983A (en) Automatic gear shifting prediction method and system for special vehicle
Dong et al. An online health monitoring framework for traction motors in high-speed trains using temperature signals
Liu et al. A hybrid LSSVR-HMM based prognostics approach
CN112257348A (en) Method for predicting long-term degradation trend of lithium battery
CN113469013B (en) Motor fault prediction method and system based on transfer learning and time sequence
CN116892932A (en) Navigation decision method combining curiosity mechanism and self-imitation learning

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