CN114065433A - Bearing residual service life prediction method based on SMA (shape memory alloy) optimization algorithm - Google Patents

Bearing residual service life prediction method based on SMA (shape memory alloy) optimization algorithm Download PDF

Info

Publication number
CN114065433A
CN114065433A CN202111386170.4A CN202111386170A CN114065433A CN 114065433 A CN114065433 A CN 114065433A CN 202111386170 A CN202111386170 A CN 202111386170A CN 114065433 A CN114065433 A CN 114065433A
Authority
CN
China
Prior art keywords
bearing
lstm
hyper
sma
algorithm
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
CN202111386170.4A
Other languages
Chinese (zh)
Inventor
刘冉冉
藏传涛
颜海彬
郑恩兴
郭威
李丽
蒋益锋
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.)
Jiangsu University of Technology
Original Assignee
Jiangsu University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Jiangsu University of Technology filed Critical Jiangsu University of Technology
Priority to CN202111386170.4A priority Critical patent/CN114065433A/en
Publication of CN114065433A publication Critical patent/CN114065433A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/17—Mechanical parametric or variational design
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • 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
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/04—Ageing analysis or optimisation against ageing

Landscapes

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

Abstract

The invention discloses a bearing remaining service life prediction method based on an SMA (shape memory alloy) optimization algorithm, which comprises the steps of data preprocessing, feature vector extraction, model construction label and data set division, SMA-LSTM prediction model construction and training and the like, wherein horizontal vibration signals containing more bearing information are selected from a bearing data set to serve as a data set of an experiment during data preprocessing, time domain and frequency domain parameters capable of reflecting bearing degradation performance are extracted from the vibration signals in the data set during feature vector extraction, and the extracted feature parameters are subjected to normalization processing. Has the advantages that: aiming at the problems that the optimization of the hyperparameter of the neural network is difficult and the optimization algorithm is easy to fall into the local optimum, the slime mold algorithm with the dynamic search structure is provided, the balance can be kept between the global search and the local search, the local search capability is very strong, the falling into the local optimum can be effectively avoided, and the prediction precision of the model is improved.

Description

Bearing residual service life prediction method based on SMA (shape memory alloy) optimization algorithm
Technical Field
The invention relates to the technical field of motor equipment fault prediction and health management, in particular to a bearing residual service life prediction method based on an SMA (shape memory alloy) optimization algorithm.
Background
With the rapid improvement of the productivity and the electronic information technology level of the modern manufacturing industry, mechanical equipment is developed not only in the direction of unidirectional mechanization and automation, but also in the direction of high integration and high intelligence, and therefore new requirements on the operation reliability of the mechanical equipment are necessarily provided. Over 70% of all mechanical equipment needs to use rolling bearings, and the operation quality of the rolling bearings directly influences whether the mechanical equipment can stably operate, so that the prediction of the Remaining service Life (RUL) of the bearings is particularly important.
Wherein the selection of the prediction model is the most critical part in the bearing RUL prediction. Currently, the commonly used prediction models mainly include three types based on physical analysis, statistical analysis and artificial intelligence, but the three types of prediction models have certain problems at the present stage. In the aspect of a traditional mathematical prediction model, Jiali and the like analyze the contact stress and the service life of a self-aligning roller bearing of a cement ball mill [ J ]. bearing, 2021(03), 6-10, the influence of gear thrust and temperature on the residual service life of the bearing is researched by calculating the load of the self-aligning roller bearing. The model is usually only suitable for the bearing under the current working condition, can not be used for predicting the service life of other bearings, and has poor generalization capability. In the aspect of a prediction model based on statistical regression, shangtang and the like in a method [ P ] for predicting the residual service life of a bearing of a combustion engine based on support vector regression: CN109472241A,2019-03-15 proposes a prediction method based on support vector regression to research the running state of the bearing, and establishes a vector regression model to predict the running state of the bearing. Although the model parameter identification is simple and the real-time performance is good, the degradation characteristic of the rolling bearing is complex, and the prediction accuracy of the prediction model based on statistics is poor. In the aspect of artificial intelligence prediction models, Qin and Ying and the like are a bearing residual life prediction method [ P ] based on a GAU neural network: CN111595583A,2020-08-28 proposes a GAU neural network prediction model aiming at the problem that the bearing RUL precision predicted by the traditional model is not high, but when the model is used for constructing a bearing degradation index, the research on the extraction of time-frequency domain features is too few, only the root mean square value is taken as the input of the model, and the feature extraction is single. The method for predicting the residual life of the bearing based on the GAU neural network [ P ]. Chongqing city: CN111595583A,2020-08-28 proposes a bearing RUL prediction method based on a Long Short-Term Memory (LSTM) neural network model, fuses the extracted features, and finally proves the effectiveness of the LSTM prediction model. In the rolling bearing life prediction method [ J ] vibration test and diagnosis 2020,40(02): 303-. Li Jun et al in a PSO-SVR-based bearing residual service life prediction method [ P ]. Jiangsu province: CN112699502A,2021-04-23, uses particle swarm optimization algorithm to optimize the hyperparameter in the support vector regression machine, Lijinpeng, etc. in GA and LSTM based intelligent traffic light dispatching method [ J ] Internet of things technology, 2019,9(12):50-54, uses genetic algorithm to optimize LSTM hyperparameter, the prediction effect is better than that of prediction model without optimization algorithm, and it is feasible to use optimization algorithm to improve the accuracy of prediction model. However, the optimization algorithm is easy to fall into local optimization, and overfitting is easily caused by directly using the residual life time of the bearing as a training label of the model. In summary, although domestic and foreign scholars make many researches on the prediction of the bearing RUL, the problems that the optimization of a prediction model is easy to fall into local optimization, prediction overfitting and the like still exist. The invention provides an LSTM prediction method based on SMA algorithm optimization, which optimizes the super-parameters of the LSTM by utilizing the dynamic adjustment search mode of the SMA and can avoid the algorithm from falling into a local optimal solution. The degradation coefficient of the bearing is constructed, so that the over-fitting phenomenon caused by directly using a model to predict the residual service time of the bearing can be avoided.
Disclosure of Invention
The method aims to solve the problems that the traditional optimization algorithm is easy to fall into local optimization when a bearing RUL prediction model is optimized in the prior art, and the prediction precision is reduced; the overfitting phenomenon is easily caused by directly predicting the residual service time of the bearing; the generalization capability of the prediction model is poor, and the provided method for predicting the residual service life of the bearing based on the SMA optimization algorithm is provided.
In order to achieve the purpose, the invention adopts the following technical scheme: a bearing residual service life prediction method based on an SMA (shape memory alloy) optimization algorithm comprises the following steps:
step 1, preprocessing data;
selecting a horizontal vibration signal containing more bearing information from the bearing data set as a data set of the experiment;
step 2, extracting the feature vector; extracting time domain and frequency domain parameters capable of reflecting the bearing degradation performance from the vibration signals in the data set, and performing normalization processing on the extracted characteristic parameters; reducing the dimension of the extracted feature vector by utilizing Principal Component Analysis (PCA), and removing feature information which cannot reflect degradation performance;
step 3, constructing labels of the models and dividing the data sets; constructing a degradation coefficient R to describe the residual service time of the bearing; normalizing the constructed degradation coefficient to be used as a label of a data set, and dividing the label into a training set and a test set;
step 4, constructing and training an SMA-LSTM prediction model; initializing parameters in a slime mold algorithm, inputting a training set in a data set into an LSTM model to train the model, taking Root Mean Square Error (RMSE) of prediction errors of the LSTM as a target function of the slime mold algorithm, and determining the optimal hyper-parameter combination of the LSTM through the optimal position output by the slime mold algorithm;
step 5, predicting the RUL of the rolling bearing; and inputting the training data of the test set into the optimized LSTM model, predicting the label R of the test set, and further completing the prediction of the RUL of the bearing.
In the method for predicting the residual service life of the bearing based on the SMA optimization algorithm, the degradation coefficient R in the step 3 describes the residual service time of the bearing, and a time sequence is formed by continuous w characteristic values;
wherein the i-th time series of degradation coefficients is:
Figure BDA0003367112640000041
the remaining service time was:
RULi＝Ri×(n-w)×t(0≤i≤n-w)
wherein n is the number of times of signal acquisition in the experiment; w is the length of each time series; r linearly decreases from 1 to 0 over time, indicating that the bearing starts to degrade when R is 1 and completely degrades when R is 0; and taking the bearing degradation feature vector as input and the degradation coefficient R as a label, and dividing a training set and a test set.
In the method for predicting the residual service life of the bearing based on the SMA optimization algorithm, after the LSTM model is constructed, the hyper-parameter learning rate, the number of hidden layer neurons and the training times in the LSTM model are automatically optimized by using the slime algorithm.
In the foregoing method for predicting remaining service life of a bearing based on an SMA optimization algorithm, the automatic optimization includes:
and (3) updating the hyper-parameters by taking the hyper-parameter combination learning rate, training times and the number of hidden layer neurons in the LSTM as the slime individual:
Figure BDA0003367112640000051
wherein
Figure BDA0003367112640000052
Represents a hyper-parametric combination;
Figure BDA0003367112640000053
representing a hyper-parameter with the minimum prediction error in the LSTM;
Figure BDA0003367112640000054
and
Figure BDA0003367112640000055
representing two randomly selected hyper-parameter individuals; r represents in [0, 1 ]]A random value in between;
Figure BDA0003367112640000056
in [0, 1 ]]Decrease linearly therebetween;
Figure BDA0003367112640000057
the weight coefficients representing the hyper-parameters,
Figure BDA0003367112640000058
is in the range of [ -a, a [)](ii) a Wherein the calculation formula of a and p is as follows:
p＝tanh|S(i)-DF|
Figure BDA0003367112640000059
taking the reciprocal of the prediction error RMSE of the LSTM as the fitness value of the algorithm, namely, the maximum fitness value in the algorithm corresponds to the minimum prediction error in the LSTM; s (i) represents the fitness value of the hyper-parameter, DF represents the optimal fitness value obtained by the hyper-parameter combination in the iterative process, and the optimal fitness value is calculated
Figure BDA00033671126400000510
And S (i) continuously updating the hyperparameter.
In the foregoing method for predicting remaining service life of a bearing based on an SMA optimization algorithm, the automatically optimizing further includes:
updating the optimal hyper-parameter combination: when the LSTM prediction error of the current hyper-parameter combination is lower than the prediction error in the previous iteration process, updating the optimal hyper-parameter combination, wherein the calculation formula is as follows:
Figure BDA0003367112640000061
wherein UB and LB are upper and lower boundaries for the set hyper-parameter optimization; rand is a random value between 0 and 1; the value of z is generally chosen to be 0.03, depending on the experimental circumstances.
In the foregoing method for predicting remaining service life of a bearing based on an SMA optimization algorithm, the automatically optimizing further includes:
updating the hyperparametric weight coefficient; hyper-parametric weight coefficients
Figure BDA0003367112640000062
The mathematical expression is as follows:
Figure BDA0003367112640000063
wherein condition represents the fitness value S (i) of the first half of all super-parameter ranks, bF and ω F respectively represent the best and worst fitness values obtained in the iterative process, and SA is a sequence of fitness values ordered in an ascending order;
Figure BDA0003367112640000064
the hyper-parameter combination corresponding to the maximum fitness value in the slime mold algorithm, namely the minimum LSTM prediction error; the process shows that even if the prediction error of the LSTM in the current iteration process is very small, the slime mold algorithm can still continuously update the weight coefficient to carry out a new iteration, so that the problem that the optimization algorithm is easy to have local optimum is avoided; obtained by the above processOptimized LSTM hyper-parameters.
Compared with the prior art, the invention has the advantages that:
1. aiming at the problems that the optimization of the hyperparameter of the neural network is difficult and the optimization algorithm is easy to fall into the local optimum, the slime mold algorithm with the dynamic search structure is provided, the balance can be kept between the global search and the local search, the local search capability is very strong, the falling into the local optimum can be effectively avoided, and the prediction precision of the model is improved.
2. Aiming at the problem that the phenomenon of neural network overfitting is easily caused by directly taking the residual service time of the bearing as a label of a model, the method provides a method for constructing the degradation coefficient of the bearing as a predicted value of the model, and takes the bearing data under three working conditions as training data of the model, so that the generalization capability of the model is improved.
Drawings
FIG. 1 illustrates vibration signals of a training set and a test set selected in the present invention;
FIG. 2 is a partial degradation characteristic parameter of the present invention;
FIG. 3 is a diagram showing the contribution rate of each feature index extracted in the test set according to the present invention;
FIG. 4 is a diagram of the RNN of the present invention;
FIG. 5 is a structural diagram of an LSTM in the present invention;
FIG. 6 is a flow chart of the LSTM hyperparameter optimization based on SMA in the present invention;
FIG. 7 is a flow chart of the RUL prediction of a bearing according to the present invention;
FIG. 8 is a prediction graph of the SMA-LSTM, SMA-BP, SMA-SVR models of the present invention;
FIG. 9 is a GWO, PSO, SMA optimization algorithm and LSTM prediction graph without optimization algorithm in the present invention.
Detailed Description
The following examples are for illustrative purposes only and are not intended to limit the scope of the present invention.
Examples
Referring to fig. 1-9, a bearing remaining service life prediction method based on an SMA optimization algorithm includes the following steps:
step 1, preprocessing data;
the Data set in IEEE PHM 2012Data Challenge was chosen as the Data set for this experiment. The data set typically contains temperature data, vertical acceleration, and horizontal acceleration vibration signals, wherein the temperature signals are typically only suitable for certain cases. The horizontal vibration signal is generally more informative than the vertical vibration signal, and is therefore taken herein as the experimental data set. The data set typically contains experimental data for eleven bearings under three operating conditions.
The selected data set can reflect the characteristics of stable operation of the bearing in the early stage, faults in the later stage, sudden change of vibration signals and the like. In combination with the above requirements, (refer to table 1) this patent selects bearings 1-1, 2-1, 3-2 under three different working conditions in the data set as a training set, bearings 1-4 as a test set, and the vibration signal is shown in fig. 1.
Figure BDA0003367112640000081
TABLE 1 PHM 2012 bearing data set
Step 2, extracting the feature vector; appropriate characteristic parameters can reflect the degradation state of the bearing, and the 15 common time domain and frequency domain characteristics are extracted to analyze the degradation state of the bearing.
During the life cycle of a bearing, the bearing is usually in steady operation in an early stage of operation, and as the bearing operates, the bearing begins to fail. The characteristic indexes are usually stable in the early stage, and the bearing begins to break down in the later stage, so that the characteristic indexes have sudden change values. Therefore, when the bearing characteristic fault index is extracted, the characteristic capable of reflecting the bearing degradation process needs to be selected, and the redundant characteristic not only increases the calculated amount, but also interferes the accuracy of bearing fault diagnosis. As can be seen from fig. 2, the characteristic parameters such as the pulse index do not reflect the degradation of the bearing well, so that it is necessary to eliminate the redundant features.
The PCA is used for reducing the dimension of the redundant feature vector, and the contribution rate of each feature degradation index of the bearings 1-4 is shown in figure 3.
FC in the figure1Is the variance; FC2Root mean square; FC3Is the mean square frequency; FC4Is an index of margin; FC5Is the kurtosis; FC6Is a peak index; FC7Is the standard deviation of frequency; FC8Is the kurtosis frequency; FC9Is the center of frequency;
step 3, constructing labels of the models and dividing the data sets; constructing a degradation coefficient R to describe the residual service time of the bearing; constructing a degradation coefficient R according to the characteristics to describe the residual service time of the bearing;
wherein the i-th time series of degradation coefficients is:
Figure BDA0003367112640000091
the remaining service time was:
RULi＝Ri×(n-w)×t(0≤i≤n-w)
wherein n is the number of times of signal acquisition in the experiment; w is the length of each time series; r linearly decreases from 1 to 0 over time, indicating that the bearing starts to degrade when R is 1 and completely degrades when R is 0; and taking the bearing degradation feature vector as input and the degradation coefficient R as a label, and dividing a training set and a test set.
Step 4, constructing and training an SMA-LSTM prediction model; initializing parameters in a slime mold algorithm, inputting a training set in a data set into an LSTM model to train the model, taking Root Mean Square Error (RMSE) of the LSTM as a target function of the slime mold algorithm, wherein the LSTM is an improved recurrent neural network, and can respectively retain short-term information and determine which information can be retained all the time. The advantage of long-term memory can be realized by using the LSTM network, the characteristic vector extracted in the front is used as the input of the LSTM, and the constructed degradation coefficient is predicted by using the gate structure and the cell state in the LSTM, wherein the flow of the recurrent neural network and the LSTM is shown in the figures 4 and 5.
After the LSTM model is constructed, automatically optimizing the super-parameter learning rate and the number of neurons in a hidden layer in the LSTM by using a slime algorithm, and determining the optimal super-parameter combination of the LSTM according to the optimal position output by the slime algorithm; the main process of automatic optimization is as follows:
firstly, combining the hyper-parameters in the LSTM with the learning rate, the training times and the number of hidden layer neurons as a slime individual, and updating the hyper-parameters:
Figure BDA0003367112640000101
wherein
Figure BDA0003367112640000102
Represents a hyper-parametric combination;
Figure BDA0003367112640000103
representing a hyper-parameter with the minimum prediction error in the LSTM;
Figure BDA0003367112640000104
and
Figure BDA0003367112640000105
representing two randomly selected hyper-parameter individuals; r represents in [0, 1 ]]A random value in between;
Figure BDA0003367112640000106
in [0, 1 ]]Decrease linearly therebetween;
Figure BDA0003367112640000107
the weight coefficients representing the hyper-parameters,
Figure BDA0003367112640000108
is in the range of [ -a, a [)](ii) a Wherein the calculation formula of a and p is as follows:
p＝tanh|S(i)-DF
Figure BDA0003367112640000109
taking the reciprocal of the prediction error RMSE of the LSTM as the fitness value of the algorithm, namely, the maximum fitness value in the algorithm corresponds to the minimum prediction error in the LSTM; s (i) represents the fitness value of the hyper-parameter, DF represents the optimal fitness value obtained by the hyper-parameter combination in the iterative process, and the optimal fitness value is calculated
Figure BDA00033671126400001010
And S (i) continuously updating the hyperparameter.
Secondly, updating the optimal hyper-parameter combination: when the LSTM prediction error of the current hyper-parameter combination is lower than the prediction error in the previous iteration process, updating the optimal hyper-parameter combination, wherein the calculation formula is as follows:
Figure BDA0003367112640000111
wherein UB and LB are upper and lower boundaries for the set hyper-parameter optimization; rand is a random value between 0 and 1; the value of z is generally chosen to be 0.03, depending on the experimental circumstances.
Thirdly, updating the hyperparametric weight coefficient; hyper-parametric weight coefficients
Figure BDA0003367112640000112
The mathematical expression is as follows:
Figure BDA0003367112640000113
wherein condition represents the fitness value S (i) of the first half of all super-parameter ranks, bF and ω F respectively represent the best and worst fitness values obtained in the iterative process, and SA is a sequence of fitness values ordered in an ascending order;
Figure BDA0003367112640000114
the hyper-parameter combination corresponding to the maximum fitness value in the slime mold algorithm, namely the minimum LSTM prediction error; the above process shows that even if the prediction error of the LSTM in the current iteration process is very small, the slime bacteria algorithm can still continuously update the weight coefficient to carry out a new iteration, so thatThe problem that the optimization algorithm is easy to generate local optimization is avoided; and obtaining the optimized LSTM hyperparameter through the process. The flow chart of SMA-based LSTM optimization is shown in fig. 6.
Step 5, predicting the RUL of the rolling bearing; and inputting the feature vectors in the test set into the trained SMA-LSTM model, predicting the constructed degradation coefficient R, and taking RMSE as an evaluation index of the model. In order to verify the effectiveness of SMA on LSTM model parameter optimization, a Support Vector Regression (SVR) model and a Back Propagation (BP) neural network which are optimized by SMA are selected for comparative analysis with the LSTM model; comparative analysis was performed using Particle Swarm Optimization (PSO), Grey Wolf algorithm (GWO), and LSTM using SMA optimization and without optimization. The bearing RUL prediction flow chart is shown in fig. 7.
The prediction results of the SMA algorithm optimized BP model, SVR model and LSTM model are shown in table 2 and fig. 8. An LSTM model with GWO, PSO and SMA optimization and without optimization algorithm is shown in figure 9.
Figure BDA0003367112640000121
TABLE 2 prediction error of each model
From table 2, fig. 8 and fig. 9, it can be known that the convergence rate of the algorithm-optimized LSTM is higher than that of the LSTM model without the optimization algorithm, wherein the prediction accuracy and the convergence rate of the model with the SMA optimization algorithm perform best in all the optimization algorithms, and the effectiveness of the SMA optimization algorithm is proved. The SMA-LSTM prediction model taking bearing data under different working conditions as training data is superior to the SMA-BP model and the SMA-SVR model in convergence rate and prediction accuracy, and proves that under the same optimization algorithm SMA, the LSTM model has higher prediction accuracy, and the SMA-LSTM has better generalization capability of prediction accuracy and can be used for predicting the RUL of the bearing.

Claims (6)

1. A bearing residual service life prediction method based on an SMA optimization algorithm is characterized by comprising the following steps:
step 1, preprocessing data;
selecting a horizontal vibration signal containing more bearing information from the bearing data set as a data set of the experiment;
step 2, extracting the feature vector; extracting time domain and frequency domain parameters capable of reflecting the bearing degradation performance from the vibration signals in the data set, and performing normalization processing on the extracted characteristic parameters; reducing the dimension of the extracted feature vector by utilizing Principal Component Analysis (PCA), and removing feature information which cannot reflect degradation performance;
step 3, constructing labels of the models and dividing the data sets; constructing a degradation coefficient R to describe the residual service time of the bearing; normalizing the constructed degradation coefficient to be used as a label of a data set, and dividing the label into a training set and a test set;
step 4, constructing and training an SMA-LSTM prediction model; initializing parameters in a slime mold algorithm, inputting a training set in a data set into an LSTM model to train the model, taking Root Mean Square Error (RMSE) of prediction errors of the LSTM as a target function of the slime mold algorithm, and determining the optimal hyper-parameter combination of the LSTM through the optimal position output by the slime mold algorithm;
step 5, predicting the RUL of the rolling bearing; and inputting the training data of the test set into the optimized LSTM model, predicting the label R of the test set, and further completing the prediction of the RUL of the bearing.
2. The method for predicting the residual service life of the bearing based on the SMA optimization algorithm is characterized in that the degradation coefficient R in the step 3 describes the residual service life of the bearing, and a time sequence is formed by continuous w characteristic values;
wherein the i-th time series of degradation coefficients is:
Figure FDA0003367112630000021
the remaining service time was:
RULi＝Ri×(n-w)×t(0≤i≤n-w)
wherein n is the number of times of signal acquisition in the experiment; w is the length of each time series; r linearly decreases from 1 to 0 over time, indicating that the bearing starts to degrade when R is 1 and completely degrades when R is 0; and taking the bearing degradation feature vector as input and the degradation coefficient R as a label, and dividing a training set and a test set.
3. The method for predicting the residual service life of the bearing based on the SMA optimization algorithm, as recited in claim 1, wherein after the LSTM model is constructed, the hyper-parameter learning rate, the number of neurons in a hidden layer and the training times in the LSTM model are automatically optimized by using a slime algorithm.
4. The method for predicting the residual service life of the bearing based on the SMA optimization algorithm according to claim 3, wherein the automatic optimization comprises the following steps:
and (3) updating the hyper-parameters by taking the hyper-parameter combination learning rate, training times and the number of hidden layer neurons in the LSTM as the slime individual:
Figure FDA0003367112630000022
wherein
Figure FDA0003367112630000023
Represents a hyper-parametric combination;
Figure FDA0003367112630000024
representing a hyper-parameter with the minimum prediction error in the LSTM;
Figure FDA0003367112630000025
and
Figure FDA0003367112630000026
representsRandomly selecting two hyper-parameter individuals; r represents in [0, 1 ]]A random value in between;
Figure FDA0003367112630000027
in [0, 1 ]]Decrease linearly therebetween;
Figure FDA0003367112630000028
the weight coefficients representing the hyper-parameters,
Figure FDA0003367112630000029
is in the range of [ -a, a [)](ii) a Wherein the calculation formula of a and p is as follows:
p＝tanh|S(i)-DF|
Figure FDA0003367112630000031
taking the reciprocal of the prediction error RMSE of the LSTM as the fitness value of the algorithm, namely, the maximum fitness value in the algorithm corresponds to the minimum prediction error in the LSTM; s (i) represents the fitness value of the hyper-parameter, DF represents the optimal fitness value obtained by the hyper-parameter combination in the iterative process, and the optimal fitness value is calculated
Figure FDA0003367112630000032
And S (i) continuously updating the hyperparameter.
5. The method for predicting the residual service life of the bearing based on the SMA optimization algorithm according to claim 4, wherein the automatic optimization further comprises:
updating the optimal hyper-parameter combination: when the LSTM prediction error of the current hyper-parameter combination is lower than the prediction error in the previous iteration process, updating the optimal hyper-parameter combination, wherein the calculation formula is as follows:
Figure FDA0003367112630000033
wherein UB and LB are upper and lower boundaries for the set hyper-parameter optimization; rand is a random value between 0 and 1; the value of z was chosen to be 0.03 depending on the experimental situation.
6. The method for predicting the residual service life of the bearing based on the SMA optimization algorithm according to claim 5, wherein the automatic optimization further comprises:
updating the hyperparametric weight coefficient; hyper-parametric weight coefficients
Figure FDA0003367112630000034
The mathematical expression is as follows:
Figure FDA0003367112630000035
wherein condition represents the fitness value S (i) of the first half of all super-parameter ranks, bF and ω F respectively represent the best and worst fitness values obtained in the iterative process, and SA is a sequence of fitness values ordered in an ascending order;
Figure FDA0003367112630000036
the hyper-parameter combination corresponding to the maximum fitness value in the slime mold algorithm, namely the minimum LSTM prediction error; the process shows that even if the prediction error of the LSTM in the current iteration process is very small, the slime mold algorithm can still continuously update the weight coefficient to carry out a new iteration, so that the problem that the optimization algorithm is easy to have local optimum is avoided; and obtaining the optimized LSTM hyperparameter through the process.
CN202111386170.4A 2021-11-22 2021-11-22 Bearing residual service life prediction method based on SMA (shape memory alloy) optimization algorithm Pending CN114065433A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111386170.4A CN114065433A (en) 2021-11-22 2021-11-22 Bearing residual service life prediction method based on SMA (shape memory alloy) optimization algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111386170.4A CN114065433A (en) 2021-11-22 2021-11-22 Bearing residual service life prediction method based on SMA (shape memory alloy) optimization algorithm

Publications (1)

Publication Number Publication Date
CN114065433A true CN114065433A (en) 2022-02-18

Family

ID=80278792

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111386170.4A Pending CN114065433A (en) 2021-11-22 2021-11-22 Bearing residual service life prediction method based on SMA (shape memory alloy) optimization algorithm

Country Status (1)

Country Link
CN (1) CN114065433A (en)

Similar Documents

Publication Publication Date Title
CN108898251B (en) Offshore wind farm power prediction method considering meteorological similarity and power fluctuation
US11840998B2 (en) Hydraulic turbine cavitation acoustic signal identification method based on big data machine learning
CN112734130B (en) Fault early warning method for double-fed fan main shaft
CN113567131A (en) Bearing fault diagnosis method based on S transformation and miniature convolution neural network model
CN111340282A (en) DA-TCN-based method and system for estimating residual service life of equipment
CN112434390B (en) PCA-LSTM bearing residual life prediction method based on multi-layer grid search
CN111680788A (en) Equipment fault diagnosis method based on deep learning
CN114861533A (en) Wind power ultra-short-term prediction method based on time convolution network
CN114048688A (en) Method for predicting service life of bearing of wind power generator
CN116992779B (en) Simulation method and system of photovoltaic energy storage system based on digital twin model
CN115860177A (en) Photovoltaic power generation power prediction method based on combined machine learning model and application thereof
CN115290326A (en) Rolling bearing fault intelligent diagnosis method
CN111178612B (en) LSTM load prediction method of grid user based on big data ODPS engine
CN113762591B (en) Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning
CN117132132A (en) Photovoltaic power generation power prediction method based on meteorological data
CN112163474A (en) Intelligent gearbox diagnosis method based on model fusion
CN114065433A (en) Bearing residual service life prediction method based on SMA (shape memory alloy) optimization algorithm
CN112257348A (en) Method for predicting long-term degradation trend of lithium battery
CN117113086A (en) Energy storage unit load prediction method, system, electronic equipment and medium
CN116304587A (en) Rolling bearing degradation trend prediction method based on CAE and AGRU
CN112434729B (en) Intelligent fault diagnosis method based on layer regeneration network under unbalanced sample
Kumari et al. Remaining useful life prediction using hybrid neural network and genetic algorithm approaches
Zhao et al. Remaining useful life prediction method based on convolutional neural network and long short-term memory neural network
Zhang et al. Fault Diagnosis of Rolling Bearing Based on CNN with Attention Mechanism and Dynamic Learning Rate
Zheng et al. Fault prediction of fan gearbox based on deep belief network

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