CN112036547B - Rolling bearing residual life prediction method combining automatic feature extraction with LSTM - Google Patents
Rolling bearing residual life prediction method combining automatic feature extraction with LSTM Download PDFInfo
- Publication number
- CN112036547B CN112036547B CN202010881816.5A CN202010881816A CN112036547B CN 112036547 B CN112036547 B CN 112036547B CN 202010881816 A CN202010881816 A CN 202010881816A CN 112036547 B CN112036547 B CN 112036547B
- Authority
- CN
- China
- Prior art keywords
- time
- model
- bearing
- rolling bearing
- residual life
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- 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
- 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
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Computing Systems (AREA)
- Mathematical Optimization (AREA)
- Life Sciences & Earth Sciences (AREA)
- Pure & Applied Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Business, Economics & Management (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Human Resources & Organizations (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Operations Research (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
Abstract
The invention discloses a rolling bearing residual life prediction method combining automatic feature extraction with LSTM, by which the residual service life of a rolling bearing can be predicted according to vibration signals. Firstly, cleaning data of vibration data of a rolling bearing in a full life cycle, and removing abnormal values; performing time-frequency analysis on the vibration signal to obtain a PWVD time-frequency image data set representing the degradation state of the bearing; then, based on transfer learning, automatically extracting features by using a pretrained VGG16 model; and finally, the extracted features are sent to an LSTM network to realize residual life prediction. The residual life prediction value provided by the invention has small mean square error, can monitor the degradation state of the bearing in real time, prevents major accidents, and provides reference opinion for predictive maintenance.
Description
Technical Field
The invention relates to a rolling bearing residual life prediction method combining automatic feature extraction with LSTM, belonging to the technical field of intelligent fault diagnosis.
Background
The main bearing is a common basic unit in a large number of rotating machines. The main bearing is generally a rolling bearing and plays roles of supporting and transmitting torsion. It is counted that 30% of rotating machinery failures are caused by rolling bearings, and that there are thousands of various rolling bearings or gear failures in doubly fed machines of domestic multi-family wind turbine manufacturers. And when the bearing fails and is not repaired or replaced in time, the gearbox which is tens of times higher in value than the bearing can fail or other parts are damaged. If the residual service life of the rolling bearing can be accurately predicted, the fault components can be timely isolated and replaced, and the waste of resources and economic loss can be greatly reduced.
The rolling bearing residual life prediction methods are classified into three types: a mechanism model-based method, a statistics-based method, and an artificial intelligence-based method. The advent of the big data and artificial intelligence age has prompted the rapid development of data-driven and artificial intelligence based residual life prediction methods. At present, a plurality of scholars realize a residual life prediction method based on a long-short-term memory (LongShortTermMemory, LSTM) cyclic neural network, but a large number of characteristics are manually extracted, and the characteristics are screened, so that the method has human subjectivity. There are also scholars that combine convolutional neural network (ConvolutionalNeuralNetwork, CNN) and LSTM to realize automatic feature extraction, but use frequency spectrum as input, consider only the frequency domain of the vibration signal, not consider the non-stationary characteristic of the vibration signal, and cannot accurately reflect the bearing degradation state. Therefore, automatic feature extraction is performed on a Pseudo Wigner VilleDistribution, PWVD time-frequency image of the bearing vibration signal based on transfer learning, and the residual life prediction is realized by combining LSTM, so that the prediction accuracy is improved.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a rolling bearing residual life prediction method combining automatic feature extraction with LSTM.
The invention is realized by the following technical scheme: an automatic feature extraction combined LSTM rolling bearing residual life prediction method is characterized by comprising the following steps:
step 1: collecting full life cycle vibration data of the rolling bearing, and performing time-frequency analysis on the vibration signal to obtain a two-dimensional time-frequency image data set;
step 2: migrating parameters of the VGG16 pre-training model to a target model of the automatic feature extraction module, fully utilizing the general feature learning capability of the pre-training model in a natural image dataset, taking the two-dimensional time-frequency image dataset obtained in the step 1 as input of a deep convolution network, taking the residual life proportion as model output, setting a parameter updating strategy, and finely adjusting the target model of the automatic feature extraction module;
step 3: taking the output of the last full-connection layer of the target model of the automatic feature extraction module as a feature vector, setting a time step, generating a feature vector set of the number of samples, the time step and the feature number, taking the feature vector set in the time step as the input of the LSTM residual life prediction model at a certain moment, taking the residual life proportion at the current moment as the model output, and training the residual life prediction model;
step 4: repeating the steps 1-3, collecting vibration signals of the bearing to be tested at the current moment and at the first 29 moments to obtain two-dimensional time-frequency images of the bearing to be tested, inputting the two-dimensional time-frequency images of the bearing to be tested into an automatic feature extraction depth convolution network trained in the step 3 to obtain feature vectors of the bearing to be tested, and sending the feature vectors to a residual life prediction model trained in the step 4 to obtain the residual life proportion of the bearing to be tested;
step 5: and calculating the final residual life according to the residual life proportion and the running time of the rolling bearing.
In the step 1, the time-frequency analysis is PWV analysis, and the two-dimensional time-frequency image is a PWV time-frequency image.
The PWV analysis is a time-frequency analysis method, and for a signal s (t), the WVD is defined as:
in the formula (1), x (t) is an analytic signal obtained by Hilbert transformation, x (t) is the complex conjugate of x (t), the bandwidth of the analytic signal is halved, the distortion influence of negative frequency in the signal is avoided, and meanwhile, the cross term interference can be effectively reduced, and x (t+τ/2) x (t- τ/2) is an instantaneous autocorrelation function of the signal x (t), thus W s (t, Ω) is the local power spectrum of the signal x (t);
in consideration of further reducing the influence of cross interference items, windowing is performed on the signals in the time dimension, and the sidelobe intensity of the signals can be effectively reduced by combining with a Hamming window, and the improved WVD distribution can be expressed as follows:
discretizing equation (2), let t=nt s Wherein T is s For sampling interval time, segmenting a signal according to a frequency conversion multiple, taking a time window w (k) with the length of D=2M < -1 >, dividing x (t) into P segments in a time dimension, respectively solving improved WVD distribution of each segment, and then adding and averaging to obtain time spectrum characteristics after time domain synchronous averaging, wherein the time spectrum characteristics can be obtained:
normalizing, let T s ＝1，ω＝ΩT s The method can obtain:
in the step 2, the automatic feature extraction module is realized based on transfer learning, namely, knowledge learned from a source data set is transferred to a target data set, a pre-training model is defined as a network structure which is fully trained on a large data set and obtains optimal model parameters, the feature learning ability of the pre-training model in a natural image data set is transferred to an automatic feature extraction task, and a time-frequency image is learned by using the pre-training model and automatic feature extraction is performed;
the parameter updating strategy mainly comprises the following 4 steps:
(1) Training on a source data set to obtain a neural network model, namely a source model;
(2) Creating a new neural network model with the same structure as the source model, namely a target model, and copying all parameters except an output layer of the pre-training model into the target model;
(3) Adding an output layer with the output size being the number of categories of the target data set for the target model, and randomly initializing model parameters of the layer;
(4) Training a target model on a target data set, training an output layer from the beginning, wherein parameters of the other layers are obtained based on fine adjustment of parameters of a source model;
the residual life ratio is defined as the ratio of the residual life of the rolling bearing at a certain moment to the full life cycle of the rolling bearing, and the formula is as follows:
wherein y is t For the remaining life ratio of a bearing at time t, RUL t For the remaining life of a bearing at time t, the FLC is the full life cycle of the bearing.
In the step 3, the feature vector F extracted by the automatic feature extraction module is expressed as F e R N ：
F＝[f 1 ,f 2 ,…,f N ] (6)
Wherein f 1 ～f N N-dimensional characteristics of a rolling bearing at a certain moment are shown;
the input of the LSTM residual life prediction model is specifically defined as x t ∈R M×N ：
Wherein x is t Characteristic matrix at time t for a rolling bearing for residual life prediction, whereinFor the eigenvector of the rolling bearing at time t, likewise, F t-M+1 And F t-M+2 The characteristic vectors of the rolling bearing at the time t-M+1 and the time t-M+2 are respectively;
the output of the LSTM residual life prediction model is the residual life proportion y of the rolling bearing at the moment t t 。
In the step 5, the remaining life calculating step includes:
(1) Establishing a linear equation between the predicted remaining life proportion and the running time through linear regression;
t＝a·y t +b (8)
(2) Calculating the total life cycle time of the rolling bearing, when y t When=0, the remaining service life of the bearing is 0, and the running time thereof, namely the full life cycle flc=a·y t +b＝a·0+b＝b；
(3) The remaining service life of the t bearing at any moment:
RUL t ＝FLC-t (9)
wherein t is the running time of the rolling bearing, y t For the remaining life ratio of the rolling bearing after the operation to the time t, FLC is the full life cycle of the bearing, RUL t For the remaining life of the bearing after the run to time t.
The beneficial effects of the invention are as follows: from the joint view of time and frequency, the change of the vibration non-stationary signal frequency of the rolling bearing along with the time is described, the degradation state of the rolling bearing is represented by using a two-dimensional time-frequency image, the expertise is not needed, and meanwhile, the subjectivity of manual feature extraction is avoided.
Based on deep transfer learning, a pre-training model which is completely trained on a natural image data set is used as a transfer learning object, and an automatic feature extraction module is initialized by parameters of the pre-training model. Meanwhile, a certain parameter fine adjustment strategy is set in consideration of the difference between vibration data and natural images, so that the network can extract abstract features of time-frequency images. The training is used for obtaining a fast-convergence and high-precision automatic feature extraction model.
And constructing a residual life prediction model by using an LSTM network, setting a certain time step, taking a characteristic matrix of a period of time as model input, and fully utilizing the advantages of LSTM in time sequence data prediction.
Drawings
The invention is further described below with reference to the drawings and examples.
FIG. 1 is a functional block diagram of the present invention;
FIG. 2 is a waveform diagram of the full life cycle vibration signal of the bearing 1 of the present invention;
FIG. 3 is a waveform of bearing vibration during steady operation of the present invention;
FIG. 4 is a graph of bearing vibration time frequency during steady operation of the present invention;
FIG. 5 is a waveform diagram of bearing vibration during the amplitude increase phase of the present invention;
FIG. 6 is a time-frequency plot of bearing vibration during the amplitude increase phase of the present invention;
FIG. 7 is a waveform diagram of bearing vibration during the amplitude ramp phase of the present invention;
FIG. 8 is a time-frequency plot of bearing vibration during the amplitude surge phase of the present invention;
fig. 9 shows the predicted value of the remaining life ratio of the bearing 3 according to the present invention.
Detailed Description
An automatic feature extraction combined with LSTM rolling bearing residual life prediction method as shown in fig. 1 to 9, characterized by comprising the steps of:
step 1: the rolling bearing full life cycle vibration data and time-frequency analysis are obtained, and the rolling bearing full life cycle vibration signal from the IEEE PHM 2012Data ChallengePRONOSTIA test bed is used to accelerate failure by applying additional load or increasing the rotating speed to the bearing. The experimental bearings 1-1 to 1-7 all run under the working condition that the rotating speed is 1800r/min and the load is 4000N, the acceleration sensor collects data once every 10s, and the time length of collecting the data every time is 0.1s, namely 2560 data points are collected every time. When the acceleration amplitude exceeds 20g, the test is considered to be invalid, and the test is ended. The 7 bearings under this condition all run from normal to failure. As shown in table 1, the number of samples of each bearing was taken, with bearing 3 being the test bearing and the remainder being the training bearing. As shown in fig. 2, a full life cycle waveform of the bearing 1 is shown. It can be seen that in the early stage of the experiment, the vibration amplitude is always and stably fluctuated, then the fluctuation floatability is increased, the amplitude is increased sharply until the final stage, and the whole degradation trend accords with the operation degradation process of the equipment.
The vibration signal is subjected to time-frequency analysis, wherein the time-frequency analysis is to express one-dimensional time sequence signals in a two-dimensional time-frequency joint distribution manner, so that various frequency components of the signals are obtained, and the corresponding relation of the frequency components in the time dimension can be established. WVD is a nonlinear time-frequency analysis method, and is used for carrying out Fourier transformation on a signal instantaneous autocorrelation function, and has the advantages of simplicity in calculation, good time-frequency focusing performance, strong visual sense and the like. The autocorrelation function brings serious cross-term interference, various cross-coupling term inhibition methods are sequentially proposed, wherein PWVD can inhibit cross-interference term while inheriting good time-frequency focusing of WVD, and has higher time-frequency resolution.
Wherein, for signal s (t), its WVD is defined as:
in the formula (1), x (t) is an analytic signal obtained by Hilbert transformation, and x is the complex conjugate of x (t). The bandwidth of the obtained analytic signal is halved, the distortion influence of negative frequency in the signal is avoided, and meanwhile, the cross term interference can be effectively reduced. x (t+τ/2) x (t- τ/2) is the instantaneous autocorrelation function of signal x (t), thus W s (t, Ω) is the local power spectrum of the signal x (t).
In consideration of further reducing the influence of cross interference items, windowing is performed on the signals in the time dimension, and the sidelobe intensity of the signals can be effectively reduced by combining with a Hamming window, and the improved WVD distribution can be expressed as follows:
discretizing equation (2), let t=nt s Wherein T is s Is the sampling interval time. The signal is segmented according to the frequency conversion multiple, a time window w (k) with the length of D=2M < -1 > is taken, x (t) is divided into P segments in the time dimension, and each segment is respectively improvedAnd (3) adding and averaging the WVD distribution to obtain the time spectrum characteristics after time domain synchronous averaging. The method can obtain:
normalizing, let T s ＝1，ω＝ΩT s The method can obtain:
as shown in fig. 3-8, the waveform diagrams and PWVD time-frequency images of the vibration signals of different degradation stages of the bearing 1 are shown in fig. 3, the 3 groups of diagrams respectively represent that the bearing 1 is in 3 different degradation stages, the frequency components of the signals change along with the increase of the amplitude of the vibration signals, the vibration waves are in the vicinity of 4000Hz from normal steady operation to the vicinity of 0-2000Hz in amplitude, and in addition, obvious impact is also generated in the vibration wave patterns, and meanwhile, the vibration waves are reflected in the time-frequency diagrams. By constructing the time-frequency characteristic image, the fault recognition is converted into the recognition problem of the image. The deep convolutional neural network can automatically learn the global and local characteristics of the time-frequency image through the function of a convolutional layer filter.
Step 2: and (5) constructing an automatic feature extraction module based on transfer learning. The process comprises the following steps:
(1) Training on the source data set to obtain a neural network model, namely a source model. The pretrained model of VGG16 in mxnet framework on ImageNet dataset is directly used here.
(2) A new neural network model, i.e., a target model, is created that is structurally identical to the source model. And copying all parameters except the output layer of the pre-training model into the target model.
(3) Since the output of the automatic feature extraction module is the remaining life ratio and the dimension is 1, an output layer with the output size of 1 is added to the target model, and model parameters of the layer are randomly initialized.
(4) The target model is trained on a target dataset, i.e. a rolling bearing two-dimensional time-frequency image dataset. The parameters of the first 4 convolution modules are fixed, the parameters of the last convolution module and the fully connected layer after it are fine-tuned with a small learning rate (0.001), while the parameters of the output layer are trained de novo with a 10-fold learning rate (0.01). Therefore, the extraction capacity of the pre-training model for general features is reserved, and meanwhile, the extraction capacity of the training model for abstract features of the two-dimensional time-frequency image dataset is reserved, so that the parameter migration from the pre-training model which is completely trained on the natural image dataset to the automatic feature extraction convolution model is realized.
The model output is the remaining life ratio defined in equation (5), using the L2 loss as a loss function for model training.
Step 3: the LSTM residual life prediction model is constructed, and the flow is as follows:
(1) Obtaining the output of the last full connection layer of the automatic feature extraction module as a feature vector F= [ F ] 1 ,f 2 ,…,f N ],F∈R N Here n=4096.
(2) Reconstructing feature vectors of all samples, generating feature matrices like (number of samples, time step, feature dimension), i.e. the input of LSTM residual life prediction model is the feature matrix x composed of feature vectors in time step t ∈R M×N ：
The output is the remaining life ratio defined in equation (5).
(3) A single hidden layer network is created, the number of hidden layer neurons is 30, and network parameters are initialized randomly. Setting the learning rate to be 0.1, the time step to be 30, the batch_size to be 100, and the iteration step number to be 20000 steps; using training set { x } t ,y t } T t＝1 Training the model, wherein the mean square error is used as an evaluation index.
Step 4: and for the test bearing, obtaining time-frequency image characteristics through time-frequency analysis, obtaining characteristic vectors by an automatic characteristic extraction module, and sending the characteristic vectors to an LSTM residual life prediction model through reconstruction to realize the prediction of the residual life proportion. The predicted results are shown in the figure. It can be seen that the proposed automatic feature extraction+lstm method has smaller mean square error and higher prediction accuracy when predicting the remaining life ratio.
Step 5: the residual life calculating step comprises the following steps:
(1) Establishing a linear equation between the predicted remaining life proportion and the running time through linear regression; for test bearing 3, the remaining life required by the IEEE2012 PHM major race to run to 18010s was calculated. The linear equation t= -27045.2 ·y is derived using the remaining lifetime ratio and the run time before running to 18010s t +24815；
(2) Calculating the total life cycle time of the rolling bearing, when y t When=0, the remaining service life of the bearing is 0, and the running time of the bearing is full life cycle flc= 24815s;
(3) Remaining service life of bearing 3 at this point in operation to 18010 s: RUL (continuous unit of time) 18010 =24815-18010=6805 s, the actual remaining lifetime is 5730s, and the error is 18.76%.
TABLE 1 number of samples for each bearing for working condition 1
TABLE 2 residual Life prediction error at bearing 318010s
Claims (3)
1. An automatic feature extraction combined LSTM rolling bearing residual life prediction method is characterized by comprising the following steps:
step 1: collecting full life cycle vibration data of the rolling bearing, and performing time-frequency analysis on the vibration signal to obtain a two-dimensional time-frequency image data set;
step 2: migrating parameters of the VGG16 pre-training model to a target model of the automatic feature extraction module, fully utilizing the general feature learning capability of the pre-training model in a natural image dataset, taking the two-dimensional time-frequency image dataset obtained in the step 1 as input of a deep convolution network, taking the residual life proportion as model output, setting a parameter updating strategy, and finely adjusting the target model of the automatic feature extraction module;
step 3: taking the output of the last full-connection layer of the target model of the automatic feature extraction module as a feature vector, setting a time step, generating a feature vector set of the number of samples, the time step and the feature number, taking the feature vector set in the time step as the input of the LSTM residual life prediction model at a certain moment, taking the residual life proportion at the current moment as the model output, and training the residual life prediction model;
step 4: repeating the steps 1-3, collecting vibration signals of the bearing to be tested at the current moment and at the first 29 moments to obtain two-dimensional time-frequency images of the bearing to be tested, inputting the two-dimensional time-frequency images of the bearing to be tested into an automatic feature extraction depth convolution network trained in the step 3 to obtain feature vectors of the bearing to be tested, and sending the feature vectors to a residual life prediction model trained in the step 4 to obtain the residual life proportion of the bearing to be tested;
step 5: calculating according to the proportion of the residual life and the running time of the rolling bearing to obtain the final residual life;
in the step 2, the automatic feature extraction module is realized based on transfer learning, namely, knowledge learned from a source data set is transferred to a target data set, a pre-training model is defined as a network structure which is fully trained on a large data set and obtains optimal model parameters, the feature learning ability of the pre-training model in a natural image data set is transferred to an automatic feature extraction task, and a time-frequency image is learned by using the pre-training model and automatic feature extraction is performed;
the parameter updating strategy mainly comprises the following 4 steps:
(1) Training on a source data set to obtain a neural network model, namely a source model;
(2) Creating a new neural network model with the same structure as the source model, namely a target model, and copying all parameters except an output layer of the pre-training model into the target model;
(3) Adding an output layer with the output size being the number of categories of the target data set for the target model, and randomly initializing model parameters of the layer;
(4) Training a target model on a target data set, training an output layer from the beginning, wherein parameters of the other layers are obtained based on fine adjustment of parameters of a source model;
the residual life ratio is defined as the ratio of the residual life of the rolling bearing at a certain moment to the full life cycle of the rolling bearing, and the formula is as follows:
wherein yt is the remaining life proportion of a certain bearing at a moment t, RULt is the remaining life of a certain bearing at a moment t, and FLC is the full life cycle of the bearing;
in the step 3, the feature vector F extracted by the automatic feature extraction module is expressed as F e RN:
F＝[f 1 ,f 2 ,…,f N ]
wherein f1 to fN represent N-dimensional characteristics of the rolling bearing at a certain moment;
the specific definition of the inputs of the LSTM residual life prediction model is xt epsilon RM N:
wherein xt is the characteristic matrix of the rolling bearing at time t for residual life prediction, whereinThe characteristic vector of the rolling bearing at the time t is the characteristic vector of the rolling bearing at the time t, and the Ft-M+1 and the Ft-M+2 are the characteristic vectors of the rolling bearing at the time t-M+1 and the time t-M+2 respectively;
the output of the LSTM residual life prediction model is the residual life proportion yt of the rolling bearing at the moment t.
2. The method for predicting the residual life of a rolling bearing by combining automatic feature extraction with LSTM according to claim 1, wherein:
in the step 1, the time-frequency analysis is PWV analysis, and the two-dimensional time-frequency image is a PWV time-frequency image.
3. The method for predicting the residual life of a rolling bearing by combining automatic feature extraction with LSTM according to claim 1, wherein:
in the step 5, the remaining life calculating step includes:
(1) Establishing a linear equation between the predicted remaining life proportion and the running time through linear regression;
t＝a·y t +b
(2) Calculating the total life cycle time of the rolling bearing, when y t When=0, the remaining service life of the bearing is 0, and the running time thereof, namely the full life cycle flc=a·y t +b＝a·0+b＝b；
(3) The remaining service life of the t bearing at any moment:
RUL t ＝FLC-t
wherein t is the running time of the rolling bearing, y t For the remaining life ratio of the rolling bearing after the operation to the time t, FLC is the full life cycle of the bearing, RUL t For the remaining life of the bearing after the run to time t.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010881816.5A CN112036547B (en) | 2020-08-28 | 2020-08-28 | Rolling bearing residual life prediction method combining automatic feature extraction with LSTM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010881816.5A CN112036547B (en) | 2020-08-28 | 2020-08-28 | Rolling bearing residual life prediction method combining automatic feature extraction with LSTM |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112036547A CN112036547A (en) | 2020-12-04 |
| CN112036547B true CN112036547B (en) | 2023-09-22 |
Family
ID=73586716
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010881816.5A Active CN112036547B (en) | 2020-08-28 | 2020-08-28 | Rolling bearing residual life prediction method combining automatic feature extraction with LSTM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112036547B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112486096A (en) * | 2020-12-09 | 2021-03-12 | 中国兵器装备集团自动化研究所 | Machine tool operation state monitoring method |
| CN113075546A (en) * | 2021-03-24 | 2021-07-06 | 河南中烟工业有限责任公司 | Motor vibration signal feature extraction method and system |
| CN112949097B (en) * | 2021-04-19 | 2022-09-16 | 合肥工业大学 | Bearing residual life prediction model and method based on deep migration learning |
| CN113051689B (en) * | 2021-04-25 | 2022-03-25 | 石家庄铁道大学 | Bearing residual service life prediction method based on convolution gating circulation network |
| CN113255209B (en) * | 2021-04-27 | 2022-06-07 | 燕山大学 | Method for predicting residual life of bearing of gearbox |
| CN113465923B (en) * | 2021-06-08 | 2023-04-11 | 华能国际电力股份有限公司上海石洞口第一电厂 | Residual life prediction method for grinding roller bearing of coal mill |
| CN113536982B (en) * | 2021-06-28 | 2023-02-24 | 长江大学 | Intelligent prediction method, system, equipment and medium for residual life of main reducer |
| CN113326590B (en) * | 2021-07-16 | 2021-10-29 | 北京博华信智科技股份有限公司 | Rolling bearing service life prediction method and device based on deep learning model |
| CN113984833B (en) * | 2021-10-29 | 2024-03-01 | 江苏徐工工程机械研究院有限公司 | Environment temperature equivalent and acceleration test method |
| CN114186500B (en) * | 2022-02-16 | 2022-04-29 | 华中科技大学 | Marine bearing residual life prediction method based on transfer learning and multiple time windows |
| CN115660198B (en) * | 2022-11-07 | 2023-08-01 | 中国科学院空间应用工程与技术中心 | Method for predicting residual service life of rolling bearing |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101855440A (en) * | 2007-11-07 | 2010-10-06 | 飞思卡尔半导体公司 | Knock signal detection in automotive systems |
| CN103728663A (en) * | 2012-10-16 | 2014-04-16 | 中国石油化工股份有限公司 | Time-frequency analysis method |
| CN109710636A (en) * | 2018-11-13 | 2019-05-03 | 广东工业大学 | A kind of unsupervised industrial system method for detecting abnormality based on depth migration study |
| 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 |
| CN110941928A (en) * | 2019-11-26 | 2020-03-31 | 哈尔滨理工大学 | Rolling bearing residual life prediction method based on dropout-SAE and Bi-LSTM |
| CN111579243A (en) * | 2020-06-17 | 2020-08-25 | 大连理工大学 | Rolling bearing intelligent diagnosis system based on deep migration learning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10229369B2 (en) * | 2016-04-19 | 2019-03-12 | General Electric Company | Creating predictive damage models by transductive transfer learning |
-
2020
- 2020-08-28 CN CN202010881816.5A patent/CN112036547B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101855440A (en) * | 2007-11-07 | 2010-10-06 | 飞思卡尔半导体公司 | Knock signal detection in automotive systems |
| CN103728663A (en) * | 2012-10-16 | 2014-04-16 | 中国石油化工股份有限公司 | Time-frequency analysis method |
| CN109710636A (en) * | 2018-11-13 | 2019-05-03 | 广东工业大学 | A kind of unsupervised industrial system method for detecting abnormality based on depth migration study |
| 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 |
| CN110941928A (en) * | 2019-11-26 | 2020-03-31 | 哈尔滨理工大学 | Rolling bearing residual life prediction method based on dropout-SAE and Bi-LSTM |
| CN111579243A (en) * | 2020-06-17 | 2020-08-25 | 大连理工大学 | Rolling bearing intelligent diagnosis system based on deep migration learning |
Non-Patent Citations (1)
| Title |
|---|
| 基于伪维格纳分布的图像融合;刘聪;李言俊;张科;;光学学报(10);第2716-2720页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112036547A (en) | 2020-12-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112036547B (en) | Rolling bearing residual life prediction method combining automatic feature extraction with LSTM | |
| Xu et al. | Fault diagnosis of rolling bearing of wind turbines based on the variational mode decomposition and deep convolutional neural networks | |
| CN109460618B (en) | Rolling bearing residual life online prediction method and system | |
| CN107066759B (en) | Steam turbine rotor vibration fault diagnosis method and device | |
| CN107657250B (en) | Bearing fault detection and positioning method and detection and positioning model implementation system and method | |
| CN111914883B (en) | Spindle bearing state evaluation method and device based on deep fusion network | |
| CN106197999B (en) | A kind of planetary gear method for diagnosing faults | |
| Martinez-Garcia et al. | Visually interpretable profile extraction with an autoencoder for health monitoring of industrial systems | |
| CN108256556A (en) | Wind-driven generator group wheel box method for diagnosing faults based on depth belief network | |
| Yu et al. | Image-based damage recognition of wind turbine blades | |
| CN114659790B (en) | Identification method for bearing faults of variable-rotation-speed wind power high-speed shaft | |
| CN114048688A (en) | Method for predicting service life of bearing of wind power generator | |
| CN111275108A (en) | Method for performing sample expansion on partial discharge data based on generation countermeasure network | |
| CN116226646A (en) | Method, system, equipment and medium for predicting health state and residual life of bearing | |
| CN113240022A (en) | Wind power gear box fault detection method of multi-scale single-classification convolutional network | |
| Rahimilarki et al. | Time-series deep learning fault detection with the application of wind turbine benchmark | |
| CN112163474B (en) | Intelligent gearbox diagnosis method based on model fusion | |
| CN115828088A (en) | High-voltage shunt reactor vibration abnormity fault diagnosis method based on deep learning | |
| CN115165364A (en) | Wind turbine generator bearing fault diagnosis model construction method based on transfer learning | |
| CN112307918A (en) | Diagnosis method for transformer direct-current magnetic biasing based on fuzzy neural network | |
| Bi et al. | Fault diagnosis method of wind turbine gearbox based on deep belief network | |
| Jiangtao et al. | Application of Hilbert-Huang transform and Mahalanobis-Taguchi System in mechanical fault diagnostics using vibration signals | |
| Shang et al. | Varying speed bearing fault diagnosis based on synchroextracting transform and deep residual network | |
| Yan et al. | Research on Bearing Fault Diagnosis Based on SPWVD and Grid Optimization CNN | |
| Guo et al. | A Novel Bearing Fault Diagnosis Method Based on the DLM-CNN Framework |
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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 221000 No. 6, Keji Road, Xuzhou Economic and Technological Development Zone, Xuzhou City, Jiangsu Province Applicant after: XCMG Hanyun Technology Co.,Ltd. Address before: No.6, Keji Road, Gulou District, Xuzhou City, Jiangsu Province Applicant before: JIANGSU XUGONG INFORMATION TECHNOLOGY Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |