CN114626625B - Bearing life prediction method based on multi-module U-BiLSTM network - Google Patents
Bearing life prediction method based on multi-module U-BiLSTM network Download PDFInfo
- Publication number
- CN114626625B CN114626625B CN202210302001.6A CN202210302001A CN114626625B CN 114626625 B CN114626625 B CN 114626625B CN 202210302001 A CN202210302001 A CN 202210302001A CN 114626625 B CN114626625 B CN 114626625B
- Authority
- CN
- China
- Prior art keywords
- data
- module
- time
- bearing
- predicted
- 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
- 238000000034 method Methods 0.000 title claims abstract description 81
- 238000013528 artificial neural network Methods 0.000 claims abstract description 55
- 238000012549 training Methods 0.000 claims abstract description 45
- 230000001133 acceleration Effects 0.000 claims abstract description 22
- 238000005096 rolling process Methods 0.000 claims abstract description 16
- 238000012545 processing Methods 0.000 claims abstract description 9
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 238000000605 extraction Methods 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 19
- 230000002457 bidirectional effect Effects 0.000 claims description 17
- 239000011159 matrix material Substances 0.000 claims description 15
- 210000002569 neuron Anatomy 0.000 claims description 15
- 238000005070 sampling Methods 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 11
- 230000004913 activation Effects 0.000 claims description 6
- 238000012937 correction Methods 0.000 claims description 6
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 5
- 238000013527 convolutional neural network Methods 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 5
- 238000011176 pooling Methods 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 210000005036 nerve Anatomy 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 2
- 238000005259 measurement Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 abstract description 2
- 238000012733 comparative method Methods 0.000 description 9
- 238000012360 testing method Methods 0.000 description 9
- 238000011156 evaluation Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 230000036541 health Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 241001391944 Commicarpus scandens Species 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000008485 antagonism Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- 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
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a bearing life prediction method based on a multi-module U-BiLSTM network. The method comprises the following steps: respectively acquiring vibration acceleration signal training data x S (t) and data x T (t) to be predicted of the rolling bearing in the same working condition; after standardized processing is carried out on the acquired training data and the data to be predicted, data preprocessing is carried out, and a training set and a set to be predicted are obtained; constructing a multi-module U-BiLSTM neural network life prediction model; the training set multi-module U-BiLSTM neural network life prediction model is used for obtaining a trained multi-module U-BiLSTM neural network life prediction model; and inputting the set to be predicted into a multi-module U-BiLSTM neural network life prediction model after training is finished, and outputting a prediction result of the residual service life of the bearing. The invention can realize the extraction and the utilization of deep features in the bearing life prediction process, and has better comprehensive prediction performance as a whole.
Description
Technical Field
The invention belongs to the field of life prediction of rotary machinery, and particularly relates to a bearing life prediction method based on a multi-module U-BiLSTM network.
Background
The bearing has high bearing capacity and is widely applied to various fields such as automobiles, ships, aviation and the like. However, the general bearing has a severe working environment, is easy to break down, affects the service life of equipment if the bearing is light, causes serious property loss and casualties if the bearing is heavy, and has important significance in high-precision life prediction of the bearing. In recent years, in the field of life prediction, many expert scholars have conducted research on a data driving method based on a convolutional neural network and a cyclic neural network. The type of study performs reliability analysis based on signal timing characteristics.
The conventional life prediction field data driving method comprises a life prediction method (Li X,Ding Q,Sun J Q.Remaining useful life estimation in prognostics using deep convolution neural networks[J].Reliability Engineering&System Safety,2018,172:1-11.) combining a time window method and a one-dimensional convolutional neural network and a life prediction method (Lu H,Barzegar V,Nemani V P,et al.GAN-LSTM predictor for failure prognostics of rolling element bearings[C]//2021IEEE International Conference on Prognostics and Health Management(ICPHM).IEEE,2021:1-8.), combining an antagonism network and a long-short-term memory neural network, wherein the life prediction method is mainly based on the conventional neural network, deep features of a long-time sequence are difficult to extract, and the utilization rate of historical information of the long-time sequence is low. The existing novel method (Eid A,Clerc G,Mansouri B.A Novel Deep Soft Clustering for Unsupervised Univariate Times Series[C]//2021IEEE International Conference on Prognostics and Health Management(ICPHM).IEEE,2021:1-8.) for predicting the service life of converting one-dimensional signals into two-dimensional images and extracting deep features by utilizing a U-Net neural network optimizes the long-time sequence feature extraction process, but does not consider that the U-Net neural network has strict limitation on the size of an input image, and solves the problems that the long-time sequence memory neural network consumes long time and short time, has high sensitivity to real-time information and has low utilization rate to the history information of the long-time sequence, and finally leads to the reduction of the prediction precision of the service life of the bearing.
Disclosure of Invention
The invention aims at overcoming the defects of the prior art and provides the method. The multi-module U network and the bidirectional long-short-term memory neural network are combined to realize extraction and utilization of deep features in the bearing life prediction process, and the comprehensive prediction performance is better as a whole.
The object of the invention is achieved by at least one of the following technical solutions.
A bearing life prediction method based on a multi-module U-BiLSTM network comprises the following steps:
s1, respectively acquiring vibration acceleration signal training data x S (t) and data x T (t) to be predicted of a rolling bearing in the same working condition;
s2, after standardized processing is carried out on the training data and the data to be predicted acquired in the step S1, carrying out data preprocessing on the training data and the data to be predicted by using a time window embedding strategy to acquire a training set and a set to be predicted;
S3, constructing a multi-module U-BiLSTM neural network life prediction model, wherein the multi-module U network is used for feature extraction, and the BiLSTM network is used for regression prediction;
S4, taking the percentage of the actual residual service life (RUL) of the rolling bearing in the whole life cycle as a corresponding label, inputting a training set as input data into a multi-module U-BiLSTM neural network life prediction model for iterative training until a loss function reaches a preset value, and obtaining a trained multi-module U-BiLSTM neural network life prediction model;
S5, inputting the set to be predicted into a multi-module U-BiLSTM neural network life prediction model after training is completed, and outputting a prediction result of the residual service life of the bearing.
Further, the step S1 specifically includes the following steps:
S1.1, establishing a three-dimensional space coordinate system according to a right-hand rule: the X axis is parallel to the axial direction of the shaft where the bearing is positioned, and the positive direction points to the side of the motor; the Z axis is vertical to the ground, and the positive direction is vertical upwards; the Y-axis positive direction is determined by a right hand rule;
s1.2, installing a sensor: two unidirectional acceleration sensors are respectively fixed on the horizontal direction and the vertical direction of the rolling bearing to be measured through a magnetic seat and are used for measuring vibration acceleration data of the rolling bearing to be measured; the sensor, the data acquisition system and the computer are correctly connected;
S1.3, setting data acquisition parameters: setting the total sampling time length as T and the sampling frequency as f s, then setting the corresponding sampling time interval Deltat=1/f s, the sampling point number N=f s.T, and recording the acquired time domain vibration acceleration signal as x (T);
S1.4, distinguishing the acquired time domain vibration acceleration signals x (t) of the rolling bearing according to time, wherein the time domain vibration acceleration signals x (t) in the former part of time are taken as training data x S (t), and the time domain vibration acceleration signals x (t) in the latter part of time are taken as data x T (t) to be predicted.
Further, in step S2, in order to minimize the influence of the difference of the life cycle of different bearings under the same working condition on the prediction accuracy of the provided model, the collected training data x S (t) and the data x T (t) to be predicted are respectively subjected to standardization processing, where the standardization formula is as follows:
Wherein x t is the original data obtained by collection, and comprises training data x S (t) and data x T (t) to be predicted; μ is the mean of the raw data, σ is the standard deviation of the raw data, and x t norm represents the normalized raw data.
Further, in step S2, after normalization, the obtained normalized training data and the data to be predicted are subjected to data preprocessing by using a time window embedding strategy, so as to obtain a training set X S and a set to be predicted X T, where the training set and the set to be predicted obtained by the time window embedding strategy are respectively composed of vibration signals with a certain time window size and signals with S-1 step lengths before the vibration signals, and are expressed as follows:
Wherein, For the data at time t after normalization,/>For the normalized data at time t-S+1, x t input is the sample at time t, and S is the total step size when the time window slides.
Further, in step S3, the multi-module U-BiLSTM life prediction neural network model includes an encoder, a decoder, and a regressor connected in sequence;
the encoder comprises a first module and a second module which are sequentially connected, wherein the first module is a two-dimensional convolutional neural network and extracts shallow features of vibration signal data; the second module comprises a residual block and a two-dimensional depth convolution separable nerve network which are connected in sequence;
the decoder comprises a third module, wherein the third module comprises a deconvolution layer, a pooling layer and a batch normalization layer which are sequentially connected;
The regressor comprises a bidirectional long-short-time memory neural network and a full-connection layer which are connected in sequence;
Wherein the encoder and the decoder are connected linearly in sequence and are collectively referred to as a multi-module U network, and the regressor is referred to as BiLSTM network;
The input data is extracted through the multi-module U network characteristics to obtain output data, and the obtained output data is input to a BiLSTM network to realize prediction; after the output data is input into BiLSTM networks, the output c t at the current moment is obtained through a bidirectional long-short-time memory neural network, and the formula is as follows:
ft=σg(Wf(xt,ct-1)+bf);
it=σg(Wi(xt,ct-1)+bi);
ot=tanh(Wo(xt,ct-1)+bo);
ct=ft*ct-1+it*ot;
Wherein f t is forget gate output, i t is input gate output, o t is output gate output, and c t is output at the current time t; σ g is a sigmoid activation function; tanh (·) is the tanh activation function; w f、Wi、Wo is a forgetting gate weight matrix, an input gate weight matrix and an output gate weight matrix respectively; b f、bi、bo is forgetting gate bias weight, input gate bias weight and output gate bias weight respectively; the bidirectional long-short-time memory neural network comprises a plurality of neurons which are connected in sequence, each neuron corresponds to a moment, c t represents the output of the current t-th neuron, namely the output of the current moment t, c t-1 represents the output of the t-1 th neuron, namely the output of the last moment t-1, and c t-2 represents the output of the t-2 nd neuron, namely the output of the last moment t-2;
The output c t obtained by the bidirectional long-short time memory neural network at the current time t is input into the full-connection layer to realize prediction, and the formula is as follows:
z=W*ct+b;
Wherein z is output data of the full connection layer, c t is input data of the full connection layer, and W and b are respectively a weight matrix and a bias matrix of the full connection layer.
Further, step S4 includes the steps of:
S4.1, taking the percentage of the actual residual service life (RUL) of the rolling bearing in the whole life cycle as a corresponding label;
s4.2, setting a mean square error function as a loss function of a multi-module U-BiLSTM neural network life prediction model;
And S4.3, optimizing the multi-module U-BiLSTM neural network life prediction model by using an Adam optimizer.
Further, in step S4.1, the conversion formula is:
Wherein, T sta is the starting running time of the bearing, FPT is the time from the healthy state to the degraded state of the bearing, T end is the complete failure time of the bearing, T is the current running time, and RUL true is the normalized label value.
Further, in step S4.1, the formula of the loss function is:
Wherein L is a loss function, y i is a predicted value of the residual service life of the bearing to be predicted in the ith data to be predicted, And m is the number of data quantities and is the actual measurement value of the residual service life of the bearing to be predicted in the ith data to be predicted.
Further, in step S4.3, an update step is calculated by comprehensively considering the first moment estimate (the mean value of the gradient) and the second moment estimate (the non-centered variance of the gradient) of the gradient; the formula is:
mt=β1mt-1+(1-β1)gt;
vt=β1vt-1+(1-β1)gt 2;
Where m t is the first moment estimate of the current time step, v t is the second moment estimate of the current time step, m t-1 is the first moment estimate of the last time step, v t is the second moment estimate of the last time step, g t is the gradient, Correction value estimated for the first moment of the current time step,/>Correction value estimated for the second moment of the current time step,/>For learning rate, β 1、β2, ε are constants, θ t is the dynamic constraint value for the current time step, and θ t+1 is the dynamic constraint value for the next time step.
Compared with the prior art, the invention has the following advantages and effects:
(1) Compared with the conventional data driving method in the life prediction field, the bearing life prediction method combining the module U-Net neural network and the bidirectional long-short-time memory neural network can extract deep features of long-time sequences and further improve the utilization rate of long-time sequence historical information.
(2) Compared with the existing novel method for predicting the service life of converting one-dimensional signals into two-dimensional images and extracting deep features by utilizing a U-Net neural network, the method introduces residual blocks into the encoder of the U-Net neural network, realizes multi-scale feature fusion, improves the capability of the U-Net neural network for processing long-time sequences, and reduces the requirement on the image size; the pooling layer and the normalization layer are introduced into a U-Net neural network decoder, so that the model prediction speed is improved; the multi-module U neural network is combined with the bidirectional long-short-time memory neural network, so that the utilization rate of long-time sequence historical information is improved, and the comprehensive prediction performance of the invention is better.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are required to be used in the embodiments will be briefly described below. The accompanying drawings constitute a part of this application and are provided by way of non-limiting examples embodying the inventive concepts and are not intended to be in any way limiting.
FIG. 1 is a flow chart of a life prediction method for predicting the remaining life of a bearing in an embodiment of the method of the present invention.
Fig. 2 is a graph of the prediction result of the method of the present invention for the remaining service life of the bearing 1_5.
FIG. 3 is a graph of the predicted remaining useful life of the U-Net network for bearing 1_5.
Fig. 4 is a graph of the prediction of the remaining useful life of the bearing 2_4 according to the method of the present invention.
Fig. 5 is a graph of the predicted remaining life of the U-Net network for bearing 2_4.
Fig. 6 is a graph of the prediction result of the method of the present invention for the remaining service life of the bearing 3_3.
Fig. 7 is a graph of the predicted remaining life of the U-Net network for bearing 3_3.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1: under the first experiment of working condition, the life prediction of the residual service life of the bearing (comparison method)
To verify the effectiveness of the design of the present invention, the present invention was verified using an accelerated life test dataset from XJTY-SY bearings. The test bearing was an LDK UER204 bearing, and the relevant parameters are shown in Table 1. The test designs 3 kinds of working conditions, and 5 bearings are arranged under each kind of working condition.
Table 1 LDK uer204 bearing parameter table
In order to test the comprehensive prediction performance of the method, the residual service life of the bearing is predicted under the working condition that the rotating speed is 2100r/min and the radial force is 12 kN.
In this embodiment, the method includes the following steps:
S1, respectively acquiring vibration acceleration signal training data x S (t) and data x T (t) to be predicted of a bearing in the same working condition, wherein the method specifically comprises the following steps of:
S1.1, establishing a three-dimensional space coordinate system according to a right-hand rule: the X axis is parallel to the axial direction of the shaft where the bearing is positioned, and the positive direction points to the side of the motor; the Z axis is vertical to the ground, and the positive direction is vertical upwards; the Y-axis positive direction is determined by a right hand rule;
S1.2, installing a sensor: two unidirectional acceleration sensors are respectively fixed on the horizontal direction and the vertical direction of the bearing to be measured through a magnetic seat and are used for measuring vibration acceleration data of the bearing to be measured; the sensor, the data acquisition system and the computer are correctly connected;
S1.3, setting data acquisition parameters: setting the total sampling duration as T, setting the sampling frequency as f s =25.6kHz, and recording the acquired time domain vibration acceleration signal as x (T) at the corresponding sampling time interval Deltat=1 min and the sampling point number N=32768;
S1.4, distinguishing the acquired time domain vibration acceleration signals x (t) of the rolling bearing according to time, wherein the time domain vibration acceleration signals x (t) in the former part of time are taken as training data x S (t), and the time domain vibration acceleration signals x (t) in the latter part of time are taken as data x T (t) to be predicted.
The specific data partitioning is shown in table 2.
Table 2 data dividing table
Data | Training data | Test data |
Working condition 1 | Bearing 1_1 bearing 1_2 bearing 1_3 bearing 1_4 | Bearing 1_5 |
S2, after standardized processing is carried out on the training data and the data to be predicted acquired in the step S1, carrying out data preprocessing on the training data and the data to be predicted by using a time window embedding strategy to acquire a training set and a set to be predicted;
In order to minimize the influence of the difference of the full life cycles of different bearings under the same working condition on the prediction precision of the provided model, the acquired training data x S (t) and the data x T (t) to be predicted are respectively subjected to standardized processing, and the standardized formula is as follows:
Wherein x t is the original data obtained by collection, and comprises training data x S (t) and data x T (t) to be predicted; mu is the mean value of the original data, sigma is the standard deviation of the original data, and x t norm represents the normalized original data;
After normalization, the obtained normalized training data and the data to be predicted are subjected to data preprocessing by utilizing a time window embedding strategy to obtain a training set X S and a set X T to be predicted, wherein the training set and the set to be predicted obtained by the time window embedding strategy are respectively composed of vibration signals with a certain time window size and signals with the previous S-1 step sizes, and the signals are expressed as follows:
xt input=(xt-S+1,...,xt-1,xt);
Wherein, For the data at time t after normalization,/>For the normalized data at time t-S+1, x t input is the sample at time t, and S is the total step size when the time window slides.
4. The method according to claim 1, wherein in step S3, the multi-module U-BiLSTM life prediction neural network model includes an encoder, a decoder, and a regressor connected in sequence;
the encoder comprises a first module and a second module which are sequentially connected, wherein the first module is a two-dimensional convolutional neural network and extracts shallow features of vibration signal data; the second module comprises a residual block and a two-dimensional depth convolution separable nerve network which are connected in sequence;
the decoder comprises a third module, wherein the third module comprises a deconvolution layer, a pooling layer and a batch normalization layer which are sequentially connected;
The regressor comprises a bidirectional long-short-time memory neural network and a full-connection layer which are connected in sequence;
Wherein the encoder and the decoder are connected linearly in sequence and are collectively referred to as a multi-module U network, and the regressor is referred to as BiLSTM network;
The input data is extracted through the multi-module U network characteristics to obtain output data, and the obtained output data is input to a BiLSTM network to realize prediction; after the output data is input into BiLSTM networks, the output c t at the current moment is obtained through a bidirectional long-short-time memory neural network, and the formula is as follows:
ft=σg(Wf(xt,ct-1)+bf)
it=σg(Wi(xt,ct-1)+bi)
ot=tanh(Wo(xt,ct-1)+bo)
ct=ft*ct-1+it*ot
Wherein f t is forget gate output, i t is input gate output, o t is output gate output, and c t is output at the current time t; σ g is a sigmoid activation function; tanh (·) is the tanh activation function; w f、Wi、Wo is a forgetting gate weight matrix, an input gate weight matrix and an output gate weight matrix respectively; b f、bi、bo is forgetting gate bias weight, input gate bias weight and output gate bias weight respectively; the bidirectional long-short-time memory neural network comprises a plurality of neurons which are connected in sequence, each neuron corresponds to a moment, c t represents the output of the current t-th neuron, namely the output of the current moment t, c t-1 represents the output of the t-1 th neuron, namely the output of the last moment t-1, and c t-2 represents the output of the t-2 nd neuron, namely the output of the last moment t-2;
The output c t obtained by the bidirectional long-short time memory neural network at the current time t is input into the full-connection layer to realize prediction, and the formula is as follows:
z=W*ct+b;
Wherein z is output data of the full connection layer, c t is input data of the full connection layer, and W and b are respectively a weight matrix and a bias matrix of the full connection layer.
5. The method according to claim 1, wherein step S4 comprises the steps of:
S4.1, taking the percentage of the actual residual service life (RUL) of the rolling bearing in the whole life cycle as a corresponding label, and converting the formula into:
wherein, T sta is the starting running time of the bearing, FPT is the time from the healthy state to the degradation state of the bearing, T end is the complete failure time of the bearing, T is the current running time, and RUL true is the normalized label value;
s4.2, setting a mean square error function as a loss function of a multi-module U-BiLSTM neural network life prediction model, wherein the formula is as follows:
Wherein L is a loss function, y i is a predicted value of the residual service life of the bearing to be predicted in the ith data to be predicted, The measured value of the residual service life of the bearing to be predicted in the ith data to be predicted is m, which is the number of data quantity;
s4.3, optimizing a multi-module U-BiLSTM neural network life prediction model by using an Adam optimizer, and calculating an update step size by comprehensively considering a first moment estimation (average value of gradient) and a second moment estimation (non-centralized variance of gradient) of the gradient; the formula is:
mt=β1mt-1+(1-β1)gt
vt=β1vt-1+(1-β1)gt 2
Where m t is the first moment estimate of the current time step, v t is the second moment estimate of the current time step, m t-1 is the first moment estimate of the last time step, v t is the second moment estimate of the last time step, g t is the gradient, Correction value estimated for the first moment of the current time step,/>Correction value estimated for the second moment of the current time step,/>For learning rate, β 1、β2, ε are constants, θ t is the dynamic constraint value for the current time step, and θ t+1 is the dynamic constraint value for the next time step.
S5, respectively inputting the set to be predicted into a multi-module U-BiLSTM neural network life prediction model after training is finished, and outputting a prediction result of the residual service life of the bearing to be predicted. Meanwhile, the existing bearing life prediction method (method II) for converting one-dimensional signals into two-dimensional images and extracting deep features by using a U-Net neural network is used for comparing prediction results. As can be seen from fig. 2-3, the residual life prediction result curve of the method of the present invention is closer to the true residual life curve than the comparative method.
Setting a prediction result evaluation index, and measuring the accuracy of the prediction result by using Root-mean-square-error (RMSE) of the prediction result and a true value; the method and the method II are used for measuring the prediction speed by the time required for completing the prediction. The RMSE values for the inventive and comparative methods are shown in table 3. The time required for the completion of the prediction by the method and the comparison method is shown in Table 4.
Table 3 RMSE value comparison
Table 4 comparison of the time taken to complete the prediction
Bearing | Bearing 1_5 |
The method of the invention | 16min36s |
Method II | 251min20s |
As can be seen from tables 3 and 4, the inventive method has significantly lower overall prediction error, i.e. higher prediction accuracy, than the comparative method, i.e. the RMSE value is lower; compared with a comparison method, the method provided by the invention has the advantages that the time required for completing prediction is shorter, and the prediction speed is faster. By the comparison, the method has better comprehensive prediction performance.
Example 2: under the second working condition experiment, the life prediction of the residual service life of the bearing (comparison method)
The rotation speed of the second working condition is 2250r/min, the radial force is 11kN, the training data are the bearing 2_1, the bearing 2_2, the bearing 2_3 and the bearing 2_5, the test data are the bearing 2_4, and the rest test settings and the step flow are the same as those of the embodiment 1.
The evaluation index of the prediction result was the same as in example 1. The RMSE values for the inventive and comparative methods are shown in table 5. The time required for the prediction of the method and the comparison method is shown in Table 6.
Table 5 RMSE value comparison
Bearing | Bearing 2_4 |
The method of the invention | 2.307 |
Method II | 26.124 |
Table 6 comparison of the time taken to complete the prediction
Bearing | Bearing 2_4 |
The method of the invention | 10min38s |
Method II | 239min36s |
As can be seen from fig. 4-5, the residual life prediction result curve of the method of the present invention is closer to the true residual life curve than the comparative method. As can be seen from tables 5 and 6, the inventive method has significantly lower overall prediction error, i.e. higher prediction accuracy, than the comparative method, i.e. the RMSE value is lower; compared with a comparison method, the method provided by the invention has the advantages that the time required for completing prediction is shorter, and the prediction speed is faster. By the comparison, the method has better comprehensive prediction performance.
Example 3: under the third experiment of working condition, the life prediction of the residual service life of the bearing (comparison method)
The rotation speed of the third working condition is 2400r/min, the radial force is 10kN, the training data are the bearing 3_1, the bearing 3_2, the bearing 3_4 and the bearing 3_5, the test data are the bearing 3_3, and the rest test settings and the step flow are the same as those of the embodiment 1.
The evaluation index of the prediction result was the same as in example 1. The RMSE values for the inventive and comparative methods are shown in table 7. The time required for the prediction of the method and the comparison method is shown in Table 8.
Table 7 RMSE value comparison
Bearing | Bearing 3_3 |
The method of the invention | 30.530 |
Method II | 52.853 |
Table 8 comparison of the time taken to complete the prediction
Bearing | Bearing 3_3 |
The method of the invention | 12min20s |
Method II | 246min47s |
As can be seen from fig. 6-7, the residual life prediction result curve of the method of the present invention is closer to the true residual life curve than the comparative method. As can be seen from tables 7 and 8, the inventive method has significantly lower overall prediction error, i.e. higher prediction accuracy, than the comparative method, i.e. the RMSE value is lower; compared with a comparison method, the method provided by the invention has the advantages that the time required for completing prediction is shorter, and the prediction speed is faster. By the comparison, the method has better comprehensive prediction performance.
In summary, the bearing life prediction method based on the multi-module U-BiLSTM network has the following advantages:
(1) Compared with the conventional data driving method in the life prediction field, the bearing life prediction method combining the multi-module U network and the bidirectional long-short-time memory neural network can extract deep features of long-time sequences and further improve the utilization rate of long-time sequence historical information.
(2) Compared with the existing novel method for predicting the service life of converting one-dimensional signals into two-dimensional images and extracting deep features by utilizing a U-Net neural network, the method introduces residual blocks into the encoder of the U-Net neural network, realizes multi-scale feature fusion, improves the capability of the U-Net neural network for processing long-time sequences, and reduces the requirement on the image size; the pooling layer and the normalization layer are introduced into a U-Net neural network decoder, so that the model prediction speed is improved; the multi-module U network is combined with the bidirectional long-short-time memory neural network, so that the utilization rate of long-time sequence historical information is improved, and the comprehensive prediction performance of the invention is better.
The above description is only of the preferred embodiments of the present invention, but the protection scope of the present invention is not limited thereto, and any person skilled in the art can make equivalent substitutions or modifications according to the technical scheme and the inventive concept thereof within the scope of the present invention disclosed in the present invention, and all those skilled in the art belong to the protection scope of the present invention.
Claims (8)
1. The bearing life prediction method based on the multi-module U-BiLSTM network is characterized by comprising the following steps of:
s1, respectively acquiring vibration acceleration signal training data x S (t) and data x T (t) to be predicted of a rolling bearing in the same working condition;
s2, after standardized processing is carried out on the training data and the data to be predicted acquired in the step S1, carrying out data preprocessing on the training data and the data to be predicted by using a time window embedding strategy to acquire a training set and a set to be predicted;
S3, constructing a multi-module U-BiLSTM neural network life prediction model, wherein the multi-module U network is used for feature extraction, and the BiLSTM network is used for regression prediction;
the multi-module U-BiLSTM neural network life prediction model comprises an encoder, a decoder and a regressor which are connected in sequence;
the encoder comprises a first module and a second module which are sequentially connected, wherein the first module is a two-dimensional convolutional neural network and extracts shallow features of vibration signal data; the second module comprises a residual block and a two-dimensional depth convolution separable nerve network which are connected in sequence;
the decoder comprises a third module, wherein the third module comprises a deconvolution layer, a pooling layer and a batch normalization layer which are sequentially connected;
The regressor comprises a bidirectional long-short-time memory neural network and a full-connection layer which are connected in sequence;
Wherein the encoder and the decoder are connected linearly in sequence and are collectively referred to as a multi-module U network, and the regressor is referred to as BiLSTM network;
The input data is extracted through the multi-module U network characteristics to obtain output data, and the obtained output data is input to a BiLSTM network to realize prediction; after the output data is input into BiLSTM networks, the output c t at the current moment is obtained through a bidirectional long-short-time memory neural network, and the formula is as follows:
ft=σg(Wf(ct-2,ct-1)+bf);
it=σg(Wi(ct-2,ct-1)+bi);
ot=tanh(Wo(ct-2,ct-1)+bo);
ct=ft*ct-1+it*ot;
Wherein f t is forget gate output, i t is input gate output, o t is output gate output, and c t is output at the current time t; σ g is a sigmoid activation function; tanh (·) is the tanh activation function; w f、Wi、Wo is a forgetting gate weight matrix, an input gate weight matrix and an output gate weight matrix respectively; b f、bi、bo is forgetting gate bias weight, input gate bias weight and output gate bias weight respectively; the bidirectional long-short-time memory neural network comprises a plurality of neurons which are connected in sequence, each neuron corresponds to a moment, c t represents the output of the current t-th neuron, namely the output of the current moment t, c t-1 represents the output of the t-1 th neuron, namely the output of the last moment t-1, and c t-2 represents the output of the t-2 nd neuron, namely the output of the last moment t-2;
The output c t obtained by the bidirectional long-short time memory neural network at the current time t is input into the full-connection layer to realize prediction, and the formula is as follows:
z=W*ct+b;
Wherein z is output data of the full-connection layer, c t is input data of the full-connection layer, and W and b are respectively a weight matrix and a bias matrix of the full-connection layer;
S4, taking the percentage of the actual residual service life of the rolling bearing in the whole service life period as a corresponding label, inputting a training set as input data into a multi-module U-BiLSTM neural network service life prediction model for iterative training until a loss function reaches a preset value, and obtaining a trained multi-module U-BiLSTM neural network service life prediction model;
S5, inputting the set to be predicted into a multi-module U-BiLSTM neural network life prediction model after training is completed, and outputting a prediction result of the residual service life of the bearing.
2. The method for predicting bearing life based on the multi-module U-BiLSTM network according to claim 1, wherein step S1 specifically includes the steps of:
S1.1, establishing a three-dimensional space coordinate system according to a right-hand rule: the X axis is parallel to the axial direction of the shaft where the bearing is positioned, and the positive direction points to the side of the motor; the Z axis is vertical to the ground, and the positive direction is vertical upwards; the Y-axis positive direction is determined by a right hand rule;
s1.2, installing a sensor: two unidirectional acceleration sensors are respectively fixed on the horizontal direction and the vertical direction of the rolling bearing to be measured through a magnetic seat and are used for measuring vibration acceleration data of the rolling bearing to be measured; the sensor, the data acquisition system and the computer are correctly connected;
S1.3, setting data acquisition parameters: setting the total sampling time length as T and the sampling frequency as f s, then setting the corresponding sampling time interval delta t=1/f s, the sampling point number N=f s.T, and recording the acquired time domain vibration acceleration signal as x (T);
S1.4, distinguishing the acquired time domain vibration acceleration signals x (t) of the rolling bearing according to time, wherein the time domain vibration acceleration signals x (t) in the former part of time are taken as training data x S (t), and the time domain vibration acceleration signals x (t) in the latter part of time are taken as data x T (t) to be predicted.
3. The method for predicting the service life of a bearing based on a multi-module U-BiLSTM network according to claim 1, wherein in step S2, in order to minimize the influence of the difference of the total life cycles of different bearings under the same working condition on the prediction accuracy of the proposed model, the collected training data x S (t) and the data x T (t) to be predicted are respectively subjected to standardized processing, and the standardized formula is as follows:
Wherein x t is the original data obtained by collection, and comprises training data x S (t) and data x T (t) to be predicted; μ is the mean of the raw data, σ is the standard deviation of the raw data, and x t norm represents the normalized raw data.
4. The method for predicting bearing life based on multi-module U-BiLSTM network according to claim 3, wherein in step S2, after normalization, the obtained normalized training data and the data to be predicted are subjected to data preprocessing by using a time window embedding strategy to obtain a training set X S and a set to be predicted X T, and the training set and the set to be predicted obtained by the time window embedding strategy are respectively composed of vibration signals with a certain time window size and signals with S-1 step sizes before the vibration signals, and are expressed as:
Wherein, For the data at time t after normalization,/>For the normalized data at time t-S+1, x t input is the sample at time t, and S is the total step size when the time window slides.
5. The method for predicting bearing life based on a multi-module U-BiLSTM network according to any one of claims 1 to 4, wherein step S4 includes the steps of:
S4.1, taking the percentage of the actual residual service life (RUL) of the rolling bearing in the whole life cycle as a corresponding label;
s4.2, setting a mean square error function as a loss function of a multi-module U-BiLSTM neural network life prediction model;
And S4.3, optimizing the multi-module U-BiLSTM neural network life prediction model by using an Adam optimizer.
6. The method for predicting bearing life based on a multi-module U-BiLSTM network according to claim 5, wherein in step S4.1, the conversion formula is:
Wherein, T sta is the starting running time of the bearing, FPT is the time from the healthy state to the degraded state of the bearing, T end is the complete failure time of the bearing, T is the current running time, and RUL true is the normalized label value.
7. The method for predicting bearing life based on a multi-module U-BiLSTM network as claimed in claim 6, wherein in step S4.1, the formula of the loss function is:
Wherein L is a loss function, y i is a predicted value of the residual service life of the bearing to be predicted in the ith data to be predicted, And m is the number of data quantities and is the actual measurement value of the residual service life of the bearing to be predicted in the ith data to be predicted.
8. The method for predicting bearing life based on a multi-module U-BiLSTM network according to claim 7, wherein in step S4.3, an update step is calculated by comprehensively considering the first moment estimate and the second moment estimate of the gradient; the formula is:
mt=β1mt-1+(1-β1)gt;
vt=β1vt-1+(1-β1)gt 2;
Where m t is the first moment estimate of the current time step, v t is the second moment estimate of the current time step, m t-1 is the first moment estimate of the last time step, v t-1 is the second moment estimate of the last time step, g t is the gradient, Correction value estimated for the first moment of the current time step,/>Correction value estimated for the second moment of the current time step,/>For learning rate, β 1、β2, ε are constants, θ t is the dynamic constraint value for the current time step, and θ t+1 is the dynamic constraint value for the next time step.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210302001.6A CN114626625B (en) | 2022-03-25 | 2022-03-25 | Bearing life prediction method based on multi-module U-BiLSTM network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210302001.6A CN114626625B (en) | 2022-03-25 | 2022-03-25 | Bearing life prediction method based on multi-module U-BiLSTM network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114626625A CN114626625A (en) | 2022-06-14 |
CN114626625B true CN114626625B (en) | 2024-06-21 |
Family
ID=81903365
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210302001.6A Active CN114626625B (en) | 2022-03-25 | 2022-03-25 | Bearing life prediction method based on multi-module U-BiLSTM network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114626625B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN117077327A (en) * | 2023-10-18 | 2023-11-17 | 国网山东省电力公司鱼台县供电公司 | Bearing life prediction method and system based on digital twin |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112129534A (en) * | 2020-09-17 | 2020-12-25 | 上海交通大学 | Method and system for predicting service life of main shaft bearing of machine tool |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112152201B (en) * | 2020-08-25 | 2021-09-17 | 国网山东综合能源服务有限公司 | Electricity load prediction method and system based on convolution length time memory neural network |
CN112364567B (en) * | 2020-11-18 | 2022-04-22 | 浙江大学 | Residual life prediction method based on consistency check of similarity of degraded tracks |
CN113723007B (en) * | 2021-09-08 | 2023-09-15 | 重庆邮电大学 | Equipment residual life prediction method based on DRSN and sparrow search optimization |
-
2022
- 2022-03-25 CN CN202210302001.6A patent/CN114626625B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112129534A (en) * | 2020-09-17 | 2020-12-25 | 上海交通大学 | Method and system for predicting service life of main shaft bearing of machine tool |
Non-Patent Citations (1)
Title |
---|
深度学习在故障诊断与预测中的应用;余萍;曹洁;;计算机工程与应用;20200331(第03期);第6-23页 * |
Also Published As
Publication number | Publication date |
---|---|
CN114626625A (en) | 2022-06-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114626625B (en) | Bearing life prediction method based on multi-module U-BiLSTM network | |
CN112417954B (en) | Bearing fault mode diagnosis method and system for small sample data set | |
CN112629863B (en) | Bearing fault diagnosis method for dynamic joint distribution alignment network under variable working conditions | |
CN111523081B (en) | Aeroengine fault diagnosis method based on enhanced gate control circulating neural network | |
CN110110768B (en) | Rolling bearing fault diagnosis method based on parallel feature learning and multiple classifiers | |
CN113203566B (en) | Motor bearing fault diagnosis method based on one-dimensional data enhancement and CNN | |
CN113008559B (en) | Bearing fault diagnosis method and system based on sparse self-encoder and Softmax | |
CN109946080B (en) | Mechanical equipment health state identification method based on embedded circulation network | |
CN110991471B (en) | Fault diagnosis method for high-speed train traction system | |
CN112070322B (en) | High-voltage cable line running state prediction method based on long-short term memory network | |
CN113673346A (en) | Motor vibration data processing and state recognition method based on multi-scale SE-Resnet | |
CN111459144A (en) | Airplane flight control system fault prediction method based on deep cycle neural network | |
CN116793682A (en) | Bearing fault diagnosis method based on iCORAL-MMD and anti-migration learning | |
CN111898686A (en) | Bearing fault identification method based on gated cyclic unit network | |
CN115290326A (en) | Rolling bearing fault intelligent diagnosis method | |
CN115508073B (en) | Prototype adaptive mechanical equipment fault diagnosis method based on multi-scale attention | |
CN112834221A (en) | Bearing fault diagnosis method based on EEMD-MSE and multi-scale one-dimensional convolution neural network | |
Zhang et al. | Intelligent machine fault diagnosis using convolutional neural networks and transfer learning | |
CN112766200A (en) | Planetary gear box small sample fault diagnosis method based on one-dimensional convolutional neural network | |
CN113850161A (en) | Flywheel fault identification method based on LSTM deep noise reduction self-encoder | |
CN113758709A (en) | Rolling bearing fault diagnosis method and system combining edge calculation and deep learning | |
CN112069621B (en) | Method for predicting residual service life of rolling bearing based on linear reliability index | |
CN114861349A (en) | Rolling bearing RUL prediction method based on model migration and wiener process | |
CN114841262A (en) | Rolling bearing fault diagnosis method based on DS evidence theory | |
CN113435321A (en) | Method, system and equipment for evaluating state of main shaft bearing and readable storage medium |
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 |