CN111783242B - RVM-KF-based rolling bearing residual life prediction method and device - Google Patents
RVM-KF-based rolling bearing residual life prediction method and device Download PDFInfo
- Publication number
- CN111783242B CN111783242B CN202010554878.5A CN202010554878A CN111783242B CN 111783242 B CN111783242 B CN 111783242B CN 202010554878 A CN202010554878 A CN 202010554878A CN 111783242 B CN111783242 B CN 111783242B
- Authority
- CN
- China
- Prior art keywords
- data
- rolling bearing
- rvm
- prediction
- 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
- 238000005096 rolling process Methods 0.000 title claims abstract description 114
- 238000000034 method Methods 0.000 title claims abstract description 64
- 238000006731 degradation reaction Methods 0.000 claims abstract description 67
- 230000015556 catabolic process Effects 0.000 claims abstract description 61
- 238000012544 monitoring process Methods 0.000 claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 20
- 230000015654 memory Effects 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 10
- 230000000694 effects Effects 0.000 claims description 8
- 230000005653 Brownian motion process Effects 0.000 claims description 4
- 238000005537 brownian motion Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 description 14
- 239000011159 matrix material Substances 0.000 description 7
- 238000001914 filtration Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 238000012545 processing Methods 0.000 description 6
- 238000012360 testing method Methods 0.000 description 6
- 238000012706 support-vector machine Methods 0.000 description 5
- 230000007774 longterm Effects 0.000 description 4
- 238000005259 measurement Methods 0.000 description 4
- 239000013598 vector Substances 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 229910021389 graphene Inorganic materials 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/17—Mechanical parametric or variational design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/04—Ageing analysis or optimisation against ageing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Geometry (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Hardware Design (AREA)
- General Engineering & Computer Science (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention relates to a rolling bearing residual life prediction method and device based on RVM-KF. The method comprises the following steps: acquiring real-time monitoring data of a rolling bearing to be predicted and historical data of the whole life cycle of the rolling bearing similar to the rolling bearing to be predicted; training an RVM model using the historical data; determining a prediction starting moment according to the real-time monitoring data and the SKF, and determining a state equation of the Kalman filter according to the real-time monitoring data; inputting the predicted starting moment into a trained RVM model to obtain estimated vibration data, taking the estimated vibration data as an observation value of a Kalman filter, carrying out single-step prediction, and adding single-step predicted data into historical data to update the RVM model; and carrying out iterative prediction by using the updated RVM model until vibration data in the single-step prediction data exceeds a failure threshold value to complete the prediction of the residual life. The invention combines RVM model, kalman filter and SKF to realize the accurate prediction of the residual life of the rolling bearing in the accelerated degradation stage.
Description
Technical Field
The invention relates to a rolling bearing residual life prediction method and device based on RVM-KF, and belongs to the technical field of fault prediction and health management.
Background
The rolling bearing is used as a basic part widely applied to mechanical equipment and is one of key parts for determining safe and reliable operation of the mechanical equipment. According to the related investigation, it was found that about 30% of mechanical failures in rotating machinery were the result of rolling bearing failures. The mechanical failure not only can cause economic loss, but also can cause casualties, so that the residual service life (REMAINING USEFUL LIFE, RUL) of the rolling bearing is accurately predicted, and serious problems such as main unit failure, shutdown maintenance, productivity loss, casualties and the like caused by the failure of the rolling bearing can be effectively avoided.
The current residual life prediction of the rolling bearing is mainly divided into a model-based residual life prediction method and a data-based residual life prediction method.
The residual life prediction method based on the model utilizes Paris crack propagation models, forman crack propagation law, exponential models, random models and the like to describe the degradation process of the bearing by analyzing failure mechanisms of the bearing or summarizing experiences, and then carries out the residual life prediction of the bearing, but the degradation mechanism of the rolling bearing is difficult to accurately obtain and the selection of the model has great influence on the prediction result.
The method based on the data does not need to establish a physical model or a statistical model, but obtains the relation between the degradation state of the bearing and the residual life through a neural network, an SVM (support vector machine), an RVM (correlation vector machine) and other models in the existing monitoring data, so that the residual life of the rolling bearing is predicted, the full life test data is fully utilized by the method based on the data, and the relation between the degradation state of the bearing and the residual life of the rolling bearing is accurately established by a large amount of high-quality data, but the method cannot reflect individual differences.
For this reason, a lifetime prediction method that considers individual differences and overall characteristics of the entire lifetime has been proposed, for example: the journal is "the journal of the war industry", the journal number is 39, the journal number is 5, the journal number is 2018, the journal number is named as "the study of the model of the rest life of the mechanical parts based on the support vector machine and Kalman filtering", the method builds a nonlinear Kalman filtering state updating equation according to the SVM regression model obtained by training the prior full life test data, builds a time updating equation according to the degradation characteristics of the mechanical parts, sets an initial rest life value and variance thereof, and calculates the rest life estimated value at each moment and a confidence interval with a certain confidence level by gradual iteration. The calculation model can fully utilize the full life test data of the existing parts and the similar parts and the real-time state degradation data of the predicted parts to realize the prediction of the residual life, thereby achieving the purpose that the obtained residual life prediction model can utilize the full life data and consider the individual difference of a research object (Kalman filtering model).
However, in order to obtain an accurate nonlinear Kalman filtering state equation, the method has higher requirements on the quality of training samples and the accuracy of SVM models, namely, the dependence on models and data is larger, the accuracy in long-term prediction is not high, and the engineering requirements can not be effectively met.
Disclosure of Invention
The application aims to provide a rolling bearing residual life prediction method based on RVM-KF, which is used for solving the problem of inaccurate residual life prediction in the prior art; meanwhile, the device for predicting the residual life of the rolling bearing based on RVM-KF is also provided, so that the problem of inaccurate residual life prediction in the prior art is solved.
In order to achieve the above purpose, the invention provides a technical scheme of a rolling bearing residual life prediction method based on RVM-KF, which comprises the following steps:
1) Acquiring real-time monitoring data of a rolling bearing to be predicted and historical data of the whole life cycle of the rolling bearing similar to the rolling bearing to be predicted; the real-time monitoring data and the historical data refer to vibration data and corresponding time;
2) Training an RVM model using the historical data; the RVM model is input as time and output as vibration data;
3) Determining a prediction starting moment according to the real-time monitoring data and the SKF; determining a state equation of the Kalman filter according to the real-time monitoring data;
4) Inputting the predicted starting moment into a trained RVM model to obtain estimated vibration data, carrying out single-step prediction by taking the estimated vibration data as an observation value of a Kalman filter to obtain single-step predicted data, if the vibration data in the single-step predicted data exceeds a failure threshold value, completing prediction of the residual life, if the vibration data in the single-step predicted data does not exceed the failure threshold value, adding the single-step predicted data into the historical data of the step 2), retraining the RVM model, and updating the RVM model; the step is iteratively performed until vibration data in the single-step prediction data exceeds a failure threshold.
In addition, the invention also provides a technical scheme of the rolling bearing residual life prediction device based on the RVM-KF, which comprises a processor, a memory and a computer program which is stored in the memory and can run on the processor, wherein the processor realizes the technical scheme of the rolling bearing residual life prediction method based on the RVM-KF when executing the computer program.
The RVM-KF-based rolling bearing residual life prediction method and device have the beneficial effects that: according to the method, an RVM model is built through historical data of the same kind of rolling bearing, the moment when the rolling bearing to be predicted enters accelerated degradation is determined according to real-time monitoring data and SKF of the rolling bearing to be predicted, the prediction starting moment is further determined, meanwhile, a state equation of a Kalman filter is determined according to the real-time monitoring data, vibration data is estimated according to the prediction starting moment in combination with the built RVM model, the estimated value is used as an observation value of the Kalman filter to conduct single-step prediction, when the vibration data of the single-step prediction does not exceed a failure threshold value, the prediction data is added into the RVM model, the RVM model is retrained, and iteration prediction is conducted until the vibration data of the single-step prediction exceeds the failure threshold value, and the iteration is ended, so that the residual life is obtained. According to the invention, by continuously updating the RVM model and combining the RVM model, the Kalman filter and the SKF, the accurate prediction of the residual life of the rolling bearing in the accelerated degradation stage is realized, the uncertainty of long-term prediction of the residual life of the rolling bearing is reduced, the precision of long-term prediction of the residual life of the rolling bearing is improved, and the dependence of the prediction process on data and the model is effectively reduced by combining the Kalman filter and the state equation.
Furthermore, in the method and the device for predicting the residual life of the rolling bearing based on RVM-KF, in order to better fit and accelerate the degradation process, the state equation of the Kalman filter is a random effect index model.
Further, in the method and the device for predicting the residual life of the rolling bearing based on RVM-KF, the random effect index model is as follows:
Wherein Y (t) is vibration data at time t; alpha is a constant; θ and β are parameters; σ 2 is variance; epsilon (t) is the error term and is Brownian motion with a mean of 0 and a variance of σ 2.
Further, in the method and the device for predicting the residual life of the rolling bearing based on RVM-KF, the vibration data comprises at least one characteristic value extracted from the vibration signal.
Furthermore, in the method and the device for predicting the residual life of the rolling bearing based on the RVM-KF, in order to improve the accuracy of prediction and reduce the complexity of model training, the historical data utilized in training the RVM model is the historical data of the accelerated degradation stage.
Drawings
Figure 1 is a flow chart of the RVM-KF based rolling bearing residual life prediction method of the present invention;
FIG. 2 is a degradation curve of a rolling bearing of the present invention;
FIG. 3 is a graph comparing the predicted degradation trajectory of a rolling bearing 2-3 of the present invention with the predicted degradation trajectory of a rolling bearing 2-3 of the prior art;
FIG. 4 is a graph comparing the predicted degradation tracks of a rolling bearing 2-5 of the present invention with the predicted degradation tracks of a prior art rolling bearing 2-5;
FIG. 5 is a graph comparing the predicted remaining life of the rolling bearing 2-3 of the present invention and the predicted remaining life of the rolling bearing 2-3 of the prior art;
FIG. 6 is a graph comparing the predicted remaining life of a rolling bearing 2-5 of the present invention with the predicted remaining life of a rolling bearing 2-5 of the prior art;
Fig. 7 is a schematic structural view of a rolling bearing residual life prediction device based on RVM-KF according to the present invention.
Detailed Description
RVM-KF-based rolling bearing residual life prediction method embodiment:
The main conception of the rolling bearing residual life prediction method based on RVM-KF is that the invention combines RVM model, kalman filter and SKF to carry out iterative prediction on the residual life of the rolling bearing entering the accelerated degradation stage, thereby improving the accuracy of life prediction.
Specifically, the method for predicting the residual life of the rolling bearing based on RVM-KF is shown in figure 1, and comprises the following steps:
1) Historical data of the entire life cycle of the rolling bearing similar to the rolling bearing to be predicted is collected, wherein the historical data comprises vibration signals and corresponding time (the time is the use time).
In order to reduce individual differences, life predictions are made for rolling bearings of different models using different model parameters, so rolling bearings corresponding to historical usage data selected during modeling should be of the same type (i.e., the same type) as the rolling bearings to be predicted.
And carrying out a full life cycle test on a plurality of rolling bearings with the same type under the same test condition, periodically acquiring vibration signals (1.28 s vibration signals are acquired every 1 min) in the working process of the rolling bearings in the test process until the rolling bearings are completely invalid, and obtaining full life vibration signals of the rolling bearings, wherein each vibration signal has corresponding service time.
2) Extracting a characteristic value capable of effectively representing the degradation of the rolling bearing from the vibration signal in the step 1).
In order to realize prediction of the remaining life, characteristic values representing degradation of the rolling bearing are extracted from the vibration signals as vibration data, however, the vibration signals contain various vibration data, and other vibration data can be extracted under the condition that the vibration data representing degradation can be obtained by calculating other vibration data, and the invention is not limited to the above.
3) Inputting the characteristic values obtained in the step 2) and the time corresponding to each characteristic value in the step 1) into a pre-established SKF model, and finding out the accelerated degradation moment (namely early fault point) of the similar rolling bearing in the life cycle.
The degradation process of the rolling bearing comprises a stable degradation stage and an accelerated degradation stage as shown in fig. 2, wherein the SKF model is a dynamic Bayesian network and comprises two Kalman filters, the two Kalman filters comprise a first Kalman filter for predicting that a certain moment belongs to the stable degradation probability, and the rolling bearing is suitable for the stable degradation stage; and a second Kalman filter predicting the probability of the accelerated degradation at a certain moment, which is suitable for the accelerated degradation stage. The data input by the SKF can obtain the first probability that each moment belongs to a stable degradation stage through a first Kalman filter, and can obtain the second probability that each moment belongs to an accelerated degradation stage through a second Kalman filter.
The probability calculation process of the specific Kalman filter is as follows:
both the first and second kalman filters are basic kalman filters, and the prediction and update processes are as follows:
the prediction process is that the state prediction:
covariance prediction:
The updating process is that the residual error is measured:
Residual covariance:
Kalman gain:
and (5) updating the state:
Covariance update:
wherein y t is the characteristic value monitored at time t, For the characteristic value estimated at the time t, phi t is the state transition matrix at the time t,/>Is the transposed matrix of phi t, H t is the observation matrix at time t,/>Is the transpose matrix of H t, delta t is the measurement residual at time t, C t is the residual covariance at time t, C t -1 is the inverse matrix of C t, K t is the Kalman gain at time t, Q t is the variance of the process noise at time t, R t is the variance of the measurement noise at time t,/>For the estimated value of covariance, I is an identity matrix, x t is the estimated eigenvalue updated at time t, and P t is the estimated value of covariance updated at time t.
Then, the measurement residual of the ith Kalman filter follows a probability density function of a normal distribution with an average value of 0The method comprises the following steps:
The probability of obtaining each kalman filter is:
State weighting
Covariance weighting
Wherein,For the probability that the characteristic value belongs to the ith Kalman filter at the t moment, Z ij is the probability that the ith Kalman filter is transferred to the jth Kalman filter, n is the number of Kalman filters,/>Is the measurement residual error of the ith Kalman filter at the t moment,/>Is the residual covariance of the ith Kalman filter at time t,/>For the state of the ith Kalman filter update at time t, P t i is the covariance of the ith Kalman filter update at time t. And carrying the calculated state weights and covariance weights into the Kalman filtering process to obtain the most probable degradation state at each moment.
After the first probability and the second probability of each time are obtained, if the second probability of a plurality of continuous time (the plurality of time generally selects 5 time, of course, the number of the plurality of time can be set according to the need) is greater than the first probability, the time at the forefront of the plurality of time is taken as the accelerated degradation time, and of course, as other embodiments, the accelerated degradation time can be calculated according to the plurality of time, or the time at the rearmost end of the plurality of time is taken as the accelerated degradation time.
As other embodiments, other methods may be used to determine the stage in which the rolling bearing is to be predicted, for example: the stage can also be judged by comparing the characteristic value with the experience value. The invention is not limited in terms of how the stage in which the rolling bearing is to be predicted is determined.
4) And (3) taking the historical data obtained in the step (3) after the accelerated degradation moment as a training data set to train the RVM model.
RVM is a generalized linear model under Bayesian framework, and the training data set isThe RVM model established is:
ln＝y(tn;ω)+εn；
the likelihood function of the training samples is:
Wherein, l n is the nth target data (i.e. output data), here, the feature value; t n is the nth input data, here time (i.e., duration of use); n is the total number of data in the training data set; omega n is the weight corresponding to the nth data, and K (t, t n) is the kernel function corresponding to the nth data; epsilon n is Gaussian white noise with a mean value of 0 and a variance of sigma 2; phi is an n×n+1 matrix based on a kernel function, phi= [ phi (t 1),φ(t2),…,φ(tn),…,φ(tN)]T,φ(tn) is a vector element in K (t, t n) of kernel vectors ,φ(tn)＝[1,K(tn,t1),K(tn,t2),…K(tn,tN)]K(tn,t1),K(tn,t2),…,K(tn,tN) corresponding to nth data, and t is an input set; l is the output set; ω is the weight set, ω= (ω 1,ω2,…,ωn,…ωN)T.
The solution to ω and σ 2 directly using the maximum likelihood method results in mostly non-0 elements in ω, resulting in overfitting. To avoid this in RVMs, ω plus a precondition: their probability distribution is a normal distribution around 0:
Wherein α is a vector of n+1; alpha i is the ith hyper-parameter; n () is a normal distribution.
Thus, for a new given input t *, its predicted value l *:
p(l*|l)＝∫p(l*|ω,α,σ2)p(ω,α,σ2|l)dωdαdσ2。
In the step, the historical data after the accelerated degradation moment is optimized for training the RVM model, so that not only can the excessive data quantity be avoided, but also the RVM model can be estimated for the accelerated degradation stage, the estimated RVM model is more accurate, and as other embodiments, the RVM model can be trained by all the historical data, and the method is not limited in this aspect.
5) Acquiring real-time monitoring data of the rolling bearing to be predicted, wherein the real-time monitoring data comprises vibration signals and corresponding time (service time in the process); and extracting a real-time characteristic value capable of effectively representing the degradation of the rolling bearing from the vibration signal, judging the stage of the rolling bearing to be predicted according to the real-time characteristic value and the SKF, and determining the prediction starting moment according to the acceleration degradation moment if the rolling bearing to be predicted is judged to enter the acceleration degradation moment.
The predicted onset time is typically slightly later than the accelerated degradation time, for example: the time of accelerated degradation is the time corresponding to 200min, and the predicted starting time is the time corresponding to 220min, which can be set according to the requirement, and the invention is not limited.
The process of judging the stage in which the rolling bearing to be predicted is located is the same as that of step 3), except that the data input into the SKF is a real-time characteristic value, so that a detailed judgment process thereof is not repeated here.
6) And determining parameters of a state equation in a Kalman Filter (KF) according to the real-time monitoring data, and determining the state equation, wherein the state equation is a random effect index model.
The real-time monitoring data for determining the state equation parameters may be monitoring data near the accelerated degradation time, or monitoring data near the predicted starting time, which is not limited in the present invention. The state equation may be updated according to the update of the monitoring data before the prediction start time is not determined, but the state equation is unchanged after the prediction start time is determined.
In this embodiment, the degradation curve of the rolling bearing is shown in fig. 2, and includes a stable and stable degradation state and an accelerated degradation state, and it can be seen from the figure that the accelerated degradation curve is similar to an exponential function curve, so that the state equation of the kalman filter is represented by using a random effect exponential model, and the random coefficient exponential model is:
Wherein Y (t) is vibration data at time t (here, characteristic value of the rolling bearing); alpha is a constant; θ and β are parameters (i.e., random parameters); epsilon (t) is the error term and is Brownian motion with a mean of 0 and a variance of σ 2.
For convenient calculation, the random coefficient index model is logarithmized, and is deformed into:
In the above formula: θ' =lnθ, Θ 'and β' are parameters after deformation, ε (t) represents the error, which is the center Brownian motion with a mean of 0 and variance of σ 2.
As another embodiment, the state equation of the third kalman filter may be any existing model, and may conform to the process of accelerating degradation, which is not limited in the present invention.
7) Inputting the predicted starting time determined in the step 5) into the RVM model trained in the step 4), obtaining an estimated characteristic value of the rolling bearing to be predicted, taking the estimated characteristic value of the rolling bearing to be predicted as a measured value of a Kalman filter, carrying out single-step prediction to obtain single-step prediction data, adding the single-step prediction data into the training data set in the step 4), retraining the RVM model, and updating the RVM model.
The single-step prediction data includes a single-step predicted time: predicting a starting moment plus step length and a single-step predicted characteristic value;
8) And estimating the characteristic value at the next moment by using the updated RVM model, and carrying out iterative prediction until the characteristic value (namely the predicted characteristic value) in the single-step prediction data exceeds the failure threshold value, so as to complete the prediction of the residual life.
And 8) judging whether vibration data in the single-step prediction data exceeds a failure threshold value, if not, continuing iterative prediction, and if so, determining the residual life of the rolling bearing to be predicted according to the iteration times.
Each update of the RVM model in step 8) is to estimate the feature value of the next time of use, for example: determining the starting time of prediction to be 220min, obtaining an estimated characteristic value of 220min according to the RVM model trained in the step 4), and carrying out single-step prediction on the estimated characteristic value of 220min through KF to obtain single-step prediction data; the single-step prediction data comprises 220min and corresponding prediction characteristic values; if the predicted characteristic value corresponding to 220min does not exceed the failure threshold value, adding the single-step predicted data into the training data set in the step 4), retraining the RVM model, and updating the RVM model;
Then, estimating the characteristic value using 220min+1min according to the updated RVM model to obtain an estimated characteristic value using 221min, and carrying out single-step prediction on the estimated characteristic value using 221min through KF to obtain single-step prediction data; the single-step prediction data of the iteration comprises 221min and a corresponding prediction characteristic value;
And by analogy, performing iterative prediction, and if the predicted characteristic value in the single-step prediction data exceeds the failure threshold value after the characteristic value of 230min is estimated, then the residual life is 230min-220 min=10 min.
In order to obtain the residual life corresponding to different prediction starting moments, after the life prediction is completed at the first prediction starting moment, the life prediction is carried out again at the updated prediction starting moment, and at the moment, the state equation is also required to be updated, namely, the steps 6) -8 are repeated, so that the prediction of the residual life at the new prediction starting moment is completed. And further obtaining the residual life corresponding to different prediction starting moments.
The validity and correctness of the present invention is verified by the following examples, and dataset XJTU-SY is provided by Seisan transportation university (XJTU) institute of design science and basic building and Zhejiang Changxing Yang technology Co., ltd (SY).
The invention divides the degradation process of the rolling bearing into stable degradation and accelerated degradation, extracts the wavelet packet coefficient root mean square (eigenvalue) of 5 rolling bearings under the working condition 2 to identify the degradation state, the serial numbers of the 5 rolling bearings are respectively 2-1, 2-2, 2-3, 2-4 and 2-5, and the time when each rolling bearing enters the accelerated degradation stage is shown as the table one. The invention sets the failure threshold of the rolling bearing as 13, takes the rolling bearings with the serial numbers of 2-3 and 2-5 as the rolling bearing to be predicted, and takes the data of the rolling bearings with the serial numbers of 2-1, 2-2, 2-4 and 2-5 as the training set when the serial number of 2-3 is taken as the rolling bearing to be predicted; when the serial number 2-5 is used as the rolling bearing to be predicted, the data of the rolling bearings with serial numbers 2-1, 2-2, 2-3 and 2-4 are used as training sets; and estimating the degradation characteristic value of the rolling bearing to be predicted according to the trained RVM model by using the wavelet packet coefficient root mean square of the rest 4 rolling bearing accelerated degradation stages as an RVM training set, performing single-step prediction processing by using the estimated value of the RVM as a measured value of the KF, adding the KF filtering value into the RVM training set to retrain the RVM model for next iteration prediction until the failure threshold of the rolling bearing is reached, and completing the prediction of the rest life of the bearing. The degradation track and the residual life of the bearing are estimated by using RVM and KF and the iterative method of the invention, and the comparison results are shown in figures 3, 4, 5 and 6. And as shown in table two, the accuracy of the degradation track prediction by 3 methods is given by using the absolute error MAE and the root mean square error RMSE.
Absolute error:
root mean square error:
Where x i is the true degradation characteristic quantity of the bearing, y i is the predicted degradation characteristic quantity data, and n is the total number of data.
Table 5 accelerated degradation moments of Rolling bearing
| Bearing serial number | 2-1 | 2-2 | 2-3 | 2-4 | 2-5 |
| Entering into the accelerated degradation phase (min) | 456 | 55 | 305 | 31 | 124 |
Rolling bearing with sequence numbers 2-3 and 2-5 adopts different prediction methods of absolute error and root mean square error
As can be seen from a comparison of fig. 3 and 4, the predicted result of the method of the present invention (RVM-KF) is more similar to the true value. The invention is based on the SKF model, utilizes piecewise linear approximation rolling bearing nonlinear degradation, tracks the dynamic change of the rolling bearing degradation process, does not need to set a threshold value and a large amount of data training models, and adaptively obtains the most probable degradation model at each moment; RVM is a generalized linear model under Bayesian framework, is more suitable for processing small sample problems than neural network RVM, and can provide complete prediction distribution; meanwhile, the KF and the random effect index model are combined to better fit the degradation process of the bearing, so that the service life prediction is more accurate.
The method can accurately predict the residual service life of the bearing in early stage of the accelerated degradation in real time without accurate model and a large amount of high-quality data, and simultaneously reduces the uncertainty of long-term prediction.
RVM-KF-based rolling bearing residual life prediction device embodiment:
The rolling bearing residual life prediction device based on RVM-KF, as shown in figure 7, comprises a processor, a memory and a computer program stored in the memory and capable of running on the processor, wherein the processor realizes the rolling bearing residual life prediction method based on RVM-KF when executing the computer program.
The specific implementation process and effect of the rolling bearing residual life prediction method based on RVM-KF are described in the embodiment of the rolling bearing residual life prediction method based on RVM-KF, and are not described herein.
That is, the method in the above RVM-KF based rolling bearing remaining life prediction method embodiment should be understood that the flow of the RVM-KF based rolling bearing remaining life prediction method may be implemented by computer program instructions. These computer program instructions may be provided to a processor, such as a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus, etc., such that the instructions, which execute via the processor, create means for implementing the functions specified in the above-described method flows.
The processor in this embodiment refers to a microprocessor MCU or a processing device such as a programmable logic device FPGA;
The memory referred to in this embodiment is used for storing computer program instructions formed by implementing the RVM-KF-based rolling bearing residual life prediction method, and includes a physical device for storing information, where the information is typically stored in a medium using an electrical, magnetic or optical method after being digitized. For example: various memories, RAM, ROM and the like for storing information by utilizing an electric energy mode; various memories for storing information by utilizing a magnetic energy mode, such as a hard disk, a floppy disk, a magnetic tape, a magnetic core memory, a bubble memory and a U disk; various memories, CDs or DVDs, which store information optically. Of course, there are other ways of storing, such as quantum storing, graphene storing, etc.
The rolling bearing residual life prediction device based on RVM-KF, which is formed by the memory and the processor and is formed by storing the computer program instructions for realizing the rolling bearing residual life prediction method based on RVM-KF, is realized by executing corresponding program instructions by the processor in the computer, and the computer can be realized in an intelligent terminal by using a windows operating system, a linux system or other, for example, using android and iOS system programming languages, and is realized by processing logic based on a quantum computer.
As other embodiments, the rolling bearing residual life prediction device based on RVM-KF may further include other processing hardware, such as a database or a multi-level buffer, a GPU, etc., and the present invention is not limited to the structure of the rolling bearing residual life prediction device based on RVM-KF specifically.
Claims (4)
1. The method for predicting the residual life of the rolling bearing based on RVM-KF is characterized by comprising the following steps of:
1) Acquiring real-time monitoring data of a rolling bearing to be predicted and historical data of the whole life cycle of the rolling bearing similar to the rolling bearing to be predicted; the real-time monitoring data and the historical data refer to vibration data and corresponding time;
2) Training an RVM model using the historical data; the RVM model is input as time and output as vibration data;
3) Determining a prediction starting moment according to the real-time monitoring data and the SKF; and determining a state equation of a Kalman filter according to the real-time monitoring data, wherein the state equation of the Kalman filter is a random effect index model, and the random effect index model is as follows:
Wherein Y (t) is vibration data at time t; alpha is a constant; θ and β are parameters; σ 2 is variance; epsilon (t) is an error term, which is Brownian motion with a mean value of 0 and a variance of sigma 2;
4) Inputting the predicted starting moment into a trained RVM model to obtain estimated vibration data, taking the estimated vibration data as an observation value of a Kalman filter, carrying out single-step prediction to obtain single-step predicted data, finishing prediction of the residual life if the vibration data in the single-step predicted data exceeds a failure threshold value, adding the single-step predicted data into the historical data of the step 2) if the vibration data in the single-step predicted data does not exceed the failure threshold value, retraining the RVM model, and updating the RVM model; the step is iteratively performed until vibration data in the single-step prediction data exceeds a failure threshold.
2. The RVM-KF based rolling bearing remaining life prediction method of claim 1, wherein the vibration data includes at least one characteristic value extracted from the vibration signal.
3. The RVM-KF based rolling bearing remaining life prediction method of claim 1, wherein the historical data utilized in training the RVM model is historical data of an accelerated degradation phase.
4. An RVM-KF based rolling bearing remaining life prediction device, characterized by comprising a processor, a memory and a computer program stored in the memory and executable on the processor, the processor implementing the RVM-KF based rolling bearing remaining life prediction method according to any of claims 1-3 when executing the computer program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010554878.5A CN111783242B (en) | 2020-06-17 | 2020-06-17 | RVM-KF-based rolling bearing residual life prediction method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010554878.5A CN111783242B (en) | 2020-06-17 | 2020-06-17 | RVM-KF-based rolling bearing residual life prediction method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111783242A CN111783242A (en) | 2020-10-16 |
| CN111783242B true CN111783242B (en) | 2024-05-28 |
Family
ID=72757291
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010554878.5A Active CN111783242B (en) | 2020-06-17 | 2020-06-17 | RVM-KF-based rolling bearing residual life prediction method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111783242B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112949204B (en) * | 2021-03-22 | 2022-12-09 | 西安交通大学 | Rolling bearing residual life prediction method with data and model adaptive matching |
| CN113609685B (en) * | 2021-08-09 | 2023-04-07 | 电子科技大学 | Bearing residual life prediction method based on optimized RVM and mixed degradation model |
| CN114662226B (en) * | 2022-02-24 | 2024-03-29 | 北京工业大学 | Rolling bearing residual service life prediction method based on time-varying Kalman filtering |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1705542A1 (en) * | 2005-03-24 | 2006-09-27 | Abb Research Ltd. | Estimating health parameters or symptoms of a degrading system |
| CN104318043A (en) * | 2014-02-20 | 2015-01-28 | 河南科技大学 | Rolling bearing vibration performance reliability variation process detection method and rolling bearing vibration performance reliability variation process detection device |
| CN104615866A (en) * | 2015-01-21 | 2015-05-13 | 北京航空航天大学 | Service life prediction method based on physical statistic model |
| CN105300692A (en) * | 2015-08-07 | 2016-02-03 | 浙江工业大学 | Bearing fault diagnosis and prediction method based on extended Kalman filtering algorithm |
| CN105956236A (en) * | 2016-04-22 | 2016-09-21 | 西安交通大学 | Dual-updating four-factor random degeneration model gear life prediction method |
| KR20170067292A (en) * | 2015-12-08 | 2017-06-16 | 한양대학교 산학협력단 | Device and method for estimating remaining life of mechanical system |
| CN106934125A (en) * | 2017-02-28 | 2017-07-07 | 西安交通大学 | A kind of exponential model plant equipment method for predicting residual useful life of trapezoidal noise profile |
| CN107145720A (en) * | 2017-04-19 | 2017-09-08 | 浙江大学 | It is continuous to degenerate and the unknown equipment method for predicting residual useful life impacted under collective effect |
| CN108846241A (en) * | 2018-07-04 | 2018-11-20 | 西安交通大学 | The rolling bearing life prediction technique to be linked based on Fu Leixie apart from dynamic digital-to-analogue |
| CN109543317A (en) * | 2018-04-28 | 2019-03-29 | 北京航空航天大学 | A kind of method and device of PEMFC remaining life prediction |
| 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 |
| CN110095731A (en) * | 2019-05-06 | 2019-08-06 | 哈尔滨工业大学 | Remaining life DIRECT FORECASTING METHOD applied to long-life space lithium ion battery |
| KR20200049295A (en) * | 2018-10-31 | 2020-05-08 | 한국전력공사 | A method to predict health index transition and residual life for turbomachinery |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11262272B2 (en) * | 2018-10-10 | 2022-03-01 | Palo Alto Research Center Incorporated | Adaptive remaining useful life estimation method using constraint convex regression from degradation measurement |
| TWI669617B (en) * | 2018-10-12 | 2019-08-21 | 財團法人工業技術研究院 | Health monitor method for an equipment and system thereof |
-
2020
- 2020-06-17 CN CN202010554878.5A patent/CN111783242B/en active Active
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1705542A1 (en) * | 2005-03-24 | 2006-09-27 | Abb Research Ltd. | Estimating health parameters or symptoms of a degrading system |
| CN104318043A (en) * | 2014-02-20 | 2015-01-28 | 河南科技大学 | Rolling bearing vibration performance reliability variation process detection method and rolling bearing vibration performance reliability variation process detection device |
| CN104615866A (en) * | 2015-01-21 | 2015-05-13 | 北京航空航天大学 | Service life prediction method based on physical statistic model |
| CN105300692A (en) * | 2015-08-07 | 2016-02-03 | 浙江工业大学 | Bearing fault diagnosis and prediction method based on extended Kalman filtering algorithm |
| KR20170067292A (en) * | 2015-12-08 | 2017-06-16 | 한양대학교 산학협력단 | Device and method for estimating remaining life of mechanical system |
| CN105956236A (en) * | 2016-04-22 | 2016-09-21 | 西安交通大学 | Dual-updating four-factor random degeneration model gear life prediction method |
| CN106934125A (en) * | 2017-02-28 | 2017-07-07 | 西安交通大学 | A kind of exponential model plant equipment method for predicting residual useful life of trapezoidal noise profile |
| CN107145720A (en) * | 2017-04-19 | 2017-09-08 | 浙江大学 | It is continuous to degenerate and the unknown equipment method for predicting residual useful life impacted under collective effect |
| CN109543317A (en) * | 2018-04-28 | 2019-03-29 | 北京航空航天大学 | A kind of method and device of PEMFC remaining life prediction |
| CN108846241A (en) * | 2018-07-04 | 2018-11-20 | 西安交通大学 | The rolling bearing life prediction technique to be linked based on Fu Leixie apart from dynamic digital-to-analogue |
| KR20200049295A (en) * | 2018-10-31 | 2020-05-08 | 한국전력공사 | A method to predict health index transition and residual life for turbomachinery |
| 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 |
| CN110095731A (en) * | 2019-05-06 | 2019-08-06 | 哈尔滨工业大学 | Remaining life DIRECT FORECASTING METHOD applied to long-life space lithium ion battery |
Non-Patent Citations (6)
| Title |
|---|
| New Particle Filter Based on GA for Equipment Remaining Useful Life Prediction;ke li et al.;sensors;20170328;第17卷(第4期);全文 * |
| Satellite lithium-ion battery remaining useful life estimation with an iterative updated RVM fused with the KF algorithm;Yuchen SONG et al.;Chinese Journal of Aeronautics;第31卷(第1期);全文 * |
| 基于UKF的轴承剩余寿命预测方法研究;阙子俊;金晓航;孙毅;;仪器仪表学报(09);全文 * |
| 基于支持向量机和卡尔曼滤波的机械零件剩余寿命预测模型研究;于震梁;孙志礼;曹汝男;王鹏;;兵工学报;20180531(05);全文 * |
| 滚动轴承性能保持可靠性预测;夏新涛等;轴承(第6期);全文 * |
| 航空发动机的健康指标构建与剩余寿命预测;彭开香;皮彦婷;焦瑞华;唐鹏;;控制理论与应用(04);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111783242A (en) | 2020-10-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112629863B (en) | Bearing fault diagnosis method for dynamic joint distribution alignment network under variable working conditions | |
| CN111783242B (en) | RVM-KF-based rolling bearing residual life prediction method and device | |
| CN110378052B (en) | Equipment residual life prediction method considering future working conditions based on cyclic neural network | |
| CN108984893B (en) | Gradient lifting method-based trend prediction method | |
| CN111414703B (en) | Method and device for predicting residual life of rolling bearing | |
| Fan et al. | A novel machine learning method based approach for Li-ion battery prognostic and health management | |
| CN113722985B (en) | Method and system for evaluating health state and predicting residual life of aero-engine | |
| CN111595583B (en) | GAU neural network-based bearing residual life prediction method | |
| CN114841078B (en) | Method and system for predicting residual service life of bearing | |
| Zhao et al. | Bearing health condition prediction using deep belief network | |
| CN106407998A (en) | Probability time-varying seawater hydraulic pump fault prediction method | |
| CN108961460B (en) | Fault prediction method and device based on sparse ESGP (Enterprise service gateway) and multi-objective optimization | |
| CN108804334A (en) | A kind of discrete type software reliability growth test and evaluation method based on adaptive sampling | |
| CN114048688A (en) | Method for predicting service life of bearing of wind power generator | |
| CN114740388A (en) | Lithium battery residual life state evaluation method based on improved TCN | |
| CN110610226A (en) | Generator fault prediction method and device | |
| CN111400964B (en) | Fault occurrence time prediction method and device | |
| CN111523727B (en) | Method for predicting remaining life of battery by considering recovery effect based on uncertain process | |
| CN114166509A (en) | Motor bearing fault prediction method | |
| CN117406100A (en) | Lithium ion battery remaining life prediction method and system | |
| CN112379274A (en) | Method for predicting residual life of power battery | |
| CN117521512A (en) | Bearing residual service life prediction method based on multi-scale Bayesian convolution transducer model | |
| CN114861349A (en) | Rolling bearing RUL prediction method based on model migration and wiener process | |
| CN117371331A (en) | KDIN-based method for predicting residual life of aeroengine | |
| CN115898927A (en) | Fusion device molecular pump fault diagnosis and prediction method based on machine learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |