CN113468801A - Method for predicting residual life of gear by estimating nuclear density - Google Patents
Method for predicting residual life of gear by estimating nuclear density Download PDFInfo
- Publication number
- CN113468801A CN113468801A CN202110631246.9A CN202110631246A CN113468801A CN 113468801 A CN113468801 A CN 113468801A CN 202110631246 A CN202110631246 A CN 202110631246A CN 113468801 A CN113468801 A CN 113468801A
- Authority
- CN
- China
- Prior art keywords
- degradation
- sample
- time
- density
- gear
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000006731 degradation reaction Methods 0.000 claims abstract description 146
- 230000015556 catabolic process Effects 0.000 claims abstract description 145
- 238000012360 testing method Methods 0.000 claims abstract description 69
- 230000007704 transition Effects 0.000 claims abstract description 31
- 238000012546 transfer Methods 0.000 claims abstract description 30
- 230000001133 acceleration Effects 0.000 claims abstract description 23
- 238000012544 monitoring process Methods 0.000 claims abstract description 11
- 230000005540 biological transmission Effects 0.000 claims abstract description 5
- 238000005299 abrasion Methods 0.000 claims abstract description 3
- 238000009499 grossing Methods 0.000 claims description 30
- 238000005070 sampling Methods 0.000 claims description 27
- 238000004364 calculation method Methods 0.000 claims description 18
- 239000011159 matrix material Substances 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 8
- 230000001186 cumulative effect Effects 0.000 claims description 7
- 238000005315 distribution function Methods 0.000 claims description 6
- 238000004458 analytical method Methods 0.000 claims description 3
- 238000011160 research Methods 0.000 claims description 3
- 238000012423 maintenance Methods 0.000 description 5
- 230000003044 adaptive effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000004026 adhesive bonding Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000010998 test method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/08—Probabilistic or stochastic CAD
-
- 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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Probability & Statistics with Applications (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
A method for predicting the residual life of a gear by estimating the nuclear density belongs to the technical field of mechanical reliability and comprises the following specific implementation steps: 1. monitoring the degradation of an internal gear of the main test gearbox in real time by using an acceleration sensor; 2. extracting the characteristics of the degradation state of the gear, and evaluating the degradation performance of the gear in terms of abrasion; 3. clustering data representing different degradation states in a degradation stage by using a time sequence density peak value clustering algorithm, and dividing different degradation state modes according to a real-time clustering result to realize dynamic transfer of the system degradation state; 4. carrying out nonparametric estimation on the probability density function of the continuous degradation state of the gear by using a kernel density estimation method to obtain the probability density function of the degradation state of the gear; 5. and predicting the residual life of the gear by considering the residual life prediction model of the degradation state transition to obtain the residual life distribution of the gear. The advantages are that: the service life of the gear in gear transmission can be predicted, and the gear failure can be prevented, so that economic loss is caused.
Description
Technical Field
The invention belongs to the technical field of mechanical reliability, in particular to a method for predicting the residual life of gear nuclear density estimation,
background
The gear is a key component in a mechanical equipment transmission system widely applied in the mechanical industry, when the gear has faults of tooth breakage, tooth surface fatigue, gluing and the like, the catastrophic damage of the whole mechanical equipment is often caused, therefore, a reasonable and effective maintenance scheme provided for the gear becomes an urgent problem to be solved, in the whole maintenance scheme making process, the residual life prediction of the gear is a key and difficult point, with the development of the information sensing technology, the running state of the gear can be monitored in real time, the degradation state and the residual life of the system can be more accurately predicted by using a large amount of received real-time monitoring information, the key information related to the mechanical health state is provided, the occurrence of faults can be identified and managed, the maintenance activities can be planned, and a basis is provided for more reasonably making a maintenance strategy based on the state,
disclosure of Invention
The invention aims to provide a method for predicting the residual life of gear nuclear density estimation, which provides a scientific basis for maintenance and repair, and is implemented as follows, wherein the method is characterized by comprising the following implementation steps of: the implementation steps are shown in figure 1:
the test bench shown in figure 2 is adopted and is formed by connecting a main test gear box 1 and an auxiliary test gear box 2, the center distance a is 150cm, the test adopts mechanical lever loading, the torque is measured by a torque and rotating speed sensor, the vibration, the acceleration, the temperature and the noise of the main test box body and the auxiliary test box body are monitored in the test process, a pair of gears which are staggered and lapped in the front and the back are arranged in the main test gear box body, the broken tooth state of the gears is equivalent to the failure of the gears,
13 sensors are arranged on the test bench, 1,2, 3 and 4 acceleration sensors 1#, 2#, 3# and 4# are respectively arranged at the radial position of the bearing seat in the main test gear box 1, 7 and 8 acceleration sensors 7# and 8# are respectively arranged at the axial position of the bearing seat in the main test gear box 1, and 5 and 6 acceleration sensors 5# and 6# are respectively arranged at the radial position of the bearing seat in the auxiliary test gear box 2; 9. the No. 10 sound sensors 9# and 10# are respectively hung at the positions 40cm above the main test gear box 1 and the auxiliary test gear box 2; no. 11 temperature sensor 11# is arranged inside the main test gear box; a number 12 rotating speed sensor 12# is arranged at the output end of the driving motor 8; the No. 13 torque sensor 13# is arranged at the coupling shafts of the main and the auxiliary gear boxes 1 and 2,
in the test, eight-stage loads are loaded, wherein the eight-stage loads respectively have 349.5 torque, 430.7 torque, 492.2 torque, 555.6 torque, 612.9 torque, 693.4 torque, 734 torque and 822.7 torque, the running time of each stage of load is 10 hours, tooth breakage occurs during the eighth stage of load, the acceleration data of the gear is recorded by a No. 4 sensor 4#, the residual life prediction selects the acceleration test whole time domain signal from the eighth stage of load to tooth breakage for analysis, and the sampling information is as follows: the sampling frequency is 25.6kHz, each sampling lasts for 60 seconds, and the text sampling is recorded every 9 minutes;
in the formula: Σ is a summation number, n is the number of sampling points per sampling period, yiIs tiMean square amplitude, y, of the state information of the time gearjData for each sampling period of the gear;
3, clustering data representing different degradation states at a degradation stage by utilizing a proposed time sequence density peak value clustering algorithm suitable for real-time manifold data clustering according to initial data of the gear in the main test gear box, and dividing different degradation state modes according to a real-time clustering result to realize dynamic transfer of the system degradation state;
according to sample xpK neighbor information of (1), defining sample xpPart ofDensity ppAdaptive truncation distance h in (1)pThe cluster center of the sample is searched, the real distribution information of the data set sample can be reflected better, and the large truncation distance of the place with large sample density is selected, so that the local density of the sample becomes higher, and the cluster center is found accurately; where the density of the sample is small, hpThe value is small, the local density becomes smaller, and outliers are easy to find;
sample xpLocal density p ofpComprises the following steps:
sample xpAdaptive truncation distance h ofpComprises the following steps:
hp=dmax+dmin-dknn(xp) (3)
dmaxand dminIs a set { dknn(xp) 1,2, m } maximum and minimum values, dknn(xp) Represents the average distance to its k nearest neighbors as shown in equation (4):
wherein, KNN (x)p)={d1(xp),d2(xp),...dk(xp) Is the calculation point xpK-nearest neighbor method of Euclidean distance of each adjacent sample, which is to sample point xpDefining:
d1(xp)≤d2(xp)≤…≤dk(xp) (5)
equation (5) is from sample xpDistance to neighbor in increasing order: d1(xp) Is xpEuclidean distance to nearest neighbor, d2(xp) Is xpEuclidean distance to next neighbor, and so on, k is generally taken asn is the total number of samples;
from the formulas (3) and (4), h is shownpIn the area with high density, the local density is increased by using a large truncation distance; in the area with small density, the local density is reduced by using a small truncation distance, so that the center of the cluster is easier to select;
the transfer distance is the distance between samples determined by the maximum gap in the connecting path between samples, and at least one path exists between every two points in the data set, so that for any two sample points xp,xqDefine a connection path X ═ X with M verticesu1xu2...xuMWherein x isu1=xp,xuM=xqAt this time for sample xpAnd sample xqThe distance between is then defined as:
any two sample points xp,xqSelecting the maximum gap in each connection path X as the path Xp,xqAt a distance between the connection pathsSelect each path xp,xqDistance minimum as transfer distance:
in the formulaIs the transfer distance of the two points,is an arbitrary sample xpAnd sample xqAll connection paths between;
constructing a new transfer matrix to distribute the residual samples, wherein the calculation method of the transfer matrix T comprises the following steps:
1) forming an undirected graph G (X, E) with Euclidean distance as a weight value for all samples of the data, and obtaining the directed graph by adopting a Kruskal minimum spanning tree algorithm
2) To be provided withMedium maximum transmission distanceCutting apartCan obtain the productAndmiddle arbitrary sample point xpToMiddle arbitrary sample point xqIs defined as the length of the cut edgeT is a transfer matrix;
3) repeating the step 2), and cuttingAnduntil there is no subtree, then obtaining the transfer matrix T between all sample points;
4) allocating sample points by a transfer matrix;
the kernel density estimation method is a nonparametric estimation method which does not make any assumption on the form of data distribution and researches the data distribution characteristics from the data;
let Δ x1,Δx2,...,ΔxnIs an independent and identically distributed random variable for representing the degradation increment of the system per unit time, and the probability density function is f (delta x), then the functionThe nuclear density estimate of f (Δ x) is written as:
where h is a smoothing parameter, K (-) is a kernel function, n is the number of samples of a random variable, any function can be used as a kernel function in principle when the samples are large enough, and a Gaussian kernel function is generally selected in consideration of the shape influence and smooth weight of the functionThe smoothing parameter is an important parameter for smoothing the whole estimation in kernel density estimation, and the width of each kernel function can be determined so as to determine the kernel densityThe accuracy of the degree estimate;
the degenerated increments are independently and identically distributed in unit time, then tkThe distribution of the time characteristic degradation amount is [0, t ]k]Distribution of cumulative sums of characteristic degradation increments per unit time, i.e. tkProbability density function of accumulated time characteristic degradation X (t)Is equal to the probability density function of the degradation increment in unit timeK-deconvolution of (a), i.e.:
then go through tlTime of day, future tk+tlThe characteristic cumulative degradation amount at the time is:
X(tk+tl)=X(tk)+X(tl) (11)
then tk+tlThe probability density function of the accumulated time characteristic degradation amount is as follows:
defining a component first time to failure threshold time TfWith the current time tkThe difference is the remaining life tRULAccording to the nuclear density estimation model of the degradation distribution, a fault time distribution function F (t) is obtained as follows:
wherein L is a fault threshold, a remaining life distribution function F can be obtainedk(t) is:
probability density function f of real-time residual life can be predictedk(t):
Through the modeling process, the probability density function of the residual life can be predicted in a nuclear density estimation mode after new degradation characteristic incremental data are obtained
After clustering, dividing different degradation stages, and setting a smoothing parameter h in formula (9) as degradation increment delta X of each stagenThe function of the surrounding density, the smoothing parameter of the kernel function is adjusted by the distance between the sample point and the nearest neighbor, and is recorded asTaking the average Euclidean distance of m nearest neighbors selected in a new degradation stage after sample clustering as the window width of kernel estimation, namely:
wherein d isi(delta x) is the distance from the retrogression incremental sample delta x to the nearest neighbor of the retrogression degraded state of clustering, and in a new degradation stage after clustering, the smoothing parameter adapts to the local density to carry out kernel density estimation, so that the smoothing parameter at the high density part is smaller, the smoothing parameter at the low density part is larger, and smoother estimation is obtained, and therefore, the kernel density estimation is adopted for establishing the self-adapting smoothing parameter degradation incremental probability density function for the data of different degradation states, and the probability density function is as follows:
the dynamic transition of the degradation state is realized after the real-time received degradation information is clustered, the degradation information of the last stage selected by the update sample improves the self-adaptive update efficiency of the smooth parameter, and the accuracy of the prediction of the residual life of the system is improved;
new degradation stage after sample clustering, current tkProbability density function with time instant updated to m state degradation incrementsCan be expressed as:
it can be known that the probability density function of the degradation state at the current moment after the sample clustering can be expressed as:
it can be known from the formula (18) and the formula (19) that after the system degradation state transition real-time clustering is considered, the kernel density estimation model needs to perform convolution calculation again under different degradation modes, the calculation amount is larger and larger, and therefore the t is passedk-1Estimating the nuclear density of the degradation state at the moment to obtain tkThe kernel density estimation of the dynamic transition of the time degradation state realizes the real-time update of the probability density function of the residual service life, and is very necessary to reduce the calculation complexity;
substituting equation (20) into equation (14) can obtain the remaining life distribution and predict the probability density function of the real-time remaining life considering the transition of the degradation stateComprises the following steps:
when the m-th sample is received at the current moment, the self-adaptive smoothing parameter of the m-th sample is updatedIn thatCan obtain the real-time residual life distribution at the current moment on the basisTherefore, after new degradation characteristic increment data are continuously obtained, the predicted residual life probability density function is updated in real time, and the calculation complexity is reduced;
and 5, predicting the residual life of the gear by considering the residual life prediction model of the degraded state transition:
the invention has the advantages and positive effects that:
the invention is applied to the residual life prediction of the gear, establishes a prediction model, can effectively predict the residual life of the gear,
drawings
FIG. 1 is a flowchart of a method for predicting remaining life of a gear according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a test rig in an embodiment of the present invention;
FIG. 3 is a graph of eigenvalue change with monitoring time in an embodiment of the present invention;
FIG. 4 is a 20-hour regression state T-CFDP clustering in an example of the present invention;
FIG. 5 shows the T-CFDP clustering for 30 hours of degradation in an embodiment of the present invention;
FIG. 6 shows the T-CFDP clustering for 40 hours of degradation in an embodiment of the present invention;
FIG. 7 shows the T-CFDP clustering for 50 hours of degradation in an embodiment of the present invention;
FIG. 8 is a T-CFDP clustering of 60 hours of degradation in accordance with an embodiment of the present invention;
FIG. 9 shows T-CFDP clustering for 70 hours of degradation in an embodiment of the present invention;
FIG. 10 is a graph illustrating predicted remaining life at various monitoring times in accordance with an embodiment of the present invention;
FIG. 11 is a probability density distribution of 30h remaining life in an embodiment of the present invention;
FIG. 12 is a probability density distribution of 60h remaining life in an embodiment of the present invention;
in the figure: 1-a main test gear box, 2-an auxiliary test gear box, 3-a first coupler, 4-a second coupler, 5-a third coupler, 6-a mechanical lever, 7-a detection system, 8-a driving motor, and a-the center distance between the main test gear box and the auxiliary test gear box;
1# -8 # -1 st to 8 th acceleration sensors, 9#, 10# -1 st and 2 nd sound sensors, 11# -temperature sensor, 12# -rotating speed sensor and 13# -torque sensor;
Detailed Description
The embodiments of the invention will be further explained with reference to the drawings in which:
fig. 2 is a schematic diagram of a test bench for the test application, and the test method comprises the following steps:
the test bench shown in figure 2 is adopted and is formed by connecting a main test gear box 1 and an auxiliary test gear box 2, the center distance a is 150cm, the test adopts mechanical lever loading, the torque is measured by a torque and rotating speed sensor, the vibration, the acceleration, the temperature and the noise of the main test box body and the auxiliary test box body are monitored in the test process, a pair of gears which are staggered and lapped in the front and the back are arranged in the main test gear box body, the broken tooth state of the gears is equivalent to the failure of the gears,
13 sensors are arranged on the test bench, 1,2, 3 and 4 acceleration sensors 1#, 2#, 3# and 4# are respectively arranged at the radial position of the bearing seat in the main test gear box 1, 7 and 8 acceleration sensors 7# and 8# are respectively arranged at the axial position of the bearing seat in the main test gear box 1, and 5 and 6 acceleration sensors 5# and 6# are respectively arranged at the radial position of the bearing seat in the auxiliary test gear box 2; 9. the No. 10 sound sensors 9# and 10# are respectively hung at the positions 40cm above the main test gear box 1 and the auxiliary test gear box 2; no. 11 temperature sensor 11# is arranged inside the main test gear box; a number 12 rotating speed sensor 12# is arranged at the output end of the driving motor 8; the No. 13 torque sensor 13# is arranged at the coupling shafts of the main and the auxiliary gear boxes 1 and 2,
in the test, eight-stage loads are loaded, wherein the eight-stage loads respectively have 349.5 torque, 430.7 torque, 492.2 torque, 555.6 torque, 612.9 torque, 693.4 torque, 734 torque and 822.7 torque, the running time of each stage of load is 10 hours, tooth breakage occurs during the eighth stage of load, the acceleration data of the gear is recorded by a No. 4 sensor 4#, the residual life prediction selects the acceleration test whole time domain signal from the eighth stage of load to tooth breakage for analysis, and the sampling information is as follows: the sampling frequency is 25.6kHz, each sampling lasts for 60 seconds, and the text sampling is recorded every 9 minutes;
in the formula: Σ is a summation number, n is the number of sampling points per sampling period, yiIs tiStatus information of the time gear, yjData for each sampling period of the gear;
processing the acceleration and noise data by a mean square amplitude method to obtain a graph of the degradation of the gearbox along with the change of monitoring time as shown in figure 3, wherein the threshold value y of the gearbox when a fault occurs is 76.325mm/s2,
And 3, clustering data representing different degradation states in a degradation stage by using the proposed time sequence density peak value clustering algorithm suitable for real-time manifold data clustering, and dividing different degradation state modes according to real-time clustering results to realize dynamic transfer of system degradation states:
according to sample xpK neighbor information of (1), defining sample xpLocal density p ofpAdaptive truncation distance h in (1)pThe cluster center of the sample is searched, the real distribution information of the data set sample can be reflected better, and the large truncation distance of the place with large sample density is selected, so that the local density of the sample becomes higher, and the cluster center is found accurately; where the density of the sample is small, hpThe value is small, the local density becomes smaller, and outliers are easy to find;
sample xpLocal density p ofpComprises the following steps:
sample xpAdaptive truncation distance h ofpComprises the following steps:
hp=dmax+dmin-dknn(xp) (3)
dmaxand dminIs a set { dknn(xp) 1,2, m } maximum and minimum values, dknn(xp) Represents the average distance to its k nearest neighbors as shown in equation (4):
wherein, KNN (x)p)={d1(xp),d2(xp),...dk(xp) Is the calculation point xpK-nearest neighbor method of Euclidean distance of each adjacent sample, which is to sample point xpDefining:
d1(xp)≤d2(xp)≤…≤dk(xp) (5)
equation (5) is from sample xpDistance to neighbor in increasing order: d1(xp) Is xpEuclidean distance to nearest neighbor, d2(xp) Is xpEuclidean distance to next neighbor, and so on, k is generally taken asn is the total number of samples;
from the formulas (3) and (4), h is shownpIn the area with high density, the local density is increased by using a large truncation distance; in the area with small density, the local density is reduced by using a small truncation distance, so that the center of the cluster is easier to select;
the transfer distance is the distance between samples determined by the maximum gap in the connecting path between samples, and at least one path exists between every two points in the data set, so that for any two sample points xp,xqDefine a connection path X ═ X with M verticesu1xu2…xuMWherein x isu1=xp,xuM=xqAt this time for sample xpAnd sample xqThe distance between is then defined as:
any two sample points xp,xqSelecting the maximum gap in each connection path X as the path Xp,xqAt a distance between the connection pathsSelect each path xp,xqDistance minimum as transfer distance:
in the formulaIs the transfer distance of the two points,is an arbitrary sample xpAnd sample xqAll connection paths between;
constructing a new transfer matrix to distribute the residual samples, wherein the calculation method of the transfer matrix T comprises the following steps:
1) forming an undirected graph G (X, E) with Euclidean distance as a weight value for all samples of the data, and obtaining the directed graph by adopting a Kruskal minimum spanning tree algorithm
2) To be provided withMedium maximum transmission distanceCutting apartCan obtain the productAndall kinds ofThis point xpToMiddle arbitrary sample point xqIs defined as the length of the cut edgeT is a transfer matrix;
3) repeating the step 2), and cuttingAnduntil there is no subtree, then obtaining the transfer matrix T between all sample points;
4) allocating sample points by a transfer matrix;
finally, obtaining a degradation state cluster map of the system shown in the figures 4-9 at different times;
the kernel density estimation method is a nonparametric estimation method which does not make any assumption on the form of data distribution and researches the data distribution characteristics from the data;
let Δ x1,Δx2,...,ΔxnIs an independent and identically distributed random variable for representing the degradation increment of the system per unit time, and the probability density function is f (delta x), then the functionThe nuclear density estimate of f (Δ x) is written as:
wherein h is a smoothing parameter and K (-) isKernel function, n is the number of samples of random variables, when the samples are large enough, in principle any function can be used as kernel function, and the Gaussian kernel function is generally selected in consideration of the shape influence and smooth weightThe smoothing parameter is an important parameter for smoothing the whole estimation in the kernel density estimation, and the width of each kernel function can be determined so as to determine the accuracy of the kernel density estimation;
the degenerated increments are independently and identically distributed in unit time, then tkThe distribution of the time characteristic degradation amount is [0, t ]k]Distribution of cumulative sums of characteristic degradation increments per unit time, i.e. tkProbability density function of accumulated time characteristic degradation X (t)Is equal to the probability density function of the degradation increment in unit timeK deconvolution of (i), i.e.
Then go through tlTime of day, future tk+tlThe characteristic cumulative degradation amount at the time is:
X(tk+tl)=X(tk)+X(tl) (11)
then tk+tlThe probability density function of the accumulated time characteristic degradation amount is as follows:
defining a component first time to failure threshold time TfWith the current time tkThe difference is the remaining life tRULThe fault can be obtained according to the nuclear density estimation model of the degradation distributionThe time distribution function F (t) is:
wherein L is a fault threshold, a remaining life distribution function F can be obtainedk(t) is:
probability density function f of real-time residual life can be predictedk(t):
Through the modeling process, the probability density function of the residual life can be predicted in a nuclear density estimation mode after new degradation characteristic incremental data are obtained
After clustering, dividing different degradation stages, and setting a smoothing parameter h in formula (9) as degradation increment delta X of each stagenThe function of the surrounding density, the smoothing parameter of the kernel function is adjusted by the distance between the sample point and the nearest neighbor, and is recorded as Taking the average Euclidean distance of m nearest neighbors selected in a new degradation stage after sample clustering as the window width of kernel estimation, namely:
wherein d isi(Δ x) is the cluster degradation stateAnd transferring a retrogression incremental sample delta x to the nearest neighbor distance of the retrogression incremental sample delta x, and in a new degradation stage after clustering, adapting the smoothing parameters to local density to carry out kernel density estimation, so that the local smoothing parameters with high density are smaller, the local smoothing parameters with low density are larger, and smoother estimation is obtained, and thus, adopting kernel density estimation to establish a self-adaptive smoothing parameter degradation incremental probability density function for data in different degradation states:
the dynamic transition of the degradation state is realized after the real-time received degradation information is clustered, the degradation information of the last stage selected by the update sample improves the self-adaptive update efficiency of the smooth parameter, and the accuracy of the prediction of the residual life of the system is improved;
new degradation stage after sample clustering, current tkProbability density function with time instant updated to m state degradation incrementsCan be expressed as:
it can be known that the probability density function of the degradation state at the current moment after the sample clustering can be expressed as:
it can be known from the formula (18) and the formula (19) that after the system degradation state transition real-time clustering is considered, the kernel density estimation model needs to perform convolution calculation again under different degradation modes, the calculation amount is larger and larger, and therefore the t is passedk-1Estimating the nuclear density of the degradation state at the moment to obtain tkThe kernel density estimation of the dynamic transition of the time degradation state realizes the real-time update of the probability density function of the residual service lifeAnd it is very necessary to reduce the computational complexity;
substituting equation (20) into equation (14) can obtain the remaining life distribution and predict the probability density function of the real-time remaining life considering the transition of the degradation stateComprises the following steps:
when the m-th sample is received at the current moment, the self-adaptive smoothing parameter of the m-th sample is updatedIn thatCan obtain the real-time residual life distribution at the current moment on the basisTherefore, after new degradation characteristic increment data are continuously obtained, the predicted residual life probability density function is updated in real time, and the calculation complexity is reduced;
and 5, predicting the residual life of the gear by considering the residual life prediction model of the degraded state transition:
the predicted value result of the residual life of the dynamic transfer gear considering the degradation state shown in fig. 10 is finally obtained:
the comparison error between the predicted value and the actual value of the residual life of the gear for the dynamic transition considering the degradation state and the dynamic transition not considering the degradation state under the same parameter is shown in table 1:
TABLE 1 dynamic transition with degradation state considered and dynamic transition model with degradation state not considered error comparison of predicted values to true values
Comparing the data in the table 1, it can be found that as the operation time of the system increases, the state monitoring information increases, the absolute error between the predicted value and the actual value of the residual life is gradually reduced, meanwhile, the prediction of the residual life considering the dynamic transition of the degradation state is closer to the true value, which shows that the prediction value of the real-time residual life is more accurate by considering the adaptive kernel density estimation of the dynamic transition of the degradation state, which shows that the method provided by the invention can well predict the residual life in real time,
in order to further verify the effectiveness of the method provided by the invention, residual life of the gearbox is predicted by adopting a residual life prediction method based on Gamma distribution on the residual life of the same monitoring point under the same initial sample, and the two predicted values are compared to obtain the results of the graphs in FIGS. 11 and 12;
from fig. 11 and fig. 12, it can be seen that the variance of the residual life probability density function predicted by the kernel density estimation method is smaller than that of the residual life probability density function predicted based on the Gamma distribution model, and the predicted value of the residual life is closer to the actual value, which indicates that the residual life prediction predicted based on the kernel density estimation method is more accurate than that based on the Gamma distribution model, and meanwhile, as the monitoring time increases, the performance of the gearbox deteriorates continuously, the data representing the state degradation of the gearbox increases continuously, so that the variance of the residual life probability density function becomes smaller gradually, the predicted value of the residual life is closer to the actual value gradually, which indicates that the predicted residual life value is more accurate;
fig. 11 shows the probability density distribution of the remaining lifetime of 30h, and it can be seen that the variance of the probability density distribution of the dynamic transition without considering the degradation state and the probability density distribution of the kernel density estimation with considering the degradation state are approximately the same, but the kernel density estimation with considering the degradation state is closer to the true value, fig. 12 shows the probability density distribution of the remaining lifetime of 60h, the absolute error of the kernel density estimation result of the dynamic transition without considering the degradation state is large, and the variance is small because all data at the current time and before are selected, the possible mutation and the possible degradation acceleration in the degradation process are not considered, the error of the predicted lifetime obtained by considering the kernel density estimation method of the dynamic transition with the degradation state is small, but the variance is large, because the dynamic transition of the degradation state is considered in clustering, the degradation data after the state transition at the new degradation acceleration stage is selected, and the predicted average remaining lifetime is more accurate, is closer to the true value of the image,
in conclusion, the invention provides a method for predicting the residual life of gear nuclear density estimation, which clusters data representing different degradation states in a degradation stage by using a time sequence density peak value clustering algorithm, divides different degradation state modes according to a real-time clustering result and realizes dynamic transfer of a system degradation state; carrying out nonparametric estimation on the probability density function of the continuous degradation state of the gear by using a kernel density estimation method to obtain the probability density function of the degradation state of the gear; and finally, adaptively updating smooth parameters according to the sample density, and simultaneously considering the problem of high calculation complexity in the real-time prediction process, establishing an adaptive kernel density estimation real-time residual life prediction recurrence model of the dynamic transition of the degradation state to obtain the residual life distribution of the gear.
Claims (1)
1. A method for predicting the residual life of gear nuclear density estimation is characterized by comprising the following implementation steps:
step 1, acquiring real-time monitoring data representing the state of an internal gear of a main test gear box through a test bench:
the adopted test bench is formed by connecting a main test gear box (1) and an auxiliary test gear box (2), the center distance a is 150cm, the test adopts mechanical lever loading, the torque is measured by a torque and rotation speed sensor, the vibration, acceleration, temperature and noise of the main test box body and the auxiliary test box body are monitored in the test process, a pair of gears which are staggered and lapped in the front and the back are arranged in the main test gear box, and the broken tooth state of the gears is equivalent to the failure of the gears;
13 sensors are arranged on the test bench, 1,2, 3 and 4 acceleration sensors (1#, 2#, 3#, and 4#) are respectively arranged at the radial direction of the inner bearing seat of the main test gear box (1), 7 and 8 acceleration sensors (7#, 8#) are respectively arranged at the axial direction of the inner bearing seat of the main test gear box (1), and 5 and 6 acceleration sensors (5#, 6#) are respectively arranged at the radial direction of the inner bearing seat of the auxiliary test gear box (2); 9. the No. 10 sound sensors (9#, 10#) are respectively hung at the positions 40cm above the main test gear box (1) and the auxiliary test gear box (2); the No. 11 temperature sensor (11#) is arranged inside the main test gear box; a No. 12 rotating speed sensor (12#) is arranged at the output end of the driving motor; the No. 13 torque sensor (13#) is arranged at the connecting shafts of the main and auxiliary gear boxes (1, 2);
in the test, eight-stage loads are loaded, wherein the eight-stage loads respectively have 349.5 torque, 430.7 torque, 492.2 torque, 555.6 torque, 612.9 torque, 693.4 torque, 734 torque and 822.7 torque, the running time of each stage of load is 10 hours, tooth breakage occurs during the eighth stage of load, a No. 4 sensor (4#) is used for recording the acceleration data of the gear, the residual life prediction selects the acceleration from the eighth stage of loading to the tooth breakage to test the whole time domain signal for analysis, and the sampling information is as follows: the sampling frequency is 25.6kHz, each sampling lasts for 60 seconds, and the text sampling is recorded every 9 minutes;
step 2, extracting the characteristic of the degradation state of the gear in the main test gear box, evaluating the degradation performance of the gear abrasion by using mean square amplitude, and expressing the mean square amplitude characteristic value of a sampling signal in each sampling time length as follows:
in the formula: sigma being a sum number and n being per sampling periodNumber of sampling points, yiIs tiMean square amplitude, y, of the state information of the time gearjData for each sampling period of the gear;
3, clustering data representing different degradation states at a degradation stage by utilizing a proposed time sequence density peak value clustering algorithm suitable for real-time manifold data clustering according to initial data of the gear in the main test gear box, and dividing different degradation state modes according to a real-time clustering result to realize dynamic transfer of the system degradation state;
according to sample xpK neighbor information of (1), defining sample xpLocal density p ofpAdaptive truncation distance h in (1)pThe cluster center of the sample is searched, the real distribution information of the data set sample can be reflected better, and the large truncation distance of the place with large sample density is selected, so that the local density of the sample becomes higher, and the cluster center is found accurately; where the density of the sample is small, hpThe value is small, the local density becomes smaller, and outliers are easy to find;
sample xpLocal density p ofpComprises the following steps:
sample xpAdaptive truncation distance h ofpComprises the following steps:
hp=dmax+dmin-dknn(xp) (3)
dmaxand dminIs a set { dknn(xp) 1,2, m } maximum and minimum values, dknn(xp) Represents the average distance to its k nearest neighbors as shown in equation (4):
wherein, KNN (x)p)={d1(xp),d2(xp),...dk(xp) Is the calculation point xpK-nearest neighbor method of Euclidean distance of each adjacent sample, which is to sample point xpDefining:
d1(xp)≤d2(xp)≤…≤dk(xp) (5)
equation (5) is from sample xpDistance to neighbor in increasing order: d1(xp) Is xpEuclidean distance to nearest neighbor, d2(xp) Is xpEuclidean distance to next neighbor, and so on, k is generally taken asn is the total number of samples;
from the formulas (3) and (4), h is shownpIn the area with high density, the local density is increased by using a large truncation distance; in the area with small density, the local density is reduced by using a small truncation distance, so that the center of the cluster is easier to select;
the transfer distance is the distance between samples determined by the maximum gap in the connecting path between samples, and at least one path exists between every two points in the data set, so that for any two sample points xp,xqDefine a connection path X ═ X with M verticesu1xu2…xuMWherein x isu1=xp,xuM=xqAt this time for sample xpAnd sample xqThe distance between is then defined as:
any twoA sample point xp,xqSelecting the maximum gap in each connection path X as the path Xp,xqAt a distance between the connection pathsSelect each path xp,xqDistance minimum as transfer distance:
in the formulaIs the transfer distance of the two points,is an arbitrary sample xpAnd sample xqAll connection paths between;
constructing a new transfer matrix to distribute the residual samples, wherein the calculation method of the transfer matrix T comprises the following steps:
1) forming an undirected graph G (X, E) with Euclidean distance as a weight value for all samples of the data, and obtaining the directed graph by adopting a Kruskal minimum spanning tree algorithm
2) To be provided withMedium maximum transmission distanceCutting apartCan obtain the productAndmiddle arbitrary sample point xpToMiddle arbitrary sample point xqIs defined as the length of the cut edgeT is a transfer matrix;
3) repeating the step 2), and cuttingAnduntil there is no subtree, then obtaining the transfer matrix T between all sample points;
4) allocating sample points by a transfer matrix;
step 4, the density of the clustered samples is changed, and the smooth parameters of kernel density estimation can be judged according to the local density of the samples, so that the smooth parameters are updated in a self-adaptive mode according to the clustering result to realize the residual life prediction of the degradation state transition;
the kernel density estimation method is a nonparametric estimation method which does not make any assumption on the form of data distribution and researches the data distribution characteristics from the data;
let Δ x1,Δx2,...,ΔxnIs an independent and identically distributed random variable for representing the degradation increment of the system per unit time, and the probability density function is f (delta x), then the functionThe nuclear density estimate of f (Δ x) is written as:
where h is a smoothing parameter, K (-) is a kernel function, n is the number of samples of a random variable, any function can be used as a kernel function in principle when the samples are large enough, and a Gaussian kernel function is generally selected in consideration of the shape influence and smooth weight of the functionThe smoothing parameter is an important parameter for smoothing the whole estimation in the kernel density estimation, and the width of each kernel function can be determined so as to determine the accuracy of the kernel density estimation;
the degenerated increments are independently and identically distributed in unit time, then tkThe distribution of the time characteristic degradation amount is [0, t ]k]Distribution of cumulative sums of characteristic degradation increments per unit time, i.e. tkProbability density function of accumulated time characteristic degradation X (t)Is equal to the probability density function of the degradation increment in unit timeK-deconvolution of (a), i.e.:
then go through tlTime of day, future tk+tlThe characteristic cumulative degradation amount at the time is:
X(tk+tl)=X(tk)+X(tl) (11)
then tk+tlTime of day characteristicsThe probability density function for the cumulative amount of degradation is:
defining a component first time to failure threshold time TfWith the current time tkThe difference is the remaining life tRULAccording to the nuclear density estimation model of the degradation distribution, a fault time distribution function F (t) is obtained as follows:
wherein L is a fault threshold, a remaining life distribution function F can be obtainedk(t) is:
probability density function f of real-time residual life can be predictedk(t):
Through the modeling process, the probability density function of the residual life can be predicted in a nuclear density estimation mode after new degradation characteristic incremental data are obtained
After clustering, dividing different degradation stages, and setting a smoothing parameter h in formula (9) as degradation increment delta X of each stagenThe function of the surrounding density, the smoothing parameter of the kernel function is adjusted by the distance between the sample point and the nearest neighbor, and is recorded as Taking the average Euclidean distance of m nearest neighbors selected in a new degradation stage after sample clustering as the window width of kernel estimation, namely:
wherein d isi(delta x) is the distance from the retrogression incremental sample delta x to the nearest neighbor of the retrogression degraded state of clustering, and in a new degradation stage after clustering, the smoothing parameter adapts to the local density to carry out kernel density estimation, so that the smoothing parameter at the high density part is smaller, the smoothing parameter at the low density part is larger, and smoother estimation is obtained, and therefore, the kernel density estimation is adopted for establishing the self-adapting smoothing parameter degradation incremental probability density function for the data of different degradation states, and the probability density function is as follows:
the dynamic transition of the degradation state is realized after the real-time received degradation information is clustered, the degradation information of the last stage selected by the update sample improves the self-adaptive update efficiency of the smooth parameter, and the accuracy of the prediction of the residual life of the system is improved;
new degradation stage after sample clustering, current tkProbability density function with time instant updated to m state degradation incrementsCan be expressed as:
it can be known that the probability density function of the degradation state at the current moment after the sample clustering can be expressed as:
it can be known from the formula (18) and the formula (19) that after the system degradation state transition real-time clustering is considered, the kernel density estimation model needs to perform convolution calculation again under different degradation modes, the calculation amount is larger and larger, and therefore the t is passedk-1Estimating the nuclear density of the degradation state at the moment to obtain tkThe kernel density estimation of the dynamic transition of the time degradation state realizes the real-time update of the probability density function of the residual service life, and is very necessary to reduce the calculation complexity;
substituting equation (20) into equation (14) can obtain the remaining life distribution and predict the probability density function of the real-time remaining life considering the transition of the degradation stateComprises the following steps:
when the m-th sample is received at the current moment, the self-adaptive smoothing parameter of the m-th sample is updatedIn thatCan obtain the real-time residual life distribution at the current moment on the basisThereby continuously obtaining new degradation characteristic incrementThen, the predicted residual life probability density function is updated in real time, and the calculation complexity is reduced;
and 5, predicting the residual life of the gear by considering the residual life prediction model of the degraded state transition:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110631246.9A CN113468801B (en) | 2021-06-07 | 2021-06-07 | Gear nuclear density estimation residual life prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110631246.9A CN113468801B (en) | 2021-06-07 | 2021-06-07 | Gear nuclear density estimation residual life prediction method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113468801A true CN113468801A (en) | 2021-10-01 |
CN113468801B CN113468801B (en) | 2024-03-26 |
Family
ID=77872426
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110631246.9A Active CN113468801B (en) | 2021-06-07 | 2021-06-07 | Gear nuclear density estimation residual life prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113468801B (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2001016880A2 (en) * | 1999-08-30 | 2001-03-08 | Synes Nv | Topographic map and methods and systems for data processing therewith |
CN105740625A (en) * | 2016-01-31 | 2016-07-06 | 太原科技大学 | Real time residual life prediction method of gear |
CN109883691A (en) * | 2019-01-21 | 2019-06-14 | 太原科技大学 | The gear method for predicting residual useful life that kernel estimates and stochastic filtering integrate |
CN110174261A (en) * | 2019-05-23 | 2019-08-27 | 太原科技大学 | The real-time method for predicting residual useful life of gear of more amount of degradation monitorings |
WO2019174142A1 (en) * | 2018-03-14 | 2019-09-19 | 山东科技大学 | Multi-mode degradation process modelling and remaining service life prediction method |
CN112417612A (en) * | 2020-10-15 | 2021-02-26 | 浙江工业大学 | Method for tracking degradation state and evaluating failure aggregation risk of wind power gear box |
-
2021
- 2021-06-07 CN CN202110631246.9A patent/CN113468801B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2001016880A2 (en) * | 1999-08-30 | 2001-03-08 | Synes Nv | Topographic map and methods and systems for data processing therewith |
CN105740625A (en) * | 2016-01-31 | 2016-07-06 | 太原科技大学 | Real time residual life prediction method of gear |
WO2019174142A1 (en) * | 2018-03-14 | 2019-09-19 | 山东科技大学 | Multi-mode degradation process modelling and remaining service life prediction method |
CN109883691A (en) * | 2019-01-21 | 2019-06-14 | 太原科技大学 | The gear method for predicting residual useful life that kernel estimates and stochastic filtering integrate |
CN110174261A (en) * | 2019-05-23 | 2019-08-27 | 太原科技大学 | The real-time method for predicting residual useful life of gear of more amount of degradation monitorings |
CN112417612A (en) * | 2020-10-15 | 2021-02-26 | 浙江工业大学 | Method for tracking degradation state and evaluating failure aggregation risk of wind power gear box |
Non-Patent Citations (3)
Title |
---|
张卫贞;曾建潮;石慧;董增寿;: "基于核密度估计的实时剩余寿命预测", 计算机集成制造系统, no. 07, 15 July 2020 (2020-07-15) * |
张妍;王村松;陆宁云;姜斌;: "基于退化特征相似性的航空发动机寿命预测", 系统工程与电子技术, no. 06, 29 March 2019 (2019-03-29) * |
石慧;宋仁旺;张岩;董增寿;: "基于核密度估计和随机滤波理论的齿轮箱剩余寿命预测方法", 计算机集成制造系统, no. 03, 31 December 2020 (2020-12-31) * |
Also Published As
Publication number | Publication date |
---|---|
CN113468801B (en) | 2024-03-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Filev et al. | An industrial strength novelty detection framework for autonomous equipment monitoring and diagnostics | |
Yang et al. | Machine condition prognosis based on regression trees and one-step-ahead prediction | |
Niu et al. | Intelligent condition monitoring and prognostics system based on data-fusion strategy | |
Wang et al. | A joint particle filter and expectation maximization approach to machine condition prognosis | |
Peng et al. | A prognosis method using age-dependent hidden semi-Markov model for equipment health prediction | |
CN109883691B (en) | Gear residual life prediction method integrating kernel estimation and random filtering | |
Boutros et al. | Detection and diagnosis of bearing and cutting tool faults using hidden Markov models | |
Yu et al. | Model-based health monitoring for a vehicle steering system with multiple faults of unknown types | |
Lin et al. | Using principal components in a proportional hazards model with applications in condition-based maintenance | |
CN109212966B (en) | Multi-working-condition dynamic benchmarking mechanical equipment residual life prediction method | |
Cerrada et al. | A fuzzy transition based approach for fault severity prediction in helical gearboxes | |
CN110174261B (en) | Gear real-time residual life prediction method based on multi-degradation monitoring | |
Mosallam et al. | Component based data-driven prognostics for complex systems: Methodology and applications | |
Salunkhe et al. | Prediction of Remaining Useful Life of mechanical components-a Review | |
CN113468721A (en) | Method for predicting residual life of gear and bearing in gear reduction box | |
Martin-del-Campo et al. | Dictionary learning approach to monitoring of wind turbine drivetrain bearings | |
He et al. | Prognostic analysis based on hybrid prediction method for axial piston pump | |
CN114386563A (en) | Bayesian context aggregation of neural processes | |
CN114091525A (en) | Rolling bearing degradation trend prediction method | |
Scheffer et al. | A comparative evaluation of neural networks and hidden Markov models for monitoring turning tool wear | |
Wang et al. | Multiple model particle filtering for bearing life prognosis | |
CN113468801A (en) | Method for predicting residual life of gear by estimating nuclear density | |
Zhang et al. | A multi-fault modeling approach for fault diagnosis and failure prognosis of engineering systems | |
CN118043753A (en) | System and method for estimating remaining useful life of bearing | |
Mollazade et al. | An intelligent model based on data mining and fuzzy logic for fault diagnosis of external gear hydraulic pumps |
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 |