CN109460574A - A kind of prediction technique of aero-engine remaining life - Google Patents
A kind of prediction technique of aero-engine remaining life Download PDFInfo
- Publication number
- CN109460574A CN109460574A CN201811170972.XA CN201811170972A CN109460574A CN 109460574 A CN109460574 A CN 109460574A CN 201811170972 A CN201811170972 A CN 201811170972A CN 109460574 A CN109460574 A CN 109460574A
- Authority
- CN
- China
- Prior art keywords
- sample
- remaining life
- data
- tested
- aero
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000006731 degradation reaction Methods 0.000 claims abstract description 49
- 230000015556 catabolic process Effects 0.000 claims abstract description 48
- 230000003412 degenerative effect Effects 0.000 claims abstract description 33
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 12
- 238000000513 principal component analysis Methods 0.000 claims abstract description 10
- 238000012216 screening Methods 0.000 claims abstract description 7
- 239000012634 fragment Substances 0.000 claims abstract description 4
- 239000000523 sample Substances 0.000 claims description 62
- 230000007850 degeneration Effects 0.000 claims description 21
- 239000011159 matrix material Substances 0.000 claims description 6
- 239000013074 reference sample Substances 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 230000008859 change Effects 0.000 claims description 5
- 238000013480 data collection Methods 0.000 claims description 5
- 230000001174 ascending effect Effects 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 230000007704 transition Effects 0.000 claims description 3
- 230000008569 process Effects 0.000 abstract description 5
- 238000009499 grossing Methods 0.000 abstract 1
- 230000006870 function Effects 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 6
- 238000005259 measurement Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 230000036541 health Effects 0.000 description 3
- 239000000203 mixture Substances 0.000 description 3
- 238000012423 maintenance Methods 0.000 description 2
- 230000007257 malfunction Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical group C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000008450 motivation Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
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
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M15/00—Testing of engines
- G01M15/04—Testing internal-combustion engines
- G01M15/05—Testing internal-combustion engines by combined monitoring of two or more different engine parameters
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- Chemical & Material Sciences (AREA)
- Combustion & Propulsion (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of prediction techniques of aero-engine remaining life, the method passes through the correlation Relief algorithm carries out degenerative character screening, reduces feature with principal component analysis (PCA) between multidimensional performance characteristic parameter measurable in aero-engine performance degenerative process first, the orthogonal degenerative character of more succinct, effective low-dimensional is extracted, uses the practical Degradation path of Kernel Smoothing fit to degenerative character is obtained;Then the similitude matching of Degradation path is carried out, it finds and the characteristic fragment set in one group of most like historical sample of sample to be tested path segment, data distribution is sought to the remaining life of similar features segment by the way of Density Estimator, the life prediction estimated value of sample to be tested is obtained by Density Weighted method, achievees the purpose that predict aero-engine remaining life.
Description
Technical field
The invention belongs to the life predictions of aero-engine, are related to a kind of similitude of aero-engine performance degenerative character
Modeling and predicting residual useful life technology, and in particular to a kind of prediction technique of aero-engine remaining life.
Background technique
Aero-engine is the core equipment of all kinds of aviation aircrafts, reliability, maintainability, safety, protection and
The focus of testability (RMSST) industry circle always.Predicting residual useful life is by analytical equipment historical performance degradation trend, in advance
Remaining time of the measurement equipment from current time to ultimate failure.Accurate life prediction can be improved equipment RMSST performance, reduce
Maintenance cost solves the Affordability of system maintenance.
Aero-engine is a complicated mechanical electronic hydraulic Electromagnetic Coupling System, and performance degradation is by multiple state variable dynamics
Change and intercouple as a consequence it is difficult to obtain its accurate physical failure model.Based on the technology of data-driven because it is disobeyed
Rely the physical failure mechanism in equipment, it is only necessary to acquire, analyze the history and online monitoring data of engine, in practical applications
It is very popular.
The life prediction mainstream technology of data-driven is directly to find performance degradation based on modeling and the regression forecasting of degenerating
Mapping relations between process and remaining life.The prior art much uses linear regression model foundation relationship, for example, by pair
The damage data of engine carries out linear fit, damage estimates of parameters is solved using Maximum-likelihood estimation, with test data data
Model of growth is more damaged, the probability density function of remaining time is then sought, taking its intermediate value is the remaining life of engine, real
It is difficult to directly apply to the complex object with more amount of degradations and non-linear degradation process on border.The prior art also passes through multi-source number
A comprehensive health degree index is first constructed according to fusion and sets failure threshold, is then established the Wiener-Hopf equation of nonlinear drift, is led to
It crosses remaining life distribution and carries out life prediction.But the building of health degree index and the setting of failure threshold are cores therein
And difficulties, determine the accuracy and reliability of forecasting technique in life span.In addition, the prior art commonly assumes that all samples
Degenerative character has consistent or known healthy initial value, some methods are it is also assumed that historical sample and sample to be tested degenerative character length
Unanimously, these assume all to bring many difficulties for the practical application of this kind of technology.
Summary of the invention
Goal of the invention: in view of the above shortcomings of the prior art, the present invention provides a kind of the pre- of aero-engine remaining life
Survey method, this method is according to the degenerative character of extraction aero-engine, the purpose of realization predicting residual useful life.
Technical solution: a kind of prediction technique of aero-engine remaining life, the method includes according to aero-engine
The similitude of performance degradation feature is modeled, and is included the following steps:
(1) multivariable degenerative character is screened: being returned first, in accordance with time series of the degree of degeneration to history lifetime data
Class, by Relief algorithm to two class data screening degenerative characters, the bigger tribute for indicating this feature to degeneration of the weight of feature
Offer it is bigger, otherwise contribution it is small;
(2) degenerative character is extracted: being converted, is obtained by principal component analysis to the feature screened through Relief algorithm
To the principal component feature for being free of redundancy, practical degeneration rail is fitted to principal component time series using Kernel smooth method
Mark;
(3) similarity assessment: the track indicated according to degenerative character to the Degradation path of sample to be tested and historical sample into
Row similarity measurement finds Degradation path piece similar with sample to be tested based on similarity degree in history reference sample
Duan Jihe, and calculate the remaining life of similar degradation path segment;
(4) model is comprehensive: the data point of similar degradation path segment remaining life are sought by way of Density Estimator
Cloth assigns corresponding weight to similar fragments according to density value, and the life prediction result of sample to be tested is similar degradation track piece
The weighted sum of section remaining life, is finally limited by maximum value of the segmented model of remaining life to remaining life, and amendment is pre-
Survey result.
Further, classification described in step (1) forms the low one kind of degree of degeneration by earlier time points data, when latter stage
Between point data composition degree of degeneration it is high conduct it is another kind of, setting before total time 5% is SMS message, and the data in this period move back
Change degree is low, removes time tag and sticks class label 0 for it, total time last 5% is time in latter stage, data in this period
It is that degree of degeneration is high close to failure, removes time tag and stick class label 1 for it.
Further, specific step is as follows for step (1):
(11) the preceding T that all historical engine degraded data collection are initially run is taken out0The observation data conduct at a time point
Performance normal data, Q0={ Tree0(t) | t=1,2 ..., T0, t indicates runing time point;
(12) take out the performance failure data of all historical engine degraded data collection, the data be each data set most
T afterwards1The observation data at a time point,tEFor the observation end time of historical sample
Point;
(13) using Relief algorithm to Q0∪Q1Data set screens degenerative character collection F'={ f1,f2,…,fm}。
Further, specific step is as follows for the step (2):
(21) to new feature set F'={ f1,f2,…,fmData use principal component analysis, will wherein there is correlation
Degenerate variable is converted into p linear incoherent principal component characteristic Y=(y1,y2,…,yp)T,It is p-th
The time series of principal component records transition matrix
(22) it is denoised using time series of the Kernel smooth method to principal component feature, eliminates the interference of noise, obtain
To the practical Degradation path f (t) of degenerative character.
Further, the step (3) includes the following steps:
For c-th of sample to be tested keeping characteristics collection F ',The Degradation path obtained after transformation and Kernel are smooth,
Its p-th it is main at state for time sequence be denoted asIts time observed length is tI, calculate sample to be tested with
Similarity between first of historical sample are as follows:
Wherein, p indicates p-th of principal component, σpFor its standard deviation, τ is delay parameter, sets maximum delay as τmax, remember τ*
=min (tE-tI,τmax), sample to be tested is calculated for l history reference sample in τ ∈ [0, τ*] similarity in range Calculating be expressed as:
Sample to be tested and all history are calculated referring to the similarity between Degradation pathIt will
SimcIn all values by ascending sort from small to large, select top n value, calculate and the most like N number of degeneration rail of sample to be tested
The remaining life r of mark segmentc=(r1,r2,…,rN)T, wherein the calculation expression of remaining life is as follows:
R=tE-tI-τ+1。
Further, the step (4) includes:
(41) similar sample residual service life r is sought using Density EstimatorcData distribution, Density Estimator data distribution
Mode are as follows:
Wherein, K () is kernel function, it is desirable that meets symmetry and ∫ K (r) dr=1, with the minimum principle of mean square error
Window width h is selected, determines that weighting function is probability density function, the predicting residual useful life result of sample to be tested is similar degradation track
The remaining life Density Weighted of segment:
Remaining life due to seeking similar sample here is distributed, the requirement based on data distribution, the data of similar sample
Amount should meet N >=20;
(42) segmented model for using remaining life, i.e., first undergo performance normal phase, moves back after running a period of time
Change, remaining life is divided into two stages: constant stage and linear decrease stage, the maximum value of remaining life being limited
System, it is assumed that maximum remaining life is RULmax, finally prediction result is corrected are as follows:
The utility model has the advantages that its significant effect is compared with prior art: the first, the present invention is using based on data
Mathematical Modeling Methods, degenerative character screening is carried out based on the measurable multivariable performance parameter of aero-engine and is extracted, and
Degradation path is fitted on the basis of degenerative character, can effectively describe the degenerative process of engine;The second, the present invention is similar in sample
The uncertain problem of initial health is changed into the delay problem of Degradation path segment, with practical hair in property assessment algorithm
Motivation degenerate case is more consistent;Third, sample to be tested remaining life added by the density of multiple most like Degradation path segments
Power avoids the Biased estimator generated by single model, improves the robustness of precision of prediction and prediction model;4th, institute of the present invention
The method of proposition is support with data, does not need to establish complicated mathematical model, is predicted from the similitude angle between sample remaining
Service life has certain practicability.
Detailed description of the invention
Fig. 1 is the flow chart of prediction technique of the present invention;
Fig. 2 is the practical Degradation path figure of three history reference samples;
Fig. 3 is the schematic diagram of similarity assessment of the present invention;
Fig. 4 is the true lifetime of sample to be tested and the comparing result schematic diagram of prediction result.
Specific embodiment
In order to which technical solution disclosed by the invention is described in detail, with reference to the accompanying drawings of the specification and specific embodiment do into
The elaboration of one step.
A kind of prediction technique of aero-engine remaining life of the invention, belongs to aero-engine performance degenerative character
Similitude modeling and predicting residual useful life technology.The present invention passes through Relief algorithm to the multi-dimensional time sequence of aero-engine first
Column data carries out degenerative character screening, principal component analysis (PCA), Kernel smoothly extract the orthogonal degenerative character of low-dimensional and feature
Practical Degradation path;Then the matching of Degradation path similitude is carried out, one group most like with sample to be tested path segment is found and goes through
Degradation path set of segments in history sample obtains the life prediction estimated value of sample to be tested, side using Density Weighted method
Method flow chart is as shown in Figure 1.
The C-MAPSS aero-engine of present invention combination NASA emulates data set, which is divided into training set and test
Collection, training set respectively include 100 groups of units for running to malfunction, and for the running simulation of each unit, experience is complete
Arrive malfunction (with different degrees of running-in wear and manufacture variation) from normal, record should each airborne period in the process
3 operating condition values and 21 measurement value sensors, as shown in table 1.
The description of 1 aero-engine sensor data set of table
Detailed process includes the following steps:
Step 1, the screening of multivariable degenerative character: it is carried out first, in accordance with time series of the degree of degeneration to history lifetime data
Sort out, earlier time points data form the low one kind of degree of degeneration, and latter stage time point data forms the high one kind of degree of degeneration.It adopts
With Relief algorithm to two class data screening degenerative characters, the weight of feature is bigger, indicates that this feature is bigger to the contribution of degeneration,
Conversely, contribution is small.
Step 2, degenerative character are extracted: being converted, obtained without superfluous using PCA to the feature screened through Relief
The principal component feature of remaining information is smoothly fitted practical Degradation path to principal component time series using Kernel.
Step 3, similarity assessment: the degeneration from the track that degenerative character indicates, to sample to be tested and historical sample
Track carries out similarity measurement, and based on similarity degree, degeneration similar with sample to be tested is found in history reference sample
Path segment set, and calculate the remaining life of similar degradation path segment.
Step 4, model are comprehensive: the data of similar degradation path segment remaining life are sought by way of Density Estimator
Distribution assigns corresponding weight to similar fragments according to density value, and the life prediction result of sample to be tested is similar degradation track
The weighted sum of segment remaining life finally corrects prediction result by the segmented model of remaining life.
1) multivariable degenerative character is screened:
Assuming that the degraded data sample size of existing historical engine is L, experience E occur from starting to failure for each sample
View of time measuring point ti(i=1,2 ... E), characteristic variable has M dimension.tiThe observation x at placeiIt is made of M dimensional feature value, i.e. xi=(x1,
x2,…,xM)T, and it is t that every one-dimensional characteristic variable, which is all length,ETime series.
The preceding T that all historical engine degraded data collection are initially run is taken out first0The observation data conduct at a time point
Performance normal data Q0={ Tree0(t) | t=1,2 ..., T0, t indicates that runing time point similarly takes out each data set most
T afterwards1The observation data at a time point as performance failure data,Using
Relief algorithm is to Q0∪Q1Data set screens degenerative character collection, and u is arest neighbors number, and k is cycle-index (k≤L), specific steps
It is as follows:
(11) the weight w of each feature is initializedi=0;
(12)RjFor the sample randomly selected in X, j is followed badly to k from 1, is found with RjCentered on, u in similar sample
Nearest samples form u nearest samples in nearhit and inhomogeneity and form nearmiss:
(13) to all characteristic value xi(i=1,2 ..., M) is calculated by following formula update weight respectively:
(14) circulation terminates, according to wiTo feature xiIt sorts by size.
Wherein, diff (xi,Rj, nearhit) and indicate RjWith an example in nearhit for feature xiDifference, can adopt
With Euclidean distance, similarly diff (xi,Rj, nearmiss) and indicate RjWith an example in nearmiss for feature xiDifference.
Avg () indicates the average value of difference between all these samples.It is arranged feature weight threshold value λ=avg (w), through above-mentioned algorithm
Filter out the m feature that feature weight is greater than λ, new feature set F'={ f1,f2,…,fm}。
2) degenerative character is extracted:
Data set after Feature Selection has m dimensional feature variable,Wherein,Indicate the time series of j-th of characteristic variable, PCA extracts the specific steps of principal component to m dimension data are as follows:
(21) to X standardization, (mean value returns 0,1) variance is returned, and matrix is X after standardization first*;
(22) X is sought*The eigenvalue λ of covariance matrix (m*m dimension)j(j=1,2 ..., m) and feature vector;
(23) to λjValue is arranged by descending from big to small, i.e. λ1≥λ2≥…≥λm, calculate contribution rate of accumulative totalIt usually requires that θ is greater than 85%, takes the corresponding feature vector of preceding p characteristic value sequentially composition characteristic vector matrix
(24) by X*?It projects, the calculation formula of the principal component Y after obtaining dimensionality reduction are as follows:
Wherein, Y is defined as the principal component feature of X, y1Referred to as first principal component, y2Referred to as Second principal component, and so on;
P principal component of reservation should retain most variation characteristics (general threshold value takes 85% or more) of initial data, and each other
Between it is uncorrelated,For the time series of p-th of principal component, transition matrix is recorded
(25) F is definedlFor the degradation model of first of historical sample, FlFunction representation by principal component feature about the time:
Fl: y=f (t)+ε, 0≤t≤tE (3)
Wherein, ε is white Gaussian noise, and f (t) is characterized the practical Degradation path changed over time, FlModel is thought to observe
The degenerative character arrived, is here main composition characteristics, is made of actual characteristic Degradation path plus noise.The present invention uses Kernel
Smooth extracts Degradation path to degeneration orthogonal characteristic, to principal componentSmooth manner are as follows:
Wherein, K () is kernel function, usually using Gaussian kernel:
ρ is width parameter in above formula, and the mode for generalling use cross validation chooses suitable width parameter.Low-dimensional is orthogonal to move back
The result for changing its Degradation path of feature is as shown in Figure 2.
3) similarity assessment:
For c-th of sample to be tested keeping characteristics collection F ',The Degradation path obtained after transformation and Kernel are smooth,
Its p-th it is main at state for time sequence be denoted asIts time observed length is tI, calculate sample to be tested with
Similarity between first of historical sample are as follows:
Wherein, p indicates p-th of principal component, σpFor its standard deviation, τ is delay parameter.Remaining life based on similitude is pre-
Survey technology schematic diagram is as shown in figure 3, theoretically the maximum value of τ is tE-tIIf but the observation time point of current sample terminate compared with
It is early, i.e. tIIt is smaller, lead to tE-tIIt is very big.And the original state of sample to be tested and historical sample will not have a long way to go, and therefore, setting
Maximum delay is τmax, remember τ*=min (tE-tI,τmax), sample to be tested and first of history reference sample are calculated in τ ∈ [0, τ*]
Similarity in range Calculation are as follows:
Sample to be tested and all history are calculated referring to the similarity between Degradation pathIt will
SimcIn all values by ascending sort from small to large, select top n value, calculate and the most like N number of degeneration rail of sample to be tested
The remaining life r of mark segmentc=(r1,r2,…,rN)T, the wherein calculation of remaining life are as follows:
R=tE-tI-τ+1 (8)
4) model is comprehensive:
Similar sample residual service life r is sought using Density EstimatorcData distribution, the side of Density Estimator data distribution
Formula are as follows:
K () is kernel function, it is desirable that meets symmetry and ∫ K (r) dr=1, selects window with the minimum principle of mean square error
Wide h.Determine that weighting function is probability density function, the predicting residual useful life result of sample to be tested is similar degradation path segment
Remaining life Density Weighted:
Remaining life due to seeking similar sample here is distributed, the requirement based on data distribution, the data of similar sample
Amount should meet N >=20.
Using the segmented model of remaining life, performance normal phase is specially first undergone, is moved back after running a period of time
Change.Remaining life is divided into two stages: constant stage and linear decrease stage by the present invention, to the maximum value of remaining life into
Row limitation, it is assumed that maximum remaining life is RULmax, finally prediction result is corrected are as follows:With this
Invention is to result and its true lifetime comparative analysis result of the progress life prediction of all samples to be tested as shown in figure 4, pre- in detail
Survey the results are shown in Table 2.
The prediction result of 2 sample to be tested of table
Claims (6)
1. a kind of prediction technique of aero-engine remaining life, it is characterised in that: the method includes according to aero-engine
The similitude of performance degradation feature is modeled, and is included the following steps:
(1) multivariable degenerative character is screened: being sorted out first, in accordance with time series of the degree of degeneration to history lifetime data, is led to
Relief algorithm is crossed to two class data screening degenerative characters, the weight of feature is bigger to indicate this feature to the contribution of degeneration more
Greatly, otherwise contribution is small;
(2) degenerative character is extracted: being converted, is obtained not by principal component analysis to the feature screened through Relief algorithm
Principal component feature containing redundancy is fitted practical Degradation path to principal component time series using Kernel smooth method;
(3) similarity assessment: the track indicated according to degenerative character carries out phase to the Degradation path of sample to be tested and historical sample
It is measured like property, based on similarity degree, Degradation path segment collection similar with sample to be tested is found in history reference sample
It closes, and calculates the remaining life of similar degradation path segment;
(4) model is comprehensive: the data distribution of similar degradation path segment remaining life, root are sought by way of Density Estimator
Corresponding weight is assigned to similar fragments according to density value, the life prediction result of sample to be tested is that similar degradation path segment is remaining
The weighted sum in service life is finally limited by maximum value of the segmented model of remaining life to remaining life, corrects prediction result.
2. a kind of prediction technique of aero-engine remaining life according to claim 1, it is characterised in that: step (1)
The classification forms the low one kind of degree of degeneration by earlier time points data, and it is high that latter stage time point data forms degree of degeneration
As another kind of, setting before total time 5% is SMS message, and the data degradation degree in this period is low, and removal time tag is simultaneously
Class label 0 is sticked for it, total time last 5% is the time in latter stage, and data are that degree of degeneration is high close to failure in this period, is gone
Class label 1 is sticked except time tag and for it.
3. a kind of prediction technique of aero-engine remaining life according to claim 1, it is characterised in that: step (1)
Specific step is as follows:
(11) the preceding T that all historical engine degraded data collection are initially run is taken out0The observation data at a time point as performance just
Regular data, Q0={ Tree0(t) | t=1,2 ..., T0, t indicates runing time point;
(12) the performance failure data of all historical engine degraded data collection are taken out, the data are the last T of each data set1
The observation data at a time point,tEFor the observation end time point of historical sample;
(13) using Relief algorithm to Q0∪Q1Data set screens degenerative character collection F'={ f1,f2,…,fm}。
4. a kind of prediction technique of aero-engine remaining life according to claim 1, it is characterised in that: the step
(2) specific step is as follows:
(21) to new feature set F'={ f1,f2,…,fmData use principal component analysis, will wherein there is the degeneration of correlation
Variable is converted into p linear incoherent principal component characteristic Y=(y1,y2,…,yp)T,For p-th it is main at
The time series divided records transition matrix
(22) it is denoised using time series of the Kernel smooth method to principal component feature, eliminates the interference of noise, moved back
Change the practical Degradation path f (t) of feature.
5. a kind of prediction technique of aero-engine remaining life according to claim 1, which is characterized in that the step
(3) include the following steps:
For c-th of sample to be tested keeping characteristics collection F ',The Degradation path obtained after transformation and Kernel are smooth, pth
It is a it is main at state for time sequence be denoted asIts time observed length is tI, calculate sample to be tested and first
Similarity between historical sample are as follows:
Wherein, p indicates p-th of principal component, σpFor its standard deviation, τ is delay parameter, sets maximum delay as τmax, remember τ*=min
(tE-tI,τmax), sample to be tested is calculated for l history reference sample in τ ∈ [0, τ*] similarity in range Calculating be expressed as:
Sample to be tested and all history are calculated referring to the similarity between Degradation pathBy SimcIn
All values by ascending sort from small to large, select top n value, calculate and the most like N number of Degradation path segment of sample to be tested
Remaining life rc=(r1,r2,…,rN)T, wherein the calculation expression of remaining life is as follows:
R=tE-tI-τ+1。
6. a kind of prediction technique of aero-engine remaining life according to claim 1, it is characterised in that: the step
(4) include:
(41) similar sample residual service life r is sought using Density EstimatorcData distribution, the side of Density Estimator data distribution
Formula are as follows:
Wherein, K () is kernel function, it is desirable that meets symmetry and ∫ K (r) dr=1, with the minimum principle selection of mean square error
Window width h determines that weighting function is probability density function, and the predicting residual useful life result of sample to be tested is similar degradation path segment
Remaining life Density Weighted:
Remaining life due to seeking similar sample here is distributed, and the data volume of the requirement based on data distribution, similar sample is answered
Meet N >=20;
(42) segmented model for using remaining life, i.e., first undergo performance normal phase, degenerates after running a period of time, will
Remaining life is divided into two stages: constant stage and linear decrease stage, limiting the maximum value of remaining life, it is assumed that
Maximum remaining life is RULmax, finally prediction result is corrected are as follows:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811170972.XA CN109460574A (en) | 2018-10-09 | 2018-10-09 | A kind of prediction technique of aero-engine remaining life |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811170972.XA CN109460574A (en) | 2018-10-09 | 2018-10-09 | A kind of prediction technique of aero-engine remaining life |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109460574A true CN109460574A (en) | 2019-03-12 |
Family
ID=65607336
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811170972.XA Pending CN109460574A (en) | 2018-10-09 | 2018-10-09 | A kind of prediction technique of aero-engine remaining life |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109460574A (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110333265A (en) * | 2019-07-11 | 2019-10-15 | 中国中车股份有限公司 | A kind of method and system for predicting locomotive engine exhaust expansion bellows remaining life |
CN110569623A (en) * | 2019-09-19 | 2019-12-13 | 西安工程大学 | Method for predicting service life of spinning frame motor |
CN110717379A (en) * | 2019-08-28 | 2020-01-21 | 南京康尼机电股份有限公司 | Health assessment method for subway car door key components based on feature fusion |
CN110757510A (en) * | 2019-10-31 | 2020-02-07 | 广东工业大学 | Method and system for predicting remaining life of robot |
CN112052624A (en) * | 2020-08-13 | 2020-12-08 | 五邑大学 | Method for estimating and predicting bridge life based on big data |
CN112149746A (en) * | 2020-09-27 | 2020-12-29 | 中国商用飞机有限责任公司北京民用飞机技术研究中心 | Method and device for training model of remaining use times of undercarriage and computer equipment |
CN112149316A (en) * | 2019-11-04 | 2020-12-29 | 中国人民解放军国防科技大学 | Aero-engine residual life prediction method based on improved CNN model |
CN112348120A (en) * | 2020-12-01 | 2021-02-09 | 江苏南高智能装备创新中心有限公司 | Numerical control machine tool degradation track mode identification method and device |
CN112560234A (en) * | 2020-12-01 | 2021-03-26 | 江苏南高智能装备创新中心有限公司 | Method and device for calculating residual life estimation probability of numerical control machine tool and network side server |
CN112731159A (en) * | 2020-12-23 | 2021-04-30 | 江苏省电力试验研究院有限公司 | Method for pre-judging and positioning battery fault of battery compartment of energy storage power station |
CN113487087A (en) * | 2021-07-06 | 2021-10-08 | 新智数字科技有限公司 | Method and device for predicting service life of equipment, computer equipment and computer-readable storage medium |
CN113742883A (en) * | 2020-11-20 | 2021-12-03 | 国网河北省电力有限公司雄安新区供电公司 | Method for dividing service life cycle of alternating current contactor based on multivariate time sequence |
CN113761722A (en) * | 2021-08-18 | 2021-12-07 | 航天科工海鹰集团有限公司 | Spacecraft multi-working-condition service life prediction method based on PCA |
CN114707098A (en) * | 2022-01-22 | 2022-07-05 | 西北工业大学 | Aircraft engine performance degradation state evaluation method based on multi-source sensor state distance |
CN115982942A (en) * | 2022-11-25 | 2023-04-18 | 浙江长龙航空有限公司 | Method, device, equipment and storage medium for predicting residual life of aviation precooler |
CN117169709A (en) * | 2023-10-09 | 2023-12-05 | 中国民航大学 | Aviation relay fatigue performance testing method, device and storage medium |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102789545A (en) * | 2012-07-12 | 2012-11-21 | 哈尔滨工业大学 | Method for predicating remaining life of turbine engine based on degradation model matching |
CN107451392A (en) * | 2017-06-23 | 2017-12-08 | 山东科技大学 | A kind of method for predicting residual useful life containing multiple dependent degeneration processes |
-
2018
- 2018-10-09 CN CN201811170972.XA patent/CN109460574A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102789545A (en) * | 2012-07-12 | 2012-11-21 | 哈尔滨工业大学 | Method for predicating remaining life of turbine engine based on degradation model matching |
CN107451392A (en) * | 2017-06-23 | 2017-12-08 | 山东科技大学 | A kind of method for predicting residual useful life containing multiple dependent degeneration processes |
Non-Patent Citations (2)
Title |
---|
TIANYI WANG: "Trajectory Similarity Based Prediction for Remaining Useful Life Estimation", 《DISSERTATION & THESES-GRADWORKS》 * |
朱丽娜 等: "基于ReliefF-PCA和SVM的发动机故障诊断方法研究", 《北京化工大学学报》 * |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110333265A (en) * | 2019-07-11 | 2019-10-15 | 中国中车股份有限公司 | A kind of method and system for predicting locomotive engine exhaust expansion bellows remaining life |
CN110717379A (en) * | 2019-08-28 | 2020-01-21 | 南京康尼机电股份有限公司 | Health assessment method for subway car door key components based on feature fusion |
CN110569623A (en) * | 2019-09-19 | 2019-12-13 | 西安工程大学 | Method for predicting service life of spinning frame motor |
CN110757510A (en) * | 2019-10-31 | 2020-02-07 | 广东工业大学 | Method and system for predicting remaining life of robot |
CN112149316A (en) * | 2019-11-04 | 2020-12-29 | 中国人民解放军国防科技大学 | Aero-engine residual life prediction method based on improved CNN model |
CN112149316B (en) * | 2019-11-04 | 2022-04-19 | 中国人民解放军国防科技大学 | Aero-engine residual life prediction method based on improved CNN model |
CN112052624A (en) * | 2020-08-13 | 2020-12-08 | 五邑大学 | Method for estimating and predicting bridge life based on big data |
CN112149746A (en) * | 2020-09-27 | 2020-12-29 | 中国商用飞机有限责任公司北京民用飞机技术研究中心 | Method and device for training model of remaining use times of undercarriage and computer equipment |
CN112149746B (en) * | 2020-09-27 | 2024-02-06 | 中国商用飞机有限责任公司北京民用飞机技术研究中心 | Landing gear remaining use number model training method and device and computer equipment |
CN113742883A (en) * | 2020-11-20 | 2021-12-03 | 国网河北省电力有限公司雄安新区供电公司 | Method for dividing service life cycle of alternating current contactor based on multivariate time sequence |
CN112348120A (en) * | 2020-12-01 | 2021-02-09 | 江苏南高智能装备创新中心有限公司 | Numerical control machine tool degradation track mode identification method and device |
CN112560234A (en) * | 2020-12-01 | 2021-03-26 | 江苏南高智能装备创新中心有限公司 | Method and device for calculating residual life estimation probability of numerical control machine tool and network side server |
CN112560234B (en) * | 2020-12-01 | 2024-03-12 | 江苏南高智能装备创新中心有限公司 | Calculation method and device of residual life estimation probability of numerical control machine tool and network side server |
CN112731159B (en) * | 2020-12-23 | 2023-09-22 | 江苏省电力试验研究院有限公司 | Method for pre-judging and positioning battery faults of battery compartment of energy storage power station |
CN112731159A (en) * | 2020-12-23 | 2021-04-30 | 江苏省电力试验研究院有限公司 | Method for pre-judging and positioning battery fault of battery compartment of energy storage power station |
CN113487087A (en) * | 2021-07-06 | 2021-10-08 | 新智数字科技有限公司 | Method and device for predicting service life of equipment, computer equipment and computer-readable storage medium |
CN113761722A (en) * | 2021-08-18 | 2021-12-07 | 航天科工海鹰集团有限公司 | Spacecraft multi-working-condition service life prediction method based on PCA |
CN114707098A (en) * | 2022-01-22 | 2022-07-05 | 西北工业大学 | Aircraft engine performance degradation state evaluation method based on multi-source sensor state distance |
CN114707098B (en) * | 2022-01-22 | 2024-03-08 | 西北工业大学 | Aeroengine performance degradation state evaluation method based on multisource sensor state distance |
CN115982942A (en) * | 2022-11-25 | 2023-04-18 | 浙江长龙航空有限公司 | Method, device, equipment and storage medium for predicting residual life of aviation precooler |
CN117169709A (en) * | 2023-10-09 | 2023-12-05 | 中国民航大学 | Aviation relay fatigue performance testing method, device and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109460574A (en) | A kind of prediction technique of aero-engine remaining life | |
CN111163057B (en) | User identification system and method based on heterogeneous information network embedding algorithm | |
CN111222549B (en) | Unmanned aerial vehicle fault prediction method based on deep neural network | |
CN111222290B (en) | Multi-parameter feature fusion-based method for predicting residual service life of large-scale equipment | |
CN110210681B (en) | Prediction method of PM2.5 value of monitoring station based on distance | |
CN108009674A (en) | Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks | |
Lai et al. | Online pattern matching and prediction of incoming alarm floods | |
CN103473540B (en) | The modeling of intelligent transportation system track of vehicle increment type and online method for detecting abnormality | |
CN108875118B (en) | Method and device for evaluating accuracy of prediction model of silicon content of blast furnace molten iron | |
CN106934237A (en) | Radar cross-section redaction measures of effectiveness creditability measurement implementation method | |
CN108446619A (en) | Face critical point detection method and device based on deeply study | |
CN113065702B (en) | Landslide displacement multi-linear prediction method based on ST-SEEP segmentation method and space-time ARMA model | |
CN111861272A (en) | Multi-source data-based complex electromechanical system abnormal state detection method | |
CN105334185A (en) | Spectrum projection discrimination-based near infrared model maintenance method | |
CN104239722A (en) | Forecasting method based on recognition of correlational relationship between factors | |
CN113836808A (en) | PM2.5 deep learning prediction method based on heavy pollution feature constraint | |
CN108022123A (en) | The automatic adjusting method and device of a kind of business model | |
CN115270266A (en) | Slope monitoring state risk judgment method based on improved analytic hierarchy process | |
Hamidisepehr et al. | Moisture content classification of soil and stalk residue samples from spectral data using machine learning | |
CN113112067A (en) | Method for establishing TFRI weight calculation model | |
CN113281229A (en) | Multi-model self-adaptive atmosphere PM based on small samples2.5Concentration prediction method | |
CN109000924B (en) | Method for monitoring state of ball screw pair based on K mean value | |
CN115048873B (en) | Residual service life prediction system for aircraft engine | |
CN110675382A (en) | Aluminum electrolysis superheat degree identification method based on CNN-LapseLM | |
CN115935283A (en) | Drought cause tracing method based on multivariate nonlinear causal analysis |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190312 |
|
RJ01 | Rejection of invention patent application after publication |