CN111553489B - Aero-engine high-pressure rotor fault diagnosis method based on ensemble learning - Google Patents

Aero-engine high-pressure rotor fault diagnosis method based on ensemble learning Download PDF

Info

Publication number
CN111553489B
CN111553489B CN202010293645.4A CN202010293645A CN111553489B CN 111553489 B CN111553489 B CN 111553489B CN 202010293645 A CN202010293645 A CN 202010293645A CN 111553489 B CN111553489 B CN 111553489B
Authority
CN
China
Prior art keywords
support vector
vector machine
model
kernel function
machine model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202010293645.4A
Other languages
Chinese (zh)
Other versions
CN111553489A (en
Inventor
张硕
李隆齐
周鹤洋
隋天举
孙希明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Dalian University of Technology
Original Assignee
Dalian University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Dalian University of Technology filed Critical Dalian University of Technology
Priority to CN202010293645.4A priority Critical patent/CN111553489B/en
Publication of CN111553489A publication Critical patent/CN111553489A/en
Application granted granted Critical
Publication of CN111553489B publication Critical patent/CN111553489B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/24323—Tree-organised classifiers
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

The invention belongs to the technical field of fault diagnosis of aero-engines, and provides an aero-engine high-pressure rotor fault diagnosis method based on integrated learning, which can be used for predicting whether the aero-engine has a bearing vibration fault in the future flight time; firstly, selecting characteristics with higher importance degree in various data by using a Gini coefficient formula in random forest; secondly, constructing a support vector machine model by using the feature data with higher importance degree; and finally, quickly solving a regularization parameter C and a kernel function parameter sigma in the support vector machine model by utilizing a particle swarm optimization algorithm to integrally improve the prediction accuracy of the model. According to the method, on one hand, the characteristics with higher importance degree are effectively selected from the data set for modeling to avoid dimension disaster, and on the other hand, the regularization constant C and the kernel function parameter sigma of the support vector machine model are optimized through the particle swarm optimization to improve the prediction accuracy of the model.

Description

Aero-engine high-pressure rotor fault diagnosis method based on ensemble learning
Technical Field
The invention belongs to the technical field of fault diagnosis of an aircraft engine, and particularly relates to an integrated diagnosis learning method designed for a vibration fault of a bearing of the aircraft engine.
Background
An aircraft engine is an extremely complex and precise mechanical system, wherein whether any part works normally or not is related to the stability and safety of the whole system, and a large catastrophic accident is caused by the failure of a small part. Moreover, the research and development period is long, the cost is high, the crystallization of modern science and technology is integrated, and the research and development process relates to the tips of various fields such as technology, materials and computer science. As a power source for an aircraft, it also directly determines various performances and levels of the aircraft. In particular, the safety problem of the aircraft engine is a big problem which needs to be solved urgently at present. An effective method for solving the problem is mainly to provide a data driving model aiming at bearing vibration data of the aircraft engine, so that whether the bearing vibration fault occurs in the future of the aircraft engine is predicted, and corresponding measures are taken to ensure the safety of the aircraft. At present, the following methods are used for predicting the problem of the vibration fault of the bearing of the aircraft engine: 1) a diagnosis learning method based on BP neural network. The method mainly utilizes aeroengine vibration data as a data sample to establish an engine fault diagnosis model based on an artificial neural network. However, since the BP algorithm is essentially a gradient descent method, and the objective function to be optimized is very complex, the efficiency of the BP neural network is not high. In addition, from the mathematical point of view, the BP algorithm is used as an optimization method for local search, but the problem to be solved is to solve the global extremum of a complex nonlinear function, and the algorithm is likely to fall into the local extremum. It is difficult to accurately and quickly predict the vibration failure of the bearing of the aeroengine. 2) An analysis method based on a support vector machine. The method is characterized in that a support vector machine model is established by using the bearing vibration parameters of the aero-engine to monitor the state of the aero-engine. I.e. the monitoring of various parameters of the aeroengine bearing according to the established model. However, because the regularization parameter C and the kernel function parameter σ in the support vector machine model are unknown, it takes a lot of time to manually adjust the parameters when the model is constructed. In addition, if less important or irrelevant features are used to construct the support vector machine, the failure diagnosis rate of the support vector machine may be reduced.
In combination with the above discussion, aiming at the problems that the parameters of the support vector machine are unknown and all data characteristics are applied, the invention designs an integrated learning method to quickly and accurately predict the vibration fault of the bearing of the aircraft engine.
Disclosure of Invention
The invention provides an improved integrated learning method aiming at the problem of limitation of a support vector machine model in the diagnosis of the vibration fault of an aircraft engine bearing, and obtains better prediction accuracy. Since an aircraft engine is a highly complex pneumatic-thermal-mechanical system, there are probably hundreds of parameters measured by sensors, and generally when feature extraction is performed, more features are extracted as much as possible, but too many features cause redundancy, the correlation degree of partial features is too high, the computational performance is consumed, and problems such as overfitting or dimension disaster are easily caused. Therefore, how to select the characteristics with larger contribution from the characteristics to predict the bearing vibration fault of the aircraft engine is a challenging problem. In addition, the regularization parameter C and the kernel function parameter σ in the support vector machine model are unknown, and a lot of time is spent in manual adjustment, so an improved ensemble learning method is provided for the above problem.
The technical scheme of the invention is as follows:
an integrated learning method for vibration faults of bearings of an aircraft engine can predict whether the aircraft engine has vibration faults of the bearings in the future flight time; firstly, selecting characteristics with higher importance degree in various data by using a Gini coefficient formula in random forest; secondly, constructing a support vector machine model by using the feature data with higher importance degree; finally, a particle swarm optimization algorithm is used for rapidly solving a regularization parameter C and a kernel function parameter sigma in the support vector machine model to integrally improve the prediction accuracy of the model;
the method comprises the following specific steps:
step 1: aiming at the vibration data of the bearing of the aircraft engine, the data with larger importance degree in each class is solved by using a Gini coefficient method in random forest;
the random forest kini coefficient model is as follows:
Figure BDA0002451365420000021
Figure BDA0002451365420000022
wherein GI m Is the Keyny coefficient of node m, sigma represents the summation symbol, c is the total number of classes, P mk Representing the proportion of class k in node m,
Figure BDA0002451365420000023
is a characteristic X j Importance at node m, GI l And GI r Respectively represent the kini coefficients of two new nodes after branching,
Figure BDA0002451365420000024
is a characteristic X j Importance of the ith Tree, M being the utilization of feature X in the ith Tree j The set of nodes of the branch is set,
Figure BDA0002451365420000025
is a characteristic X j Of importance throughout a random forest, n is the number of trees, VIM, contained in the random forest j Is a characteristic X j The degree of importance of;
step 2: constructing a support vector machine by using main features selected by formula (2) in step 1
(1) The support vector machine model is as follows:
Figure BDA0002451365420000026
wherein ω ═ ω (ω ═ ω) 1 ；ω 2 ；…；ω d ) Is a normal vector, b is a displacement, y i And y j Is a normal label or a bearing vibration fault label defined by an aircraft engine,
Figure BDA0002451365420000031
represents x i The mapped feature vector s.t. is a constraint condition, C is a regularization constant, epsilon i Is a relaxation variable;
(2) the dual problem is obtained by using Lagrange multiplier method for the above formula, namely the problem is converted into the selection of proper kernel function K (x) i ,x j ) And appropriate regularization parameters C, constructing and solving an optimization problem
Figure BDA0002451365420000032
Figure BDA0002451365420000033
Obtaining an optimal solution:
Figure BDA0002451365420000034
wherein, a i And a j Is the ith and jth lagrange multipliers;
(3) selecting a * A positive component of
Figure BDA0002451365420000035
And calculates therefrom a threshold value:
Figure BDA0002451365420000036
(4) constructing a decision function taking RBF as a kernel function:
Figure BDA0002451365420000037
solving the problem of debugging by adopting a particle swarm optimization algorithm to obtain an optimal regularization constant C and a kernel function parameter sigma of a support vector machine model; the particle swarm optimization specifically optimizes the parameter model as follows:
v i+1 ＝h*v i +C 1 *rand(0，1)*(pbest i -x i )+C 2 *rand(0，1)*(gbest i -x i ) (7)
x i+1 ＝x i +v i
wherein v is i Is the velocity of the particle, x i Is the current position of the particle, rand (0,1) is a random number between (0,1), h is called the inertia factor, and its value is non-negative; the value is large, the global optimizing capability is strong, and the local optimizing capability is weak; its value is small, global optimizing ability is weak, local optimizingThe capability is strong; adjusting the global optimizing performance and the local optimizing performance by adjusting the size of h; c1 and C2 are referred to as learning factors, the former being the individual learning factor per particle and the latter being the group learning factor per particle, pbest i Is the best position of the individual, gbest i Is the best position found for all particles in the entire population.
The invention has the beneficial effects that: random forest and particle swarm algorithms are utilized. On one hand, the characteristics with higher importance degree are effectively selected from the data set for modeling to avoid dimension disaster, and on the other hand, the regularization constant C and the kernel function parameter sigma of the support vector machine model are optimized through the particle swarm optimization to improve the prediction accuracy of the model. Therefore, the integrated learning method can accurately predict whether the bearing vibration fault occurs in future flight of the aircraft engine.
Drawings
FIG. 1 shows the importance of class A data features.
FIG. 2 shows the importance of the B-class data characteristics.
FIG. 3 shows the importance of class C data features.
FIG. 4 shows the importance of class D data features.
FIG. 5 shows the importance of the class E data feature.
Fig. 6 is a flow chart of the solution of the regularization parameter C and the kernel function σ.
Detailed Description
Step 1: according to the characteristics of a vibration fault data set of an aircraft, the data set is divided into five categories of A, B, C, D and E, the feature importance degree of each category of data is respectively calculated by using a random forest-based coefficient formula (2) (as shown in figures 1-5), then the features with higher importance degree are extracted, namely 2 features are extracted from the A category of data, 1 feature is extracted from each of the B category of data and the C category of data, 8 features are extracted from the E category of data, and 12 features are extracted in total.
Step 2: and (4) constructing a support vector machine according to formulas (3) and (6) by using the 12 features extracted in the step 1.
And step 3: and (3) solving the regularization parameter C and the kernel function parameter sigma of the support vector machine constructed in the step (2) according to the formula (7) particle swarm optimization algorithm and the flow chart of fig. 6, so that the time for constructing the support vector machine model is shortened, and the accuracy of prediction is improved.
Results of the implementation
Model classes Accuracy of classification
First class model (support vector machine + particle swarm algorithm) 85.0439％
Second type model (support vector machine + particle swarm algorithm) 81.1024％
Third type model (support vector machine + particle swarm algorithm) 81.8898％
Fourth type model (support vector machine + particle swarm algorithm) 81.1024％
Fifth type model (support vector machine + particle swarm algorithm) 88.9764％
Integrated model (random forest + support vector machine + particle swarm algorithm) 95.2756％
TABLE 1
As can be seen from Table 1, although the classification accuracy of the first-class model and the fifth-class model reaches more than 85%, the prediction accuracy of the model constructed by the integrated learning method is obviously improved to more than 90%.
Therefore, the result shows that the integrated learning method effectively improves the traditional support vector machine diagnosis method and improves the prediction accuracy and the model training speed. Meanwhile, the integrated learning algorithm is proved to have better prediction and diagnosis capability on the bearing vibration fault problem of the aircraft engine.

Claims (1)

1. An aeroengine high-pressure rotor fault diagnosis method based on integrated learning can predict whether the aeroengine will have a bearing vibration fault problem in the future flight time; firstly, selecting characteristics with higher importance degree in various data by using a Gini coefficient formula in random forest; secondly, constructing a support vector machine model by using the feature data with higher importance degree; finally, a particle swarm optimization algorithm is used for rapidly solving a regularization parameter C and a kernel function parameter sigma in the support vector machine model to integrally improve the prediction accuracy of the model;
the method is characterized by comprising the following specific steps:
step 1: aiming at the vibration data of the bearing of the aircraft engine, the data with larger importance degree in each class is solved by using a Gini coefficient method in random forest;
the random forest kini coefficient model is as follows:
Figure FDA0002451365410000011
Figure FDA0002451365410000012
wherein, GI m Is the Keyny coefficient of node m, sigma represents the summation symbol, c is the total number of classes, P mk Representing the proportion of class k in node m,
Figure FDA0002451365410000013
is a characteristic X j Importance at node m, GI l And GI r Respectively represent the kini coefficients of two new nodes after branching,
Figure FDA0002451365410000014
is a characteristic X j Importance of the ith Tree, M being the utilization of feature X in the ith Tree j The set of nodes of the branch is set,
Figure FDA0002451365410000015
is a characteristic X j Of importance throughout a random forest, n is the number of trees, VIM, contained in the random forest j Is a characteristic X j The degree of importance of;
and 2, step: constructing a support vector machine by using main features selected by formula (2) in step 1
(1) The support vector machine model is as follows:
Figure FDA0002451365410000016
wherein ω ═ ω (ω ═ ω) 1 ；ω 2 ；…；ω d ) Is a normal vector, b is a displacement, y i And y j Are the normal label and the bearing vibration fault label defined by the aircraft engine,
Figure FDA0002451365410000017
represents x i The mapped feature vector s.t. is a constraint condition, C is a regularization constant, epsilon i Is a relaxation variable;
(2) the dual problem is obtained by using Lagrange multiplier method for the above formula, namely the problem is converted into the selection of proper kernel function K (x) i ,x j ) And appropriate regularization parameters C, constructing and solving an optimization problem
Figure FDA0002451365410000021
s.t.
Figure FDA0002451365410000022
0≤a i ≤C,i＝1,…,l
Obtaining an optimal solution:
Figure FDA0002451365410000023
wherein, a i And a j Is the ith and jth lagrangian multipliers;
(3) selecting a * A positive component of
Figure FDA0002451365410000024
And from this the threshold is calculated:
Figure FDA0002451365410000025
(4) constructing a decision function taking RBF as a kernel function:
Figure FDA0002451365410000026
wherein, sigma is a kernel function parameter, and sgn is a sign function;
and 3, step 3: aiming at the problem that the regularization constant C and the kernel function parameter sigma of the support vector machine model constructed in the step 2 are unknown and need manual debugging, solving the optimal regularization constant C and the kernel function parameter sigma of the support vector machine model by adopting a particle swarm optimization algorithm; the particle swarm optimization specifically optimizes the parameter model as follows:
v i+1 ＝h*v i +C 1 *rand(0，1)*(pbest i -x i )+C 2 *rand(0，1)*(gbest i -x i )(7)
x i+1 ＝x i +v i
wherein v is i Is the velocity of the ith particle, x i Is the ith particle current position, rand (0,1) is a random number between (0,1), h is called an inertia factor, and its value is non-negative; the value is large, the global optimizing capability is strong, and the local optimizing capability is weak; the value is small, the global optimizing capability is weak, and the local optimizing capability is strong; adjusting the global optimizing performance and the local optimizing performance by adjusting the size of h; c1 and C2 are known as learning factors, the former being the individual learning factor per particle and the latter being the population learning factor per particle, pbest i Is the best position of the individual, gbest i Is the best location found for all particles in the entire population.
CN202010293645.4A 2020-04-15 2020-04-15 Aero-engine high-pressure rotor fault diagnosis method based on ensemble learning Active CN111553489B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010293645.4A CN111553489B (en) 2020-04-15 2020-04-15 Aero-engine high-pressure rotor fault diagnosis method based on ensemble learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010293645.4A CN111553489B (en) 2020-04-15 2020-04-15 Aero-engine high-pressure rotor fault diagnosis method based on ensemble learning

Publications (2)

Publication Number Publication Date
CN111553489A CN111553489A (en) 2020-08-18
CN111553489B true CN111553489B (en) 2022-09-27

Family

ID=72004325

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010293645.4A Active CN111553489B (en) 2020-04-15 2020-04-15 Aero-engine high-pressure rotor fault diagnosis method based on ensemble learning

Country Status (1)

Country Link
CN (1) CN111553489B (en)

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106295153B (en) * 2016-08-03 2018-12-28 南京航空航天大学 A kind of Fault Diagnosis of Aircraft Engine Gas Path method based on twin support vector machines
CN106503746B (en) * 2016-11-03 2018-07-06 哈尔滨工业大学 A kind of Fault Diagnosis of Aeroengines method based on offset of performance amount

Also Published As

Publication number Publication date
CN111553489A (en) 2020-08-18

Similar Documents

Publication Publication Date Title
Li et al. A directed acyclic graph network combined with CNN and LSTM for remaining useful life prediction
CN109766583B (en) Aircraft engine life prediction method based on unlabeled, unbalanced and initial value uncertain data
Long et al. Improved diagnostics for the incipient faults in analog circuits using LSSVM based on PSO algorithm with Mahalanobis distance
CN108959778B (en) Method for predicting residual life of aircraft engine based on consistency of degradation modes
WO2016107246A1 (en) Wavelet noise reduction and relevance vector machine-based method for predicting remaining life of lithium battery
CN112508442B (en) Transient stability assessment method and system based on automatic and interpretable machine learning
CN109242215B (en) Particle swarm-support vector machine-based rotating mechanical equipment operation condition prediction method
CN110987439B (en) Aeroengine fault prediction method based on Logitics regression and Xgboost model
CN111680875A (en) Unmanned aerial vehicle state risk fuzzy comprehensive evaluation method based on probability baseline model
Li et al. Domain adaptation remaining useful life prediction method based on AdaBN-DCNN
Zhao et al. Probabilistic remaining useful life prediction based on deep convolutional neural network
CN113177357B (en) Transient stability assessment method for power system
CN113569462A (en) Distribution network fault level prediction method and system considering weather factors
Che et al. Multi-head self-attention bidirectional gated recurrent unit for end-to-end remaining useful life prediction of mechanical equipment
Li et al. Aero-engine exhaust gas temperature prediction based on LightGBM optimized by improved bat algorithm
CN115392286A (en) Bearing fault diagnosis method based on 1dCNN-GWO-SVM model
CN111695288A (en) Transformer fault diagnosis method based on Apriori-BP algorithm
CN111737918B (en) Integrated learning method for fault diagnosis of high-pressure rotor of aircraft engine
CN117674119A (en) Power grid operation risk assessment method, device, computer equipment and storage medium
CN111553489B (en) Aero-engine high-pressure rotor fault diagnosis method based on ensemble learning
CN116928038A (en) Fault early warning method and device based on main bearing temperature of offshore wind turbine generator
CN117113086A (en) Energy storage unit load prediction method, system, electronic equipment and medium
Caricato et al. Prognostic techniques for aeroengine health assessment and Remaining Useful Life estimation
Si et al. Application of a Taylor expansion criterion-based pruning convolutional network for bearing intelligent diagnosis
Dutta et al. Early & accurate forecasting of mid term wind energy based on PCA empowered supervised regression model

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