Scribd
Upload
4 views2 pages

Engine Testing Fault Classification Based On The Multi-Class SVM of Auto-Regression

This paper presents a classification method for diagnosing vibration faults in aero-engines using a time series autoregressive model and a multi-class support vector machine (SVM). The proposed method effectively identifies different fault modes with high accuracy, even with limited sample sizes, by utilizing autoregressive parameters as feature vectors. Experimental results demonstrate the method's adaptability and computational efficiency in classifying vibration faults during engine testing.

Uploaded by

Isabelle Simão
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
4 views2 pages

Engine Testing Fault Classification Based On The Multi-Class SVM of Auto-Regression

This paper presents a classification method for diagnosing vibration faults in aero-engines using a time series autoregressive model and a multi-class support vector machine (SVM). The proposed method effectively identifies different fault modes with high accuracy, even with limited sample sizes, by utilizing autoregressive parameters as feature vectors. Experimental results demonstrate the method's adaptability and computational efficiency in classifying vibration faults during engine testing.

Uploaded by

Isabelle Simão
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Engine Testing Fault Classification Based

on the Multi-class SVM of Auto-regression

Xiangyang Jin, Shisheng Zhong, Gang Ding, and Lin Lin

School of Mechatronics, Harbin Institute of Technology, 150001 Harbin, China


[email protected]

Abstract. In connection with the problem of lack of an effective method for


vibration fault diagnose of aero-engine during the testing process in test cell,
this paper proposed a classification method for aero-engine’s different types of
fault modes based on a time series autoregressive (AR) model and support
vector machine (SVM) classifier. First, respectively collect 200 kinds of
vibration signals in normal state and three kinds of fault state from the engine
test cell. Establish AR model for the training set of aero-engine vibration
signals through the autocorrelation algorithm, then obtained by the feature
vectors consist of autoregressive parameters and residual variance. Then create
SVM classifier, the obtained vibration signal feature vector will be entered into
the SVM classifier, adjusting the penalty parameter c and the kernel function
parameter g through the optimization algorithm, the ideal forecasting
classification model is available. Finally, conduct classification identification
on fault types of different test sets through the obtained classification models.
Experimental results verified this method was effective to engine’s
classification for different vibration fault modes under the conditions of small
samples and had high classification accuracy.

Keywords: AR model, multi-class SVM, test fault, model classification.

1 Introduction

The nature of fault diagnosis is based on a set of actual characteristic quantity


reflecting the system’s working state, according to a certain criteria, the system is
divided into a fault mode. The commonly used methods for intelligent diagnosis are
fault tree, expert systems, neural network, fuzzy theory and so on. Fault tree analysis
has a heavy workload in handling faults, the subordinate and dependent faults are
difficult to handle, and it is easy to neglect or miss some useful information, for
events with a variety of states, it is difficult to handle; The expert system approach
has difficulties in knowledge acquisition, with a weak reasoning ability, it is either of
no memory function or robustness; Neural network method requires large amount of
data for fault samples, but of a weak generalization ability; Fuzzy diagnosis method
has strong subjective factors, such as in the improper selection, the diagnostic
accuracy will drop, thus it is rarely used alone[1].

D. Zeng (Ed.): Advances in Information Technology and Industry Applications, LNEE 136, pp. 33–38.
springerlink.com © Springer-Verlag Berlin Heidelberg 2012
34 X. Jin et al.

The SVM can achieve good classification results in the case of very few training
samples [2], it has a strong anti-interference ability and generalization, and it can
demonstrate its unique advantages in solving the small sample classification
problems. AR model is a most basic time series model with a mostly widely practical
application, autoregressive parameters in this model are sensitive to the state change
law, all the information of equipment’s dynamic characteristics and work state are
contained in model parameters [3], it is very effective to take the autoregressive
parameters of AR model as the feature vectors when analyzing the system’s state.
This paper proposed a diagnosis method for vibration fault mode based on AR
model and SVM classifier, experiments showed that this method had a good
adaptability to aero-engine’s classification for different vibration fault modes and had
a higher classification accuracy and computational efficiency.

2 Fault Classification Algorithms


This paper adopted the Least Square-Support Vector Machine (Least Square-Support
vector Machine, LS-SVM) algorithm to find the classification model of fault
diagnosis [4], the training was completed by the following formula:
1 C M
min W T W + ∑ ξi2
2 2 i =1 (1)

The constrained condition is


yi (W T g ( xi ) + b) = 1 − ε i (2)

In this formula, i = 1," , M；y = {+1, −1} ; W is the weight vector; g ( x ) is the
function that ϕi maps X from the input space to the feature space; ξ is the
relaxation coefficient of xi ; C is the boundary coefficient. Introducing the Lagrange
multiplier α i , change the above equation into an unconstrained objective function:

1 C M M
Q(W , b, α , ξ ) = W T W + ∑ ξi2 − ∑ α i [ yi (W T g ( xi ) + b) − 1 + ξi ] (3)
2 2 i =1 i =1

Solve the matrix equation (3) can be obtained by α and b, that is:
−1 −1 −1
α = Ω (1 − yb)
−1
b = (Y Ω y ) Y Ω I .
T T

The Least Square-Support Vector Machine can change the inequality constraints
into equality constraints, the training process is converted from solving the quadratic
programming problem into solving linear equations, which simplifying the
computational complexity.

You might also like