Original Article
Proc IMechE Part I:
J Systems and Control Engineering
2015, Vol. 229(10) 917–926
Ó IMechE 2015
Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/0959651815601276
pii.sagepub.com
Fault diagnosis for the landing phase of
the aircraft based on an adaptive
kernel principal component analysis
algorithm
Runxia Guo, Kai Guo and Jiankang Dong
Abstract
Kernel principal component analysis is an effective fault diagnosis algorithm proved by large amount of practices in indus-
try; however, kernel principal component analysis with constant parameters is unable to achieve satisfactory results
when the working condition is dynamic. In this article, the parameters in the proposed kernel function are adaptively
adjusted according to the maximum variance principle and the K-nearest neighbors approach. Because the variance of
the data is maximized, the proposed adaptive kernel principal component analysis approach can obtain well diagnosis
effect under the condition that only a small amount of data are available. Moreover, for the structure of the data is pre-
served by the K-nearest neighbors method, the rate of false alarm and underreporting is effectively reduced. In addition,
a novel approach with a new algorithm for forgetting factor calculation is put forward so that the fault detection model
can be updated and suits the actual situation better. Experiments on the landing phase of the aircraft have shown that
adopting the proposed algorithm can detect faults better than using static model.
Keywords
Fault diagnosis, adaptive kernel principal component analysis, forgetting factor, maximum variance, K-nearest neighbors
method, landing phase of the aircraft
Date received: 13 April 2015; accepted: 24 July 2015
Introduction
In recent years, there has been an increasing demand on
production quality and system performance; therefore,
industrial systems are more complicated. Reliability
and safety have become the most critical issues for
modern industrial systems and are receiving more and
more attention. As system faults cannot be entirely pre-
vented, timely fault detection and diagnosis become sig-
nificant. Extensive and comprehensive researches at
universities and in industry field have been dedicated to
fault diagnosis.
Generally, there are mainly two types of research
approaches for fault diagnosis: one is based on analyti-
cal models and the other is the data-based method. The
model-based methods have their advantages on fault
location, isolation and analysis.
1,2
However, because
the actual industrial system is generally complicated
and cannot be modeled precisely, the model-based
approach has significant limitation when adopted for
fault diagnosis. In contrast, the data-based technique
can extract necessary information from huge amounts
of recorded process data without modeling, which is
then used for real-time fault detection and diagnosis.
Without the efforts for modeling sophisticated sys-
tem, data-driven approaches can be easily applied and
satisfactory results can be obtained. The principal com-
ponent analysis (PCA) approach and the partial least
squares (PLS) method are the basic multivariate statis-
tical approaches which have been widely used.
3–5
PCA
could preserve the significant variability information
extracted from the process data and reduce the dimen-
sion of the data. Because of its simple form and its abil-
ity to handle large amount of data, PCA has been
widely and successfully applied in many areas, such as
College of Aeronautical Automation, Civil Aviation University of China,
Tianjin, China
Corresponding author:
Runxia Guo, College of Aeronautical Automation, Civil Aviation
University of China, Jin North Road No. 2898, Dongli District, Tianjin
300300, China.
Email: rxguoblp@163.com
at Civil Aviation Uni of China on December 29, 2015pii.sagepub.comDownloaded from