An original pattern recognition approach for the diagnosis of switch mechanisms driven by an electric motor is presented in this paper. Its main advantage is that it does not require a physical model of the system and can easily be adapted to other complex systems. The available data for this task are the signals of the electrical power consumption during the switch actuation period and the proposed method consists of two steps: the feature extraction from the signals and the recognition of different operating states (class without defect, class with minor defect and class with critical defect) using mixture discriminant analysis (MDA). This method assumes the classes to be represented by a Gaussian mixture distribution whose parameters are estimated by the maximum likelihood method, using the expectation-maximization (EM) algorithm. An experimental study performed on real measured signals covering a wide range of defects reveals some good performances of the proposed approach compared to others classification methods such as K-Nearest-Neighbors, Neural Networks and the classical Bayesian discriminant approach (with one Gaussian distribution per class).
[1]
Christopher M. Bishop,et al.
Neural Network for Pattern Recognition
,
1995
.
[2]
G. Schwarz.
Estimating the Dimension of a Model
,
1978
.
[3]
Shigeo Abe DrEng.
Pattern Classification
,
2001,
Springer London.
[4]
R. Tibshirani,et al.
Discriminant Analysis by Gaussian Mixtures
,
1996
.
[5]
Toshio Odanaka,et al.
ADAPTIVE CONTROL PROCESSES
,
1990
.
[6]
R. Bellman,et al.
V. Adaptive Control Processes
,
1964
.
[7]
David G. Stork,et al.
Pattern Classification (2nd ed.)
,
1999
.
[8]
Belur V. Dasarathy,et al.
Nearest neighbor (NN) norms: NN pattern classification techniques
,
1991
.
[9]
Sanjoy Dasgupta,et al.
Adaptive Control Processes
,
2010,
Encyclopedia of Machine Learning and Data Mining.