Development of SVM Model for Reliability Analysis of Railway System

grizzlybearcroatianAI and Robotics

Oct 16, 2013 (3 years and 9 months ago)

81 views



Researcher
Yuan Fuqing
Tel: +46 920 49 16 82
E-mail: yuafuq@ltu.se

www.ltu.se/shb/2.2048

Supervisor
Professor Uday Kumar
Tel: +46 920 49 18 26
E-mail: uday.kumar@ltu.se

www.ltu.se/shb/2.2048


Development of SVM Model for Reliability
Analysis of Railway System
Research Period: 2010 - 2011




Introduction
This research aims to develop Support Vector
Machine (SVM) models to estimate and
predict railway system reliability. The SVM
method has been found to be more accurate
and powerful in identifying the trends and
deviations in data sets. During the project,
mathematical algorithm will be developed to
identify deviation from normal operation
indicating on coming failures. This method
can also help identifying any abnormal
operation state thereby providing warning for
impending failures. During the research, SVM
models will be combined with classical
statistical techniques such as Bayesian
Inference, ANOVA, etc. The project is
financed by Banverket.


Objective
The aim of the research is to develop and
demonstrate the applicability of SVM models to
identify abnormal state and identify the onset of
failures in railway infrastructure.
Deliverables
The outcome of the research will comprise several
efficient Support Vector Machine models and their
mathematical algorithms.

Some publications
1. Fuqing Y., Kumar U., Claudio M. Rocco S. and Misra K.
B. Complex System Reliability Evaluation using Support
Vector Machine.
2. Fuqing Y., Kumar U., Misra K. B. Reliability Evaluation
Using Support Vector Machine for Incomplete Dataset.



















Sponsor
Banverket
Tube
Sample
Data
Support
Vectors
SVR Model
0.2
0.4
0.6
0.8
1 2
3
4 5
0
-2
2
4
0
4
6
8
10
12
0
2
-2
Hyperplane
Grouup1
Grouup2
20
40
60
80
1
3
5
7
Neural Network
NHPP
SVM
Real Data
T
N
Predict Next Failure
cum
Failure
0.6444
No.Reliability (Real) SVM NF GRNN MLP(Gaussian) RBF
36
37
38
39
40
0.6345
0.6245
0.6145
0.6046
NF: Neural Fuzzy Network
GRNN:Generalized regression neuralwork
MLP: Multilayer perceptron neural network
RBF: Radial basis function neural network
ARIMA: Autoregressive integrated moving average
MAPE: Mean absolute percentage error.
NRMSE: Normalized root mean square error.
MAPE(%)
NRMSE
0.6446
0.6346
0.6248
0.6148
0.6049
0.6485 0.6533 0.6515 0.6466
0.6344
0.6236
0.6163
0.6070
0.6389
0.6446 0.6369
0.6270 0.6383
0.6270
0.6049
0.6328 0.6170
0.5989 0.6278 0.6072
0.0387 0.2972 0.9960 2.3437
0.3914
0.0004
0.00369 0.01085 0.02497 0.00391
Perfomance Comparision among various techniques
(Adopted from Journal of Reliability
engineering and system safety)
Support vector regressor and its performance