In the field of structural health monitoring, researchers focus on the design of systems and techniques capable of detecting damage in structures. However, most traditional detection methods fail under environmental and operational variations that tend to distort the signals and masquerade as damage. In this paper, we investigate the applications of machine learning techniques to developing a damage detection system robust to changes in the internal air pressure of a pipe. From each of the 240 experimental datasets, we extract 167 features and implement three classification algorithms for detecting damage: adaptive boosting, support vector machines, and a method combining the two. The performances of the three classifiers are evaluated over 30 detection trials with different combinations of training and testing data, resulting in the average accuracies of 87.7%, 92.5% and 93.5%, respectively. The combined method is a promising classifier for damage detection. Through feature selection, we also demonstrate the effectiveness of features related to the curve length, the shift-invariant correlation coefficient and the peak amplitude of the signal.
[1]
Yoav Freund,et al.
A decision-theoretic generalization of on-line learning and an application to boosting
,
1997,
EuroCOLT.
[2]
LinLin Shen,et al.
Gabor Feature Selection for Face Recognition Using Improved AdaBoost Learning
,
2005,
IWBRS.
[3]
Liana G. Apostolova,et al.
Comparison of AdaBoost and Support Vector Machines for Detecting Alzheimer's Disease Through Automated Hippocampal Segmentation
,
2010,
IEEE Transactions on Medical Imaging.
[4]
J. Michaels,et al.
Feature Extraction and Sensor Fusion for Ultrasonic Structural Health Monitoring Under Changing Environmental Conditions
,
2009,
IEEE Sensors Journal.
[5]
José M. F. Moura,et al.
Time reversal for damage detection in pipes
,
2010,
Smart Structures and Materials + Nondestructive Evaluation and Health Monitoring.