9 March 2014 Lamb wave based damage detection using Matching Pursuit and Support Vector Machine classifier
Author Affiliations +
Abstract
In this paper, the suitability of using Matching Pursuit (MP) and Support Vector Machine (SVM) for damage detection using Lamb wave response of thin aluminium plate is explored. Lamb wave response of thin aluminium plate with or without damage is simulated using finite element. Simulations are carried out at different frequencies for various kinds of damage. The procedure is divided into two parts - signal processing and machine learning. Firstly, MP is used for denoising and to maintain the sparsity of the dataset. In this study, MP is extended by using a combination of time-frequency functions as the dictionary and is deployed in two stages. Selection of a particular type of atoms lead to extraction of important features while maintaining the sparsity of the waveform. The resultant waveform is then passed as input data for SVM classifier. SVM is used to detect the location of the potential damage from the reduced data. The study demonstrates that SVM is a robust classifier in presence of noise and more efficient as compared to Artificial Neural Network (ANN). Out-of-sample data is used for the validation of the trained and tested classifier. Trained classifiers are found successful in detection of the damage with more than 95% detection rate.
© (2014) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Sushant Agarwal, Mira Mitra, "Lamb wave based damage detection using Matching Pursuit and Support Vector Machine classifier", Proc. SPIE 9064, Health Monitoring of Structural and Biological Systems 2014, 906424 (9 March 2014); doi: 10.1117/12.2044022; https://doi.org/10.1117/12.2044022
PROCEEDINGS
11 PAGES


SHARE
Back to Top