A support vector machine approach for classification of welding defects from ultrasonic signals

Abstract

Defect classification is an important issue in ultrasonic non-destructive evaluation. A layered multi-class support vector machine (LMSVM) classification system, which combines multiple SVM classifiers through a layered architecture, is proposed in this paper. The proposed LMSVM classification system is applied to the classification of welding defects from ultrasonic test signals. The measured ultrasonic defect echo signals are first decomposed into wavelet coefficients by the wavelet packet transform. The energy of the wavelet coefficients at different frequency channels are used to construct the feature vectors. The bees algorithm (BA) is then used for feature selection and SVM parameter optimisation for the LMSVM classification system. The BA-based feature selection optimises the energy feature vectors. The optimised feature vectors are input to the LMSVM classification system for training and testing. Experimental results of classifying welding defects demonstrate that the proposed technique is highly robust, precise and reliable for ultrasonic defect classification.

Keywords::

• support vector machine
• bees algorithm
• wavelet packet transform
• defect classification
• ultrasonic non-destructive evaluation

Additional information

Funding

Funding

This work is supported by the National Natural Science Funds [grant number 51074121], the Scientific Research Program Funded by Shaanxi Provincial Education Department of China [grant number 11JK0776] and the Startup Funds for Doctors of Xi'an University of Science and Technology of China [grant number 2010QDJ026].