Weld defect classification using 1-D LBP feature extraction of ultrasonic signals

A method based on the one-dimensional local binary pattern (1-D LBP) algorithm to extract features of ultrasonic defect signals and perform multi-class defect classification was proposed. The ultrasonic defect echo signals were first decomposed into wavelet coefficients by the wavelet packet decomposition. The 1-D LBP algorithm was employed to extract LBP features of components at low and high frequencies, respectively. Subsequently, these LBP statistical feature sets were regarded as feature vectors of defect classification. Weld defects were then classified automatically by using the radial basis function support vector machine. Defects of slag inclusion, porosity and incomplete penetration in a steel plate butt weld were used for experiments and feature extraction and defect classification were performed. The results show that the class separability of 1-D LBP features used for defect classification is superior to that of the traditional features. Moreover, the accuracy of defect classification reached 98.3%, providing an efficient tool for ultrasonic defect classification.