Support vector machine classification of ultrasonic shaft inspection data using discrete wavelet transform

Abstract

While many non-destructive ultrasonic test scenarios involve shallow surfaces, but when signals for testing come from long shafts a major problem of mode-converted reflection emerges. These reflections are echoes that do not correspond to cracks in the material, neither to characteristics in the shaft. Moreover, the length of the signals demands the application of efficient feature extraction mechanisms to reduce the dimension of pattern vectors for feasible automated classification. Experimental evidence has shown that the Discrete Wavelet Transform (DWT) provides faster and more reliable extraction for Artificial Neural Network (ANN) in these long signals. This paper demonstrates that DWT is not only highlighting more adequate vectors for ANN but it is more general and in particular it also improves the performance of Support Vector Machines (SVM).