Feature extraction for defect detection in strands by guided ultrasonic waves

Multi-wire strands are widely used in civil structures as tensioning members in prestressed concrete, cable-stayed and suspension bridges. In recent years, various researchers including the authors have investigated guided ultrasonic waves as a tool for monitoring the health of the strands. Magnetostrictive ultrasonic transducers have been successfully used to excite and detect the guided waves in these components. This study improves the general guided-wave technique with magnetostrictive transducers for the detection and sizing of defects in strands. The improvement consists of extracting sensitive and robust features of the wave signals, and then using these features to construct a damage index (D.I.). The proposed D.I. is unaffected by accidental variations in the excitation power or by changes in the electromechanical coupling efficiency of the transducers. The specific defect under investigation is a notch cut at varying depths in a strand that is subjected to a typical operational load. Features extracted after discrete wavelet processing of the wave signals result in a D.I. that is robust against noise and is linearly related to the notch depth in a logarithmic scale. In particular, the variance of the signal's wavelet coefficients results in the largest sensitivity to notch depth.