Wavelet-based outlier analysis for guided wave structural monitoring: application to multi-wire strands

This work describes a method based on Outlier Analysis and the Wavelet Transform for structural damage detection based on guided ultrasonic waves. The basic idea is to de-noise and compress the ultrasonic signals by the Discrete Wavelet Transform and use the relevant wavelet coefficients to construct a uni-dimensional or multi-dimensional damage index. The damage index is then fed to an outlier analysis to detect anomalies that are representative of structural defects. By extracting the essential information from the ultrasonic signals, the dimension of the damage index is kept at a minimum value, as desirable for continuous structural monitoring. The general framework is applied to the detection of notch-like defects in a seven-wire strand by using built-in magnetostrictive devices for ultrasound transduction. Random noise is digitally added to the raw ultrasonic measurements to create statistical populations of the baseline (undamaged) conditions and the damaged conditions. The object of the present study is to apply novelty detection to Guided Ultrasonic Waves (GUWs) analyzed by the Discrete Wavelet Transform (DWT) as a basis to construct the damagesensitive features. The purpose of the DWT in the present work is to de-noise and compress GUW measurements taken from structural components with finite cross-sectional dimensions, as opposed to vibration measurements representative of the global behavior of the structure. Recent works used the DWT coefficients of strain time series to formulate a vector dynamic regression model for the identification of outlier events in a bridge. This study extends these previous works by incorporating the novelty detection as the unsupervised learning method of damage detection and sizing. The method is applied to the structural monitoring of seven-wire strands used in prestressed concrete structures and cable-stayed or suspension bridges.