Evaluation of Independent Component Analysis for Lamb-wave Crack Detection in Metallic Plates

Abstract

Guided ultrasound waves represent an auspicious approach for usage in future intelligent sensor systems tackling the task of automated detection of defects in critical thinwalled mechanical structures. In these structural health monitoring (SHM) applications, permanently attached active piezoceramic transducers enable repeated non-destructive interrogations, which result in detailed time-dependent multivariate information about the examined structure. In the present work, a novel statistical signal processing scheme based on independent component analysis (ICA) is presented as well as evaluated concerning the task of defect detection. ICA is capable of separating signal components caused by defects from coherent noise. ICA therefore enables subsequent clustering of signals into the damaged and undamaged cases. The evaluation is carried out on numerically generated test data stemming from metallic sheets, which optionally exhibit crack-type defects. For the computer simulation of ultrasonic Lamb-wave propagation, a simple numerical method is introduced which permits the rapid generation of a diverse data set. Results of the evaluation are discussed and an outlook is finally given.