State Space Feature Extraction Applied to Diffuse Ultrasonic Signals Using Simulated Chaotic Excitations

Abstract

Diffuse ultrasonic waves for structural health monitoring offer the advantages of simplicity of signal generation and reception, sensitivity to damage, and large area coverage. However, one of the difficulties associated with these complex signals is extraction of robust features that can be related to progression of damage. This paper investigates feature extraction techniques in embedded state space, which is an approach that has been proposed for vibration‐based structural health monitoring. A continuous chaotic signal is typically used to excite the specimen to obtain a non‐periodic steady state response in the time domain. However, diffuse waves are naturally transient and are usually formed using an impulsive or burst excitation. In this paper, a computer‐generated continuous chaotic signal is convolved with the measured transient diffuse signal to simulate the response to a chaotic excitation. This convolved signal is embedded into state space and various features are extracted from the resulting phase portrait. This convolution method is applied to data from two experiments on aluminum plates in which artificial flaws are introduced and incrementally enlarged. Selected state space features show the ability to track increasing flaw growth.