Statistical Active-Sensing Structural Health Monitoring via Stochastic Time-Varying Time Series Models.

In the context of acousto-ultrasound guided wave-based damage diagnosis, the vast majority of existing methods are deterministic in nature and face significant challenges when exposed to real-life situations, potentially varying environmental and operating conditions, and stochastic time-varying structural response and uncertainty. These factors limit the applicability and widespread adoption of structural health monitoring (SHM) methods for aerospace, mechanical, and civil engineering systems. Thus, there lies a need for accurate and robust damage diagnosis methods for assessing the structural health under uncertainty. In this work, a novel statistical method for structural damage detection and identification, collectively referred to as damage diagnosis, via ultrasonic guided waves is postulated using stochastic time-varying time series models. Ultrasonic guided waves, that are dispersive in nature, are represented via recursive maximum likelihood time-varying autoregressive (RML-TAR) and functional series time-varying autoregressive (FS-TAR) models. Next, the estimated time-varying model parameters are employed within a statistical decision-making framework to tackle damage detection and identification under predetermined type I error probability levels. Both damage intersecting and non-intersecting paths are considered in a multi-sensor aluminum plate in pitch-catch configuration for the complete experimental assessment. The detailed damage diagnosis results are presented and the method's robustness, effectiveness, and limitations are discussed.