Improving Damage Localization Through Gaussian Process Regression and Enhanced Acoustic Emissions Time of Arrival Estimation

Accurate localization of structural flaws through time of arrival (ToA) estimation is fundamental for structural health monitoring (SHM). However, current methods suffer from inaccuracies in noisy environments, particularly when using standard statistical approaches like the Akaike information criterion (AIC). To overcome these limitations, we propose a new approach built on an adaptive picking technique (APT) combined with gaussian process (GP) regression. APT employs frequency-based mode separation and adaptive envelope analysis to reliably identify first arrivals without prior modal knowledge, while GP regression provides probabilistic damage localization with confidence intervals. Experimental validation on an aluminum plate instrumented with six sensors demonstrates superior performance across multiple conditions, achieving a median error of 13.7 mm with 4.9× improvement over AIC-only alternatives. Under noise conditions down to 18 dB), representative of operational scenarios, our approach not only maintains significant robustness, improving up to 8.2× with respect to AIC, but also generalizes across training and testing conditions, confirming its reliability for real-world SHM applications.