Towards Unified Probabilistic Rotorcraft Damage Detection and Quantification via Non-parametric Time Series and Gaussian Process Models

The complex dynamics of rotorcraft structures under varying operational and environmental conditions demand the development of accurate and robust-to-uncertainties structural health monitoring (SHM) approaches. The inherent uncertainty within monitoring data makes it difficult for conventional methods to accurately and robustly detect and quantify damage without the need for a large number of data sets. In addition, due to the time-varying nature of rotorcraft operations, such conventional metrics might still fail even with abundance of data. In this paper, we propose a unified probabilistic damage detection and quantification framework for active-sensing, guided-wave SHM that focuses on monitoring rotorcraft structural "hotspots". The proposed framework involves three stages: The first stage incorporates statistical damage detection based on stochastic non-parametric time series (NP-TS) models of ultrasonic wave propagation signals within a hotspot sensor network configuration. The second stage involves the statistical path selection, where a NP-TS representation is used for the sole purpose of identifying damage-intersecting signal (wave propagation) paths, that is the paths that are most sensitive to damage, in order to use them in the subsequent damage quantification stage. That last stage achieves probabilistic damage quantification, where the results of the NP-TS models are used to train Bayesian Gaussian Process regression and classification models. This unified framework ensures accurate and robust damage detection and quantification in a data-efficient manner since only damage-intersecting paths are selected and used in the analysis. The performance of the proposed framework is compared to that of conventional state-of-the-art damage indices (DIs) in detecting and quantifying simulated damage in two representative coupons: a Carbon Fiber Reinforced Polymer (CFRP) coupon and a stiffened aluminum (Al) panel. It is shown that the proposed framework outperforms conventional DI-based active-sensing guided-wave SHM methods.