| The development of real-world structural health management (SHM) systems requires the design and implementation of robust damage estimation, diagnosis, and prognosis methodologies. In the last few decades, research efforts have resulted in data-driven methods that depend only on sensor-based measurements, as well as physics-based methods that depend only on structural physical properties. Although the process of damage nucleation and evolution is stochastic, and uncertainty is prevalent in any practical scenario due to possible variability in material microstructure, loading, and environmental conditions, physics-based models do not account for uncertainty in structural parameters. As a result, a reliable progressive damage estimation strategy should rely on the combination of physics-based models and information collected real-time from sensor measurements.;In this dissertation, a novel approach is developed that uses a probabilistic state-space setting to formulate and solve the progressive damage estimation problem in an optimal Bayesian framework. The damage evolution state model is described using physics-based models. On the other hand, the measurement models are obtained using hidden Markov models that depend on time-frequency features extracted from structural sensor measurements. The damage state estimation is performed efficiently using sequential Monte Carlo techniques. The utility of the proposed methods is demonstrated by their application to estimate progressive fatigue damage in a compact tension sample.;The performance of the progressive damage estimation methods is further improved by proposing optimal sensor scheduling algorithms. In particular, when multiple sensors are employed for measurements, the sensors are adaptively configured by minimizing the predicted estimation error. The performance of the adaptive sensor scheduling algorithm is demonstrated by estimating the progressive fatigue damage in a notched laminate and a lug joint sample.;When the sensor measurements are very noisy, the probability of detection is low. It is thus important to suppress noise in real-world SHM systems for robust estimation. Novel noise suppression methods are proposed based on probabilistic measurement screening techniques in order to increase damage estimation performance. The developed approaches are demonstrated by comparing estimation results with and without noise suppression. |
