The aerospace industry increasingly employs composite materials due to their superior strength-to-weight ratios and corrosion resistance. However, the heterogeneous and anisotropic nature of composites introduces complex failure mechanisms such as delamination, matrix cracking and fiber breakage, which make modeling of degradation processes challenging. In addition, variability in material quality due to manufacturing processes, along with operational uncertainties, further challenges the reliable prediction of remaining useful life of life (RUL) of composites. As a result uncertainty management is essential for enabling informed decision-making in prognostics and health management (PHM).

This thesis presents a novel uncertainty management approach specifically targeting past state uncertainties that stem from variability in composite manufacturing processes, such as embedded defects and material quality inconsistencies. By leveraging advanced ultrasonic imaging through Dolphicam technology, internal structural variations in aerospace-grade carbon fiber-reinforced polymer (CFRP) composites were quantified. Subsequently, a similarity-informed methodology was developed, employing both spatial pyramid histogram (SPH) and convolutional neural network (CNN)-based embedding techniques, to group specimens based on internal structural quality. This approach assumes that structurally similar composites exhibit analogous degradation behavior under fatigue loading, enabling targeted and more reliable prognostic modeling.

Experimental validation involved fabricating CFRP laminates with intentionally embedded defects, followed by detailed nondestructive inspections (NDI) and fatigue testing with digital image correlation (DIC)-based strain measurements. The similarity-informed prognostic model demonstrated significant improvements in prediction accuracy and uncertainty quantification compared to conventional methods, confirming the viability and advantages of integrating similarity-based past state uncertainty management in aerospace composite structures to improve reliability in prognostics.

Overall, the results underscore the value of high-fidelity NDI for material quality characterization and highlight the potential of similarity learning techniques to reduce prognostic uncertainty. By enhancing the credibility of RUL predictions, this work contributes to more reliable decision-making capabilities and supports the advancement of predictive maintenance strategies for aerospace composite structures.