Probabilistic system-level prognostics in aerospace applications

Abstract

System-Level Prognostics (SLP) is essential for mission success, as it aims to predict the Remaining Useful Life (RUL) of an entire component rather than of its smaller structural subsystems, referred to here as coupons. Unlike coupon-level prognostics, SLP must capture degradation interactions among coupons while quantifying inherent uncertainties. Component-scale failure data are costly to obtain, and existing methods often rely on oversimplified assumptions or computationally intensive simulations. To address these challenges, this paper introduces the RUL Inoperabilities Model (RIM), a probabilistic framework inspired by the Inoperability Input-Output Model (IIM) that operates directly on coupon-level RUL predictions. The RIM is prognostic-model agnostic, interpretable, and data-efficient, requiring only coupon-level data to train base predictors and a single component-level degradation history for adaptation. By propagating probabilistic coupon predictions to the component level, RIM enables uncertainty-aware SLP. The method is validated on a three-coupon aluminum component using two different base predictors, Hidden Semi-Markov Model (HSMM) and physics-guided Particle Filter (PF), both trained only on single-coupon data, and consistently improves component-level RUL accuracy and uncertainty quantification over a naive baseline.