Machine Learning for Covariate-Driven Changes in Weibull Scale Parameter

This research addresses the limitations of conventional, static survival models for Remaining Useful Life (RUL) estimation in Prognostics and Health Management (PHM) by proposing and implementing a data-driven framework. The study posits that a system's characteristic life, represented by the Weibull scale parameter (λ), is not static but a dynamic variable influenced by time-varying operational and environmental factors, or covariates. Unlike traditional methods, such as the Accelerated Failure Time (AFT) model, which rely on restrictive linear assumptions to link covariates to the lifetime model, this methodology employs supervised Machine Learning (ML) models to quantify this complex and potentially non-linear relationship. The methodology is validated using the National Aeronautics and Space Administration (NASA) Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) aircraft engine degradation dataset. The results demonstrate that ensemble-based methods like Light Gradient Boosting Machine (LightGBM) and Extreme Gradient Boosting (XGBoost) significantly outperform classical statistical models across a suite of standard regression metrics and custom-developed Weibull-centric error metrics. The consistent performance rankings across all metrics validate the effectiveness of the proposed approach and confirm its capability to capture the intricate degradation dynamics of real-world systems.