Balancing Interpretability and Uncertainty in Prognostic Models: A TOPSIS-Based Comparative Analysis of N-CMAPSS DS02 Methods

Abstract

Prognostic models are vital for predictive maintenance, enabling accurate prediction of remaining useful life (RUL) in complex systems. However, balancing model interpretability, accuracy, and robust uncertainty quantification remains a significant challenge. This study addresses these issues using the DS02 dataset of New Commercial Modular Aero-Propulsion System Simulation (N-CMAPSS) by developing a systematic framework that integrates interpretability, predictive accuracy, and uncertainty quantification. A key contribution is the use of the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) to rank and evaluate prognostic models based on accuracy, interpretability, and uncertainty. Additionally, the study introduces methods to separately quantify aleatory and epistemic uncertainties, offering deeper insights into model reliability. By analyzing 62 methods from 21 literature sources, this research identifies gaps, synthesizes best practices, and introduces an interpretability-accuracy map to guide model selection. Recommendations for hybrid data-driven and physics-informed approaches further enhance model robustness and applicability. This work advances the development of interpretable, accurate, and reliable prognostic systems aligned with real-world operational needs.