Simplified Models of Remaining Useful Life Based on Stochastic Orderings

Luciano Sanchez, Nahuel Costa, Inés Couso

Keywords: Remaining Useful Life, Prognostics, Health Management, Stochastic Ordering, Monotonic Neural Networks, Simplified Ageing Model

TL;DR: We propose a method to design simplified RUL models for resource-constrained systems by jointly learning a health state model and a stochastic ordering function, ensuring comonotonicity with true RUL and competitive accuracy with reduced complexity.

Abstract: We introduce a method for designing \emph{simplified} models of Remaining Useful Life (RUL) that is especially suited for deployment in resource-constrained environments. Instead of accurately predicting the RUL via complex nonlinear functions, our approach jointly learns (i) a probabilistic health state model and (ii) a stochastic ordering function --- represented by a monotonic neural network --- so that the resulting health indicator is \emph{comonotonic} with the true RUL. Notably, this work is the first study where the learning task simultaneously optimizes both the predictive model and the criterion for comparing model quality. By co-optimizing these elements, our method selects the simplest representation that preserves the essential ordering of degradation, as measured by a smooth approximation of Kendall's \(\tau\) statistic. Experiments on the CMAPSS benchmark and real-world datasets (including turbofan engines and road tunnel fans) demonstrate that our approach achieves competitive prediction accuracy with a drastic reduction in the number of parameters.

Submission Number: 7