General Value Functions for Remaining Useful Life and Failure-Mode Prediction

2026-07-24

AI Summary

This study addresses the challenge of modeling temporal dependencies in predictive maintenance when observations are incomplete or equipment identities are unavailable, hindering accurate remaining useful life (RUL) prediction and fault mode classification. To overcome this, the work introduces vector-valued general value functions (GVFs) combined with temporal-difference (TD) learning for the first time in this domain. By formulating degradation as an absorbing Markov process and jointly learning RUL and fault mode probabilities through multi-step TD(𝑛, λ) and linear projected TD algorithms, the approach ensures temporal consistency. Notably, it requires only fragmented, identity-agnostic partial degradation data, eliminating the need for complete run-to-failure labels. Experiments on event-triggered multimodal simulations and NASA C-MAPSS datasets with sparse annotations demonstrate significant improvements over supervised Monte Carlo baselines with identical architectures, particularly excelling under label-scarce conditions.

Abstract

Remaining useful life (RUL) prediction and failure-mode classification are central tasks in predictive maintenance. Many data-driven pipelines use fixed-window supervised learning with complete terminal labels; such routes do not naturally encode the temporal recursion linking successive degradation-state predictions when observations are partial or unit identities are unavailable. We formulate prognostics as vector General Value Function (GVF) prediction on an absorbing degradation process, treating RUL and failure-mode probabilities as temporally consistent targets rather than independent window-level labels, and estimate them with a multi-step temporal-difference estimator, TD($n,λ$). Supporting theory identifies the Bellman fixed point of the vector GVFs, characterizes the linear projected-TD limit and its relation to complete-return Monte Carlo regression under realizability, and explains when bootstrapped TD targets are less variable than Monte Carlo returns. On an event-triggered multimode simulation and NASA C-MAPSS label-scarce stitch data, TD improves RUL and failure-mode prediction relative to a supervised same-backbone Monte Carlo control, especially under scarce complete labels. Practically, fragmented, identity-free degradation records can contribute local Bellman transitions instead of being discarded until complete run-to-failure labels are available.

Problem

• Remaining Useful Life
• Failure-mode Prediction
• Predictive Maintenance
• Temporal Recursion
• Degradation Process

Innovation

• General Value Functions
• Temporal-Difference Learning
• Remaining Useful Life Prediction
• Failure-Mode Classification
• Predictive Maintenance