Uncertainty Quantification as a Complementary Latent Health Indicator for Remaining Useful Life Prediction on Turbofan Engines

2025-07-09

AI Summary

To address the susceptibility of health indicators (HIs) to aleatoric and epistemic uncertainties in turbofan engine remaining useful life (RUL) prediction, this paper proposes a novel HI construction method that explicitly quantifies and embeds uncertainty into the latent space of autoencoders. Our approach innovatively separates and jointly models both uncertainty types via cross-coupled health information to enhance representation robustness. We instantiate uncertainty-aware latent HIs using both standard autoencoders and variational autoencoders, then integrate them into downstream machine learning models for RUL estimation. Evaluated on the NASA C-MAPSS dataset, the proposed method significantly outperforms conventional handcrafted HIs and end-to-end deep learning models, achieving prediction accuracy comparable to state-of-the-art RUL approaches. Results empirically validate that explicit uncertainty modeling is critical for improving HI quality and, consequently, RUL prediction reliability.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Uncertainty Representations
• Intelligent Robots: State Estimation

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Health Indicators (HIs) are essential for predicting system failures in predictive maintenance. While methods like RaPP (Reconstruction along Projected Pathways) improve traditional HI approaches by leveraging autoencoder latent spaces, their performance can be hindered by both aleatoric and epistemic uncertainties. In this paper, we propose a novel framework that integrates uncertainty quantification into autoencoder-based latent spaces, enhancing RaPP-generated HIs. We demonstrate that separating aleatoric uncertainty from epistemic uncertainty and cross combining HI information is the driver of accuracy improvements in Remaining Useful Life (RUL) prediction. Our method employs both standard and variational autoencoders to construct these HIs, which are then used to train a machine learning model for RUL prediction. Benchmarked on the NASA C-MAPSS turbofan dataset, our approach outperforms traditional HI-based methods and end-to-end RUL prediction models and is competitive with RUL estimation methods. These results underscore the importance of uncertainty quantification in health assessment and showcase its significant impact on predictive performance when incorporated into the HI construction process.

Problem

• Integrates uncertainty quantification into autoencoder-based latent spaces
• Separates aleatoric and epistemic uncertainties for accurate RUL prediction
• Enhances health indicators using standard and variational autoencoders

Innovation

• Integrates uncertainty quantification into autoencoder latent spaces
• Separates aleatoric and epistemic uncertainties for accuracy
• Uses standard and variational autoencoders for health indicators