Answers are grounded in this paper and cite their sources.
Prognostics and Health Management (PHM) plays a crucial role in maximizing operational efficiency, minimizing maintenance costs, and enhancing system reliability. Predicting Remaining Useful Life (RUL) is a key aspect of PHM, inherently incorporating uncertainty. This paper focuses on uncertainty quantification (UQ) within Data-Driven Models (DDMs), particularly Machine Learning (ML), such as Long Short-Term Memory (LSTMs), and stochastic models namely Hidden Markov Models (HMMs). While ML models emphasize accuracy, stochastic models offer a different paradigm for prognostics, directly addressing uncertainty. Traditional categorizations of uncertainty as aleatory and epistemic face challenges in practical implementation. This paper explores how, in prognostics, HMMs primarily tackle aleatory uncertainty, whereas LSTMs predominantly address epistemic uncertainty. It also discusses the complexities of uncertaintymanagement in prognostics and analyzes further an already proposed alternative approach to categorize uncertainties. Despite theoretical advancements, practical implementation remains challenging, especially for DL models due to their limited interpretability. This study sheds light on UQ challenges and offers insights for future research directions in prognostics.
No takes yet. Share an insight, caveat, or question.
Salinas-Camus et al. (2024) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: