Remaining useful life estimation utilizes historical data and advanced analytic techniques to estimate the remaining lifespan of systems, components or assets, for enabling proactive decision-making and efficient resource management. However, existing data-driven prognostic technologies often provide deterministic values of remaining useful life which do not represent uncertainties. In practice, uncertainties are due to factors such as noise, measurement errors in the prediction and limited availability of data to train the predictive models. In this work, we propose a novel framework that addresses the issue of prognostic unertainty by generating prediction intervals for single-point estimators, thereby enabling robust uncertainty quantification. The proposed framework overcomes the drawbacks of previous conformal prediction methods, which lacked interval adaptivity and relied on the assumption of data exchangeability. To validate our framework, we performed experiments using the Commercial Modular Aero-Propulsion System Simulation (CMAPSS) datasets. Results demonstrate that both conventional machine learning estimators and modern deep learning estimators yield reasonable prediction intervals within the proposed framework. The superiority of our framework over previous conformal prediction methods is demonstrated.
A General Data-Driven Framework for Remaining Useful Life Estimation with Uncertainty Quantification Using Split Conformal Prediction
Xu W.;Zio E.
2023-01-01
Abstract
Remaining useful life estimation utilizes historical data and advanced analytic techniques to estimate the remaining lifespan of systems, components or assets, for enabling proactive decision-making and efficient resource management. However, existing data-driven prognostic technologies often provide deterministic values of remaining useful life which do not represent uncertainties. In practice, uncertainties are due to factors such as noise, measurement errors in the prediction and limited availability of data to train the predictive models. In this work, we propose a novel framework that addresses the issue of prognostic unertainty by generating prediction intervals for single-point estimators, thereby enabling robust uncertainty quantification. The proposed framework overcomes the drawbacks of previous conformal prediction methods, which lacked interval adaptivity and relied on the assumption of data exchangeability. To validate our framework, we performed experiments using the Commercial Modular Aero-Propulsion System Simulation (CMAPSS) datasets. Results demonstrate that both conventional machine learning estimators and modern deep learning estimators yield reasonable prediction intervals within the proposed framework. The superiority of our framework over previous conformal prediction methods is demonstrated.File | Dimensione | Formato | |
---|---|---|---|
A_General_Data-Driven_Framework_for_Remaining_Useful_Life_Estimation_with_Uncertainty_Quantification_Using_Split_Conformal_Prediction.pdf
Accesso riservato
:
Publisher’s version
Dimensione
1.21 MB
Formato
Adobe PDF
|
1.21 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.