An Optimal Dual Data-Driven Framework for RUL Prediction With Uncertainty Quantification

The prognosis of remaining useful life (RUL) plays a crucial role in the domain of predictive maintenance, particularly for high-end equipment. Data-driven methods are commonly employed for RUL prediction. However, deep learning (DL) approaches provide only point estimates and lack reliability. Statistical models, used for uncertainty quantification, suffer from inaccurate parameter estimation due to incomplete degradation data. To address these issues, we introduce an optimal dual data-driven framework (ODDF) for RUL prediction with uncertainty quantification. This framework incorporates a deep bidirectional recursive gated dual attention unit (DBR-GDAU) as its foundational network, providing robust multistep point estimation. A meta-heuristic optimization algorithm is employed to adaptively select optimal hyperparameters for the optimal DBR-GDAU. Utilizing the multistep predictions from the optimal DBR-GDAU, we establish a nonlinear adaptive Wiener process (AWP) model, with model parameters estimated through maximum likelihood estimation. The probability density function of RUL is derived through the first passage time, and the 95% confidence interval is defined as the interval estimation for RUL. We demonstrate the efficacy and superiority of the proposed method by applying it to real bearing datasets and comparing its performance with state-of-the-art prediction methods.