A Hybrid Remaining Useful Life Prediction Framework Based on Improved Autoencoder and Nonlinear Wiener Process

Modern maintenance strategies increasingly rely on accurate estimation of equipment lifespan to ensure operational reliability and reduce unexpected failures. Remaining useful life (RUL) prediction provides a data‐driven basis for making informed maintenance decisions. Considering that single RUL point prediction results cannot reflect the uncertainty risks of prediction, this paper proposes a novel hybrid prediction framework for aero‐engine remaining life named AG‐ECA‐UAE‐Wiener to achieve uncertainty quantification of prediction results. First, based on an improved U‐Net autoencoder, the Efficient Channel Attention (ECA) mechanism is integrated into multi‐scale convolutional modules to strengthen feature extraction, while the Attention Gate (AG) attention gating mechanism is introduced at the skip connections of the encoder and decoder, thereby autonomously learning and constructing high‐quality health indicators (HI) from massive monitoring data. Second, a Wiener‐based stochastic degradation model, enhanced with effects and an acceleration factor, is adopted to capture the progression of deterioration, which can not only capture the heterogeneity of individual degradation rates but also provide a probability density function (PDF) while outputting RUL prediction values, thus achieving uncertainty quantification of prediction results. Finally, the proposed hybrid prediction framework is validated on the C‐MAPSS turbofan engine dataset, and experiments show that this method effectively quantifies the uncertainty of prediction results while ensuring high‐precision prediction.