RUL Prediction and Uncertainty Quantification Based on 2D Convolution Neural Network and Approximate Bayesian Framework

The prediction of mechanical equipment's' Remaining Useful Life (RUL) and Prognostics and Health Management (PHM) is attracting increasing attention. However, most studies provide only point predictions of RUL, which may lack reliability in complex operational scenarios where mechanical equipment experiences fluctuating conditions, such as high loads or impacts. To address this limitation, this paper proposes a novel approach that integrates a sliding-window sampling method with exponential smoothing for data preprocessing, transforming one-dimensional feature sequences into two-dimensional feature maps. Then developed a 2D convolutional neural network (CNN) based on these feature maps to predict RUL. In addition, an approximate Bayesian framework combined with Kernel Density Estimation (KDE) is employed to get the probability density function (PDF) of the RUL prediction. The proposed approach is validated using the C-MAPSS dataset for turbofan engine degradation, with comparative results demonstrating that the method can accurately provide both RUL point predictions and their associated PDFs.