A Novel Dual-branch Hybrid Convolutional Transformer Uncertainty Quantification Framework for RUL Prediction

The prediction of aero-engine remaining useful life (RUL) plays a pivotal role in safeguarding flight safety and optimizing maintenance strategies. However, most of the existing RUL prediction approaches predominantly focus on point estimation, while neglecting the quantification of uncertainty. This limitation renders them inadequate for addressing the requirements of practical industrial applications. To address this gap, this paper presents a novel dual-branch hybrid convolutional transformer uncertainty quantification framework (DHCT-UQ). The proposed DHCT-UQ framework constructs a parallel architecture integrating a convolutional neural network (CNN) and a Transformer via the CNN-Transformer Fusion Network (CTF-Net) model. The CNN component is designed to capture local spatial features, whereas the Transformer is tailored to model global temporal dependencies. Through the utilization of a gated fusion mechanism, the framework enables adaptive weighting of the extracted features. Additionally, the Monte Carlo Dropout (MC Dropout) strategy is incorporated. By conducting multiple rounds of stochastic forward propagation, the framework can generate the probability distribution of RUL predictions. Extensive experiments conducted on the CMAPSS public dataset demonstrate that the DHCT-UQ framework outperforms state-of-the-art methods, validating its effectiveness and superiority in aero-engine RUL prediction tasks.