A Bayesian Deep Neural Network Model for Remaining Useful Life Prediction Using a Multi-Point Active Query Strategy  
Author Ying Chen

 
Co-Author(s) Taofeng Fan; Mingchih Chen; Xufeng Zhao

 
Abstract Accurate remaining useful life (RUL) prediction is critical for ensuring the reliability and safety of mechanical systems, particularly in aerospace and industrial applications where unexpected failures can lead to severe consequences. While deep learning (DL) has shown promise in RUL prediction, its performance heavily relies on large amounts of labeled training data, which are often scarce in real-world scenarios due to high acquisition costs and operational constraints. Active learning (AL) offers a potential solution by strategically selecting the most informative samples for annotation, thereby improving model efficiency with limited data. However, existing AL-based RUL prediction methods primarily focus on singlepoint query strategies, which may limit learning efficiency and generalization. To address this gap, this paper proposes a novel multi-point active query strategy that enhances sample selection by simultaneously acquiring multiple high-value data points in each query cycle. Furthermore, we integrate this approach with a Bayesian deep neural network to quantify prediction uncertainty, enabling more robust and interpretable RUL estimation. Extensive experiments on the C-MAPSS dataset demonstrate that the proposed method significantly outperforms conventional AL strategies and baseline models, achieving higher prediction accuracy with fewer labeled samples. This work provides a promising direction for data-efficient RUL prediction in practical industrial applications.

 
Keywords Remaining useful life; Active learning; Bayesian deep neural network; Multi-point active query
   
    Article #:  RQD2025-377
 

Proceedings of 30th ISSAT International Conference on Reliability & Quality in Design
August 6-8, 2025