Remaining useful life prediction for stratospheric airships based on a channel and temporal attention network

Abstract

Stratospheric airships are a research hotspot in the field of near space because of their durability, low cost, wide-area coverage, and relatively rapid response capabilities. Predicting the remaining useful life (RUL) of airships is the key to ensuring long-term stable residence and reducing maintenance and support costs. However, existing diagnostic and predictive techniques for airships primarily focus on individual components and consider information from a single scale, making it difficult to apply to the entire airship system. This paper proposes an end-to-end framework that utilizes both temporal and channel attention(CTA) mechanisms to extract multi-scale information from the airship data. We used the SE-ResNet module to extract high-dimensional abstract features from airship data and employed a Transformer to learn the degradation information across the temporal dimension. The effectiveness of the proposed method was verified using simulated datasets and compared with other deep-learning prediction methods through comparative analysis and ablation experiment results. The results demonstrate the superiority of the proposed approach over other predictive models in terms of effectively forecasting airship failure times.