Enhancing aircraft reliability with information redundancy: A sensor-modal fusion approach leveraging deep learning

Abstract

Redundancy design is a critical way to enhance the reliability and safety of aircraft. However, hardware redundancy significantly increases manufacturing costs and system complexity, while analytical redundancy faces challenges in establishing accurate mathematical models. To address these issues, this paper proposes an information redundancy method for flight data based on sensor-modal fusion. This method leverages deep learning networks to learn the complex coupling relationships between flight parameters from a vast amount of historical flight data. In this respect, a mapping model for flight parameters is established to replace traditional mathematical models used for analytical redundancy. First, the traditional sliding window process is improved by proposing a Fibonacci sampling to balance computational resources and historical view length. Next, a sensor-modal fusion-based prediction model is designed to avoid spatial interactions among sensor features during feature extraction. Furthermore, a sensor attention module and a modal attention module is employed to improve the interpretability of the model. Finally, a Lebesgue evaluation metric is introduced to address ineffective assessment under state balance conditions. The proposed method was validated using real flight data. The results demonstrate that the Lebesgue mean absolute error remained below 1.4 %, outperforming all comparative methods and affirming the effectiveness and superiority of the proposed method. Furthermore, this paper investigated the potential of information redundancy in enhancing aircraft reliability.