A Novel Sequence-to-Sequence Method for General Aviation Aircraft Normal Acceleration Prediction Using Flight Recording Data

Hard landing prediction is one of the most typical landing safety incidents analyses, which is the crucial part in the flight operational quality assurance of civil aviation. Although prediction performances based on quick access recorder data have been improved using machine learning and deep learning methods, there have been few studies on hard landing predictions using SD card data from general aviation aircraft recording systems. To address the problem of normal acceleration prediction in hard landing analyses, a novel sequence to sequence prediction method was developed based on SD card flight data. First, the original data was preprocessed and then an attention mechanism-based GRU encoder-decoder model was designed for the normal acceleration sequence predictions. Experiments were conducted based on real SD card flight samples for training flight aircrafts to compare the proposed method with previous approaches, the comparison results from which showed that even when using the relatively low sampling rate flight recording data, the proposed method had significant performance improvements for general aviation aircraft landing safety incidents analyses.