Deep-Learning-Based Model for Accident-Type Prediction During Approach and Landing

Abstract

Modern civil aircraft is designed for a greater workload. While the volume of air traffic has grown exponentially over the past two decades, the total number of accidents has been flat or slightly higher than past averages. However, the fatality and property losses caused by aviation accidents are unbearable even if they happen only once. As reported by a Statistical summary from Boeing in 2019, nearly 40% of the deaths occurred during the final approach and landing. A deep-learning-based model for accident-type prediction during approach and landing was established, which explores the implied association between accident types and accident factors from the risk factors affecting the occurrence of unsafe events. First, based on the investigation reports of civil aircraft fatal accidents in the recent 43 years, 20 common accident risk factors were extracted by the human factors analysis and classification system method. Implicit association of accident factors was excavated by association rules and 6 key factors were extracted from 20 factors. Back propagation neural network, Radial basis function neural network, and Elman neural network applicable to the classification were selected for multiple learning and training. The classification results showed that the comprehensive prediction accuracy of accident types reaches 86.7%, which can effectively determine the types of accidents that may occur during the approach and landing phases of civil aircraft, make emergency measures in advance, and incorporate key accident factors into the assessment scope of civil aircraft safety management system to ensure the aviation operation safety.