Enhancing aviation safety and mitigating accidents: A study on aviation safety hazard identification

Identifying promptly potential safety risks and threats in the aviation system reduces the occurrence of aviation incidents (especially accidents), protects passengers’ lives and property, and improves aviation emergency management capabilities. Such a strategy has significant implications on mitigating incidents in the aviation system and formulating efficient and orderly safety measures. To this end, this study proposes an Intelligent Aviation Safety Hazard Identification Method that combines Text Mining and Deep Learning (DL) technologies as applied on the incident data and hazard knowledge. Specifically, the Method entails Named Entity Recognition (NER) and knowledge graph visualization. Firstly, the model of Bidirectional Encoder Representations from Transformers (BERT) is employed to process aviation safety incident texts and generate word vectors based on contextual information. The trained word vectors are then input into the model of Bi-directional Long-Short Term Memory and Conditional Random Field (Bi-LSTM-CRF) to extract deep-level safety hazard entities. Next, the extracted safety hazard entities are stored and visualized using the Neo4j database to construct a knowledge graph from which the analyst can directly assess the situation. The effectiveness of this method is validated using incident records of an aviation maintenance company. The proposed method can effectively identify civil aviation safety hazard entities and uncover the intrinsic connection between incidents and safety hazards, which allows relevant personnel to quickly understand the nature and mechanism of an incident and proactively apply preventive measures, thereby providing dynamic support for strengthening aviation safety hazard management.