An aircraft high-risk subject test flight risk warning model based on multi-source transfer learning

The flight test phase is a crucial stage in the process from aircraft development to final service. During this phase, the aircraft inevitably faces high-risk flight test subjects. Due to the complexity of the subjects encountered in this stage, the aircraft is more prone to risks, and the limited flight test data makes it difficult to construct suitable machine learning warning models for effective risk prediction. In response, this study proposes a high-risk subject flight test risk warning model based on Multi Source-Tradaboost. This model uses flight data from other aircraft that are similar but of different models to the test aircraft to assist in the construction of high-risk subject warning models, improving the warning model. The case analysis constructs a warning model using actual historical flight data from a certain type of test aircraft to warn of typical hard landing risks during plateau takeoff and landing subjects. The analysis results show that this model makes full use of the flight data from two other similar models of aircraft and can achieve warning at a sufficient pre-warning altitude of 50 ft. The optimal model's recall rate reached 0.85, and a comparison with other classical machine learning methods without transfer learning fully validates the superiority and effectiveness of the multi-source transfer learning warning model.