Multi-sensor Information Fusion Method For Aircraft Icing Prediction

The ice and other pollutants attached to the aircraft would directly cause the change of aerodynamic characteristics of the aircraft, and seriously affect the safety of aircraft. Generally speaking, different icing types, degrees, and different icing positions are all the important factors that determine the icing influence extent, they would bring different levels of hazards to flight safety. Meanwhile, they also have different effects on the deicing process efficiency. Therefore, it is an important component to study aircraft ground icing data processing issue to ensure the operation working of civil aviation airport safe and effective.Firstly, the error of icing sensors was analyzed based on the icing state data obtained by the icing sensors, and the distributing diagram method was used to eliminate the careless measured error data of icing sensors. On this basis, the aircraft ground icing thickness detection model was established by using the adaptive weighted data fusion algorithm. The icing characteristic was analyzed to establish the foundation of the aircraft icing state prediction through the research of the icing detection.Secondly, the aircraft icing type prediction model was proposed in this paper, and support vector machine is applied to the aircraft icing classification. The input variables of icing type are analyzed, and then based on the analysis, the appropriate forecasting methods are chosen and an SVM model for aircraft icing type classification is established. The SVM-based classification model is employed to identify aircraft ground icing type and compared with the classification model based on BP neural network method.Finally, the nonlinear dynamics model was used to describe the change of aircraft icing thickness and icing deformation accelerations was viewed as dynamic noise based on analyzing the changing process of icing thickness to aircraft icing thickness forecast issue in this paper. Then, a dynamic prediction model of aircraft thickness is established with the theory of kalman filter. And the kalman filter method based on aircraft icing thickness prediction model was employed to forecast aircraft ground icing thickness and compared with support vector machine and BP neural network prediction method. An adaptive kalman filter based aircraft icing thickness prediction model was proposed to the situation of the tradition kalman filter not predict accurate the catastrophe point of data, and its effectiveness was showed by experiments.