Atmospheric icing measurement and online ice type recognition for aircraft utilizing optical fiber sensor and machine learning algorithms

Abstract

For the detection of atmospheric icing process, an ice type recognition and thickness measurement method based on optical fiber sensor is proposed and evaluated. Finite element models based on different structures are built for the four corresponding typical ice types in the atmospheric icing environment. Finite element analysis results show the characteristics of sensor response for different ice types and the potential of real-time ice type recognition using optical fiber sensor. In the laboratory test, under the atmospheric icing conditions, 1655 sets of valid data including glazed ice, mixed ice, rime ice and frost are collected. Cascade models based on machine learning algorithms are built and evaluated using a round-robin leave-one-out method. The model realizes instant ice type recognition and thickness measurement without relying on time-domain curve information. The evaluation results show that the proposed method achieves 92.8 % ice type recognition accuracy and 0.27 mm RMSE for ice thickness measurement.