Ultrasonic signal separation and ice thickness identification methods in dynamic icing process

Ultrasonic pulse-echo icing detectors currently lack sensitivity to the early stages of icing, failing to effectively identify the thickness of thin ice layers and issue timely ice formation warnings during flight. The thin ice layers at the initial stage of icing can result in the overlap and mixing of ultrasonic signals, posing a significant challenge for ice thickness identification. This paper initially simulates ultrasonic pulse-echo signals for ice layers of varying thicknesses using the finite element method. Subsequently, the blind source separation algorithm, FastICA, is utilized to separate the ultrasonic mixture signals at the early stages of icing, with a comparison of the signal separation effects of nonlinear functions G₁, G₂, and G₃. The results indicate that the G₂ nonlinear function exhibits poor stability in signal separation, leading to signal distortion during the process, thus favoring the use of G₁ type and G₃ type nonlinear functions. This paper further analyzes the intrinsic causes of the periodic fluctuations in the phase difference/time of flight curve through signal distribution and correlation features, and validates the feasibility of the algorithm through dynamic icing experiments, achieving effective identification of thin ice layer thickness during the initial stages of icing.