The U-net-based ice pore parameter extraction method for establishing the SAW icing sensing mechanism

Abstract

The porous ice layer formed upon the SAW (surface acoustic wave) propagation path will produce transient acoustic attenuation, and a new icing sensing method can be constructed based on this feature. Obtaining the pore characteristics of an ice layer is a prerequisite and basis for deriving the sensing mechanism of SAW icing. Employing the U-net network, a segmentation model of frozen pores is developed by collecting frozen images and calibrating sample data sets for training in this work. The MIOU and the MPA of the model reach 92.16 % and 95.15 % respectively. Then, the watershed algorithm and post-processing were used to accurately obtain the pore characteristic parameters (porosity, pore number, and average pore area) of the ice layer. On this basis, applying the Biot's theory for two-phase porous media, a theoretical model of the SAW icing sensing mechanism was established, and the acoustic sensing response under different icing types was theoretically simulated. To prove the accuracy of the theoretical simulation, a SAW sensor using a waveguide structure of SU-8/ST-90 °X quartz operating at 80 MHz was constructed and evaluated experimentally. The results demonstrate that the acoustic sensing responses of different ice types are completely consistent with the theoretical results. This fully verifies the accuracy of the ice pore parameters extracted in this work and also confirms the feasibility of using pore microscopic information to achieve ice detection and ice type discrimination.