Design of superhydrophobic anti-icing coatings guided by full-process machine learning

Superhydrophobic coatings hold immense potential in anti-icing applications. The preparation of superhydrophobic anti-icing coatings involves multi-parameter design such as coating components and substrate structures, which leads to high trial-and-error rates and unclear directionality in performance optimization. Here, we first report a full-process machine learning framework that integrates large language model, post-hoc explainable machine learning model, and Bayesian Optimization framework to guide the preparation of superhydrophobic coating with superior anti-icing performance. The FPMLF autonomously collected 217 sets of coating components data from 2315 published articles, constructed a regression model for coating components design based on the collected data, and achieved the high throughput screening and optimization of the complex process parameters of the substrate structure. Guided by this framework, we successfully fabricated a coating exhibiting a freezing delay time over 30-fold longer than bare substrate. This work provides systematic support and a methodological foundation for the rational design of advanced functional materials with coupled and complex performance requirements. • A full-process machine learning framework (FPMLF) was developed to design the superhydrophobic anti-icing coatings. • The FPMLF achieved data collection, prediction of coating components and optimization of substrate structure. • Through the FPMLF, a coating with a freezing delay time over 30-fold longer than bare substrate was obtained.