Machine learning-based splash prediction model for drops impact on dry solid surfaces

Abstract

The impact of drops on dry solid surfaces has important applications in engineering. The post-impact behavior of drops can be classified into non-splash and splash, and there is a lack of splash prediction models that well consider the independent and coupled effects of liquid properties, drop impact characteristics, and surface properties. In this study, machine learning methods of Random Forest (RF) and Support Vector Machine (SVM) are applied to build splash prediction models and analyze the effects of different features. The RF model achieves good prediction accuracy and identifies the roughness R*, Weber number We, Reynolds number Re, and contact angle θeq as the most influential parameters, with decreasing importance. The interpretability analysis shows the increasing splashing tendency with increasing We, Re, and R* and decreasing cos θeq, and a special case of non-splash by drops impact on hydrophobic surfaces with cos θeq ≈ −0.45 is found, which can be explained by the coupled effects of drop and surface features. The classical splash prediction model, K-parameter model, is improved by SVM in an explicit form and considering the effects of liquid properties, drop impact characteristics, and surface properties. The improved K-parameter model has good performance for surfaces with various roughness and wettability, and its prediction accuracy reaches 86.49%, which is significantly higher than 67.57% of the K-parameter model, 46.49% of the Riboux and Gordillo model, and 66.10% of the Zhang model. This study is expected to provide valuable insight into the control of non-splash or splash of drops according to different requirements during applications.