Determination of the single droplet post-impingement pattern on a dry wall: A data-driven approach

Abstract

Basic droplet-wall interaction such as post-impingement patterns is of great importance since it can quantitatively provide an estimation of energy transformation in spray-wall interaction dynamics. Most existing studies have developed physical-based empirical criteria to determine the droplet-wall post-impingement patterns. However, these approaches are subject to common issues such as scalability and low accuracy. In this study, a data-driven approach utilizing machine learning (ML) classification models is proposed for the identification of droplet-wall post-impingement patterns. An experimental data pool with 1093 observations of a single droplet impinging on a dry wall is established. 16 input features describing liquid properties and wall characteristics are first determined based on domain knowledge. Feature engineering methods are adopted to investigate the most informative and influential features of the splash phenomenon. The selected feature subsets/spaces are then fed into six well-known classification models (i.e., classifiers) individually for the model tuning. The performance of the classifiers is evaluated by metrics such as accuracy, precision, recall, and F1 score. The results show that ML classifiers achieve a good performance in identifying splash cases, providing a higher prediction performance than existing criterion-based approaches. More importantly, this study demonstrates the applicability of ML method on droplet/spray-wall interaction research and proposes a feasible workflow schematic for future studies.