Ensemble machine learning predicts displacement of cantilevered fibers impacted by falling drops

Abstract

In this study, we consider a simple system of water drops impacting cantilevered, circular fibers with velocity 1.0–2.4m∕s. The dynamics of the system are complicated because the motion of the drop and deflecting fiber at impact are highly coupled and the outcome is influenced by several continuous variables. The unpredictable dynamics of the system call for the use of computational tools that can reveal relationships between variables and make predictions about the physical outcomes for parameter values for which there is no experimental data. This study considers three fibers of contrasting properties, with hydrophilic and hydrophobic wetting conditions, exposed to a range of falling drop diameters and velocities. We predict maximal fiber deflection due to the impacting drop using an ensemble machine learning algorithm combining three base algorithms: a random forest regressor, a gradient boosting regressor, and a multi-layer perceptron. We train and test our algorithm with experimental datasets comprising 405 total trials using three different fibers and five input variables per fiber. The approach allows the determination of relative dominance of the input features in the prediction, reveals shortcomings of traditional scaling approaches for momentum transfer, and shows that drop momentum plays only a minor role in fiber deflection. Finally, our approach provides another example for the application of machine learning to characterize complex and coupled systems in fluid mechanics.