Prediction of air bubble entrapment during drop impact on solid hydrophobic surface: Combined machine learning and experimental validation

One of the fundamental requirements of painting and coating technologies is the controlled deposition onto the solid surface. However, an uncontrolled air bubble entrapment under the liquid layer could deteriorate the quality of the aforementioned process. Here, we present a deep-learning approach using both simulated and experimental images to detect air bubble formation during drop impact on solid parafilm (hydrophobic) surface. The simulation is performed using the phase-field modeling (PFM) approach. Several cases with varying drop diameters and impact velocities are considered to study air entrapment both numerically and experimentally. Using the simulated and the experimental results (top and side views), we have developed a deep learning approach based upon the VGG16 convolutional neural network (CNN) to predict bubble entrapment. The CNN, trained based on the experimental images, reveals better predicting capabilities than the one trained based on the simulated images. Also, for both data sets the prediction is more accurate when training the model using the side rather than the top views. The results allow for a fast and precise prediction of air bubble entrapment.