Development of machine learning based droplet diameter prediction model for electrohydrodynamic atomization systems

Due to the nature of complex multiphysics of electrohydrodynamic atomization (EHDA) system and the strong nonlinear relationship between process variables and droplet diameter, experiment-based trial and error and traditional numerical simulation have exhibited poor universality or low efficiency in analyzing such systems. In this study, an artificial neural network (ANN) model was developed to efficiently and accurately correlate the relationship between the EHDA process variables (nozzle diameter, conductivity, viscosity, dielectric constant, density, surface tension, flow rate, distance between the nozzle and the grounding electrode, and applied voltage) and droplet diameter. A database containing 8628 EHDA droplet diameter data points was collected and used for training the model. The results showed that the ANN model with 6 neurons could well predict the EHDA droplet diameter, which gives a high determination coefficient (R2) of 0.9998 and a low mean absolute error (MAE) of 0.0071. Impacts of feature inputs on the prediction performance were evaluated, suggesting that the solution properties and operating conditions should be considered as features inputs to ensure the prediction accuracy. CFD simulation was also conducted to compare efficiency and accuracy with the ANN model. Finally, the developed ANN model was used to investigate the effects of process variables. This study provides a powerful intelligent tool for efficient prediction of droplet size in EHDA systems in a green and sustainable way, which could be used in many research fields covering nanomaterial preparation, fuel spraying combustions, biomedical drug preparation, electric field assisted bioprinting etc.