Machine learning enables predictions of microfluidic step emulsification

The precise control of droplet generation in microfluidic systems has enabled breakthroughs in fields such as drug delivery and nanomaterial synthesis. Step emulsification, a pivotal method within this field, offers high-throughput production while generating highly uniform droplets, making it a proper technique for these applications. This study presents the first integration of machine learning with step emulsification, enabling precise predictions of droplet diameter and generation rate. Unlike previous approaches, which relied on simplified equations with limited accuracy, our method uses a large-scale dataset of 1244 samples generated through three-dimensional numerical simulations. This comprehensive dataset incorporates all influential factors in step emulsification, providing the most accurate prediction setup to date. The artificial neural network (ANN) model was developed and trained on this dataset and achieved an <i>R</i>² score of 0.989 for droplet diameter and 0.972 for generation rate, accompanied by low root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) values, validating its high accuracy. Furthermore, sensitivity analysis using SHapley Additive exPlanations (SHAP) method was conducted to determine and discuss the most influential factors in step emulsification. Subsequently, to streamline the iterative process of achieving target droplet sizes and generation rates for researchers and industry, we developed a graphical user interface (GUI), allowing users to input parameters such as fluid properties and device geometry, generate real-time predictions, and display results. This pioneering integration of machine learning and step emulsification accelerates microfluidic design optimization, minimizing reliance on extensive trials and enabling advancements in precision-controlled droplet generation. Integration of machine learning with step emulsification: Developed an artificial neural network (ANN) model to predict droplet diameter and generation rate for the first time in step emulsification.High-accuracy predictions: Achieved <i>R</i>² scores of 0.989 for droplet diameter and 0.972 for generation rate using a dataset of 1244 samples from 3D numerical simulations.Sensitivity analysis with SHAP: Identified and discussed the most influential parameters in step emulsification droplet generation.User-friendly graphical interface: Developed a GUI to enable real-time prediction of droplet characteristics based on input parameters. Integration of machine learning with step emulsification: Developed an artificial neural network (ANN) model to predict droplet diameter and generation rate for the first time in step emulsification. High-accuracy predictions: Achieved <i>R</i>² scores of 0.989 for droplet diameter and 0.972 for generation rate using a dataset of 1244 samples from 3D numerical simulations. Sensitivity analysis with SHAP: Identified and discussed the most influential parameters in step emulsification droplet generation. User-friendly graphical interface: Developed a GUI to enable real-time prediction of droplet characteristics based on input parameters.