Obstruction-Mediated Breakup of Shear-Thinning Droplets in Microfluidic Environment: Experiments, Simulations, and ANN Insights.

We report the breakup dynamics of shear-thinning droplets in microfluidic environment through a comprehensive integration of experiments, numerical simulations, and machine learning (ML). Particularly, we investigate the breakup of droplets interacting with a ribbed structure within a microfluidic conduit. Our findings reveal an on-demand, obstacle-driven breakup under varying orientations and rheological conditions. Based on experimental results, and supported by three-dimensional numerical simulations, we reveal that the underlying breakup dynamics depend on the intricate balance between the pressure differential and the shear-stress distribution along the droplet interface, which governs the change in the advancing curvature of the shear-thinning droplet. In addition, we developed a specialized artificial neural network (ANN) that incorporates six key input variables: the capillary and Reynolds numbers for both continuous and dispersed phases, the mother droplet length, and the obstacle aspect ratio, to predict the breakup volume of the daughter droplets. Using SHapley Additive exPlanation (SHAP) applied to our machine learning model, we quantified the relative importance of these parameters in controlling droplet breakup dynamics. We found that the obstacle aspect ratio plays the dominant role (∼70%), followed by the mother droplet length (∼10%), where these percentage contributions correspond to the mean absolute SHAP values. We believe that the development of the ANN model for the breakup of non-Newtonian droplets offers significant insights for researchers in point-of-care diagnostics, where shear-thinning fluids, such as blood, are routinely encountered.