Predicting liquid properties via droplet pinch-off and machine learning

The accurate characterisation of liquid properties is critical in various applications, specially in those that require the controlled production of droplets, such as in inkjet printing, pharmaceutical dispensing, and coatings. Conventional methods for determining fluid properties, e.g. viscosity and surface tension, rely specialized instrumentation which can limit efficiency and automation. Here, we introduce a Machine Learning framework that infers fluid properties directly from a single high-speed image of a pendant drop at the moment nearest its breakup. The method offers a scalable and efficient alternative to conventional approaches. Experiments were conducted using Newtonian fluids with a wide range of properties under controlled flow conditions. For each case, physical and operational parameters were recorded, along with images capturing the final stage of droplet breakup. Using extracted droplet contours, supervised regression models were trained to predict fluid parameters with high fidelity. The results demonstrate that the droplet geometry near pinch off has sufficient information to accurately infer fluid properties. Our approach substantially reduces measurement complexity and evaluation time and potentially facilitates integration into automation.