Predicting liquid properties and behaviour via droplet pinch-off and machine learning

Journal article

Here, we demonstrate that the time-evolving interface during droplet formation—and, in particular, the morphology near pinch-off—encodes sufficient physical information for machine-learning (ML) models to accurately infer key fluid properties, including viscosity and surface tension. A dataset was constructed from high-speed imaging of droplet formation in both dripping and and jetting regimes. The frame closest to break-up was selected and the contour of the droplet extracted and recorded, along with the corresponding flow conditions and fluid properties. Experiments were conducted using Newtonian fluids, such as silicon oils, aqueous solutions of ethanol and glycerin, and methanol, under controlled conditions, spanning 0.001 < Re < 200 and 0.01 < Oh < 20, providing 840 samples. Supervised regression models were trained to predict fluid properties from these inputs. The models accurately infer surface tension and viscosity across a wide range of fluids using the single high-speed snapshot. Moreover, apart from predicting fluid properties from a snapshot, the proposed framework can carry out the inverse problem and predict droplet shape at break-up based on fluid properties and flow conditions. Unsupervised clustering of the learned representations reveals distinct regions in the Re–Oh and Bo–Oh parameter spaces, indicating that the latent space captures meaningful physical structure and provides insight into droplet dynamics. These results establish a data-driven alternative to conventional measurement techniques, reducing experimental complexity and evaluation time while enabling integration into automated systems.

Authors: Wang, J; Chen, Q; Constante-Amores, CR; Gorse, D; Castrejon Pita, AA

• Publication status: Accepted
• Peer review status: Peer reviewed
• Publisher: American Physical Society
• Journal: Physical Review X
• Acceptance date: 2026-06-08
• DOI: 10.1103/v3p1-lmzl
• EISSN: 2160-3308
• Language: English