Automated Static and Dynamic Contact-Angle Analysis with Machine Learning

We present an autonomous machine-learning model, Conan-MLII, for fast, reproducible contact-angle measurement. The model was trained on >32 million synthetic images generated from Young–Laplace solutions spanning θ = 20◦–180◦, including the effects of gravity, multiple image scales and surface perturbations (roughness, reflection) as well as including sampling offsets to provide contact-point invariance. An improved contact-point identification (CPID) routine is integrated to enable autonomous processing of real images. The model produced results comparable to Young–Laplace (mean systematic underprediction ∼ 2◦) when tested on 172 experimental images drawn from multiple groups, and was much better at predicting asymmetric and low-angle cases where Young-Laplace fitting commonly misconverges. The total analysis time per image for the machine learning approach was three times faster than Young-Laplace fitting, the current best-practice approach. The model is extended to measurement of advancing and receding contact angles, showing excellent agreement with commercial tensiometers over the range 30◦ − 160◦. The model is provided as open-source for use and further development of contact angle measurement.