Data-driven time-dependent state estimation for interfacial fluid mechanics in evaporating droplets

Area of Science:

• Fluid Dynamics
• Physical Chemistry
• Data Science

Background:

• Sessile droplet evaporation is critical in various applications like biodiagnostics and microfabrication.
• Studying multi-component droplet evaporation experimentally is complex due to parameter space and non-axisymmetric behavior.
• Compositional changes can alter droplet properties, leading to different evaporation regimes.

Purpose of the Study:

• To analyze sessile methanol droplet evolution using data-driven classification and regression techniques.
• To develop machine learning models for predicting droplet evaporation regimes, base diameter, contact angle, and humidity levels.
• To demonstrate the capability of machine learning in capturing the physics of binary droplet evolution.

Main Methods:

• Trained machine learning models using experimental data of methanol droplet evolution.
• Employed classification algorithms (Decision Tree, Naïve Bayes) to estimate droplet evaporation regimes.
• Utilized regression algorithms to predict droplet base diameter, contact angle, and surrounding humidity over time.

Main Results:

• Decision Tree classifier outperformed Naïve Bayes for droplet regime estimation.
• Regression models showed promising performance in estimating humidity, base diameter, and contact angle.
• Models accurately predicted methanol droplet evolution under novel conditions, validating their predictive power.

Conclusions:

• Machine learning offers a powerful approach to study complex droplet evaporation phenomena.
• Data-driven models can effectively capture the underlying physics of binary droplet systems.
• This methodology provides a robust tool for analyzing and predicting droplet behavior in diverse scientific and industrial fields.