Estimating sliding drop width via side-view features using recurrent neural networks

Area of Science:

• Fluid dynamics
• Surface science
• Machine learning applications

Background:

• High-speed side-view videos are used to study sliding drop dynamics.
• Accurate measurement of drop width is crucial for understanding sliding physics and friction.
• Current methods for width measurement require cumbersome additional equipment, limiting analysis.

Purpose of the Study:

• To develop a method for estimating sliding drop width solely from side-view videos.
• To eliminate the need for front-view cameras or mirrors in drop dynamics experiments.
• To enable comprehensive analysis of sliding drops, including interactions with surface defects.

Main Methods:

• Exploration of various regression and multivariate sequence analysis (MSA) models.
• Application of Long Short-Term Memory (LSTM) neural network with a 20-frame sliding window.
• Validation of model performance using root mean square error (RMSE).

Main Results:

• The LSTM model achieved the best performance with an RMSE of 67 µm.
• This RMSE represents a prediction error of 2.4% for drop widths ranging from 1.6 to 4.4 mm.
• The LSTM model successfully estimated drop width across the entire 5 cm sliding length.

Conclusions:

• Machine learning, specifically LSTM, can accurately estimate sliding drop width from side-view videos.
• This method simplifies experimental setups and expands the scope of drop dynamics research.
• The technique allows for previously unattainable continuous width measurements during sliding.