The shape of things to come: Axisymmetric drop shape analysis using deep learning

Area of Science:

• Physical Chemistry
• Materials Science
• Computational Science

Background:

• Traditional Axisymmetric Drop Shape Analysis (ADSA) for surface tension determination faces limitations in computational speed and image quality.
• Accurate measurement of surface tension is crucial in various scientific and industrial applications.

Purpose of the Study:

• To develop and evaluate a machine learning-based approach using a convolutional neural network (CNN) for faster and more accurate pendant drop image analysis.
• To compare the performance of the CNN model against traditional ADSA in terms of precision, speed, and robustness.

Main Methods:

• A CNN model was trained to predict surface tension from pendant drop images.
• The CNN model's performance was benchmarked against traditional direct numerical integration ADSA.
• The model's accuracy in predicting other drop properties like volume and surface area was also assessed.

Main Results:

• The CNN model achieved high precision in surface tension prediction (+/-) 1.22×10^-1 mN/m at a speed of 1.50 ms^-1, exceeding traditional ADSA by over 5×10³ times.
• The model demonstrated robustness, maintaining an average error of 2.42×10^-1 mN/m even with challenging images (misaligned, out-of-focus).
• The CNN also accurately determined other drop properties such as volume and surface area.

Conclusions:

• The CNN-based approach offers a significant advancement in pendant drop analysis, providing a faster, more accurate, and robust method for determining surface tension.
• This machine learning technique holds promise for overcoming the limitations of conventional ADSA, particularly in high-throughput or resource-constrained settings.