Interfacial tension (IFT) is a an extremely important to many systems in both nature and industry. IFT governs the stability of foams and emulsions. The effects of IFT are present in everyday life with products in the food, household, cosmetic and chemical sectors, among many more.
For more than a century, a variety of techniques have been used to measure interfacial tensions between immiscible fluid phases, for example the pendant drop and Wilhelmy plate methods. However, these commonly used methods have some drawbacks. The measurements are usually batch and can take a long time while wasting a lot of materials. Some of the methods are also only suitable for a certain range of IFT values.
Taylor analysis, named after G.I. Taylor, is a microfluidic technique to measure the interfacial tension of two immiscible fluids. The technique examines the deformation of droplets as they approach a microfluidic constriction or expansion. This method allows for a more rapid and continuous characterisation of the IFT.
However, Taylor analysis has some underlying assumptions that reduce the applicability of the approach. There are three main assumptions:
- The droplet has a spheroidal shape (no deformation larger than 15%).
- The droplet is not confined within the microfluidic geometry.
- The fluids involved are Newtonian.
There are many cases were these assumptions do not hold. For example in the case where the droplets are slug-like in shape, confined to the channel walls of the microfluidic device. Also, the fluids examined are usually complex and so do not follow Newtonian assumptions.
Deep learning is a subset of machine learning that almost always take the form of neural networks (NN's). The first deep learning algorithms were established decades ago, with a recent increase in popularity as problems associated with the algorithms are overcome, such as backpropagation.
Convolutional neural networks (CNN's) are used in a wide variety of complex computer vision applications. The greatest advantages of CNN's is that they are not spatially constrained and use hierarchical learning. Spatial invariance is important due to the fact that the droplets present in images can be captured at different points along the constriction. Hierarchical learning is important due to the fact that deformation patterns of the droplets can be present at different locations along the surface, allowing the patterns to be recognised regardless.
The aim of this research project is to combine microfluidics and machine learning to create a rapid, continuously operated system to accurately measure the interfacial tension of soft matter systems. This technique will then be used to examine the static and dynamic interfacial tension of soft matter systems.