Pattern Recognition for Capillary-Driven Extensional Flows

Abstract

Understanding and analyzing capillary-driven extensional flow dynamics are crucial for applications like inkjet printing and emulsion formation. However, the spatiotemporal shapes of complex fluids as they stretch have been partially analyzed by conventional methods that measure only single points in the slender jet approximation, even though these shapes contain important rheological information. We introduce a new approach that integrates machine learning and flow visualization to classify and estimate fluid composition without relying on traditional rheological models. Our method utilizes captured images using dripping onto substrate capillary breakup extensional rheometry, which specializes in observing the spatiotemporal dynamics of capillary-driven extensional flows. Through these images, we introduce “eigenthinning” extracted via principal component analysis, enabling fluid classification and composition estimation. A k-nearest neighbor classification achieves nearly 100% accuracy using a few principal components (PCs). We extend this to multicomponent fluid composition estimation with promising results. Our model suggests potential improvement through deep learning integration and an adaptive weighting strategy. We also explore using PCs to augment training data sets, enhancing data diversity, and facilitating comprehensive analysis.