Complex Fluids Latent Space Exploration Towards Accelerated Predictive Modeling

A subtle change in complex fluids microstructures leads to a totally emergent mesoscopic response and macroscopic functionality. In most cases, the predictive structure-property relationships are missing due to the curse of dimensionality and the lack of understanding of the mechanisms that bridge the governing physics interacting across a broad range of spatiotemporal scales. In this study, we introduce a custom data intelligent technique that integrates the non-negative tensor factorization and hierarchical clustering models, dubbed as NTFh. The NTFh model decomposes a physical dataset into lower rank representations with the intention to expose "explainable" latent features. As a proof of concept, we applied NTFh to Darcy friction dataset, where the underlying physics is well established as ground truth. Our findings proved that NTFh can (i) extract the latent features as a combination of physical attributes that simplify the understanding of the dynamics, and (ii) reveal various mechanisms and transitions hidden in the data without prior knowledge. We also show that NTFh can be used to extract latent interdependencies and construct explicit structure-property relationships that bridge physics interacting across a broad range of spatiotemporal scales.