Scientific Machine Learning for Modeling and Simulating Complex Fluids

Significance The development of models that accurately describe the behavior of complex fluids under flow is a longstanding challenge in soft materials engineering. Data-driven models for these fluids have yet to make a significant impact, largely due to their inflexibility—once trained, they are able to describe only a single experiment. We propose a framework for learning accurate constitutive models that are not fixed to a particular experiment but may instead describe the fluid in any flow configuration. We demonstrate that these models may be trained on data obtained in a laboratory setting and then applied to predict fluid properties in a multidimensional simulation of an industrially relevant flow. This framework opens broad avenues to rapid soft material design and engineering.