J.G.E.M. Fraaije
Siemens Culgi,
Netherlands
Keywords: molecular modeling, machine learning
Summary:
In designing new formulation products, a particularly challenging step is from materials discovery (chemistry) to upscaling in production (engineering). Experimentally, one goes from robot-assisted screening to reactors in the factory, from small volumes to large volumes, from molecules and micro-structures to the continuum. There is also a vast difference in speed: in discovery, one can easily handle potentially thousands of formulations a year. Scaling up is a much slower process, limited to a tiny fraction of that. One of the reasons is easy to understand. To rational predict rheology on the reactor scale, one needs Computational Fluid Dynamics (CFD) equipped with Constitutive Equations (CE). But calibrating CE’s with experimental rheometry is time-consuming, expensive, and often very difficult to do. Therefore, if we could somehow predict rheological behaviors in an early discovery stage, that would be a great aid in making the value chain more efficient. The challenge is considerable. Formulations can contain many different ingredients, almost always have a delicate microstructure, and are non-Newtonian with strong non-linear viscoelastic behaviors. We introduce Stokesian Particle Dynamics (SPD) as a novel computational method focusing on the intricate microrheology of this peculiar in-between world. We borrow a soft-force model for the interactions between collections of atoms (‘beads’) from traditional coarse-grained simulations, such as Dissipative Particle Dynamics and Coarse-Grained Molecular Dynamics. But we also borrow from the continuum world: the Scallop theorem rules in the micro-world, which says that all inertial forces are negligible. In accordance, SPD is a mass-less particle dynamics simulation. In comparison with the alternative Stokesian Dynamics (SD) method (equally mass-less), we retain particles for the fluid. SPD has all the possibilities to capture molecular specificity and micro-structures but then emphasizing Stokes flow in the mesoscale regime. The method is beneficial for applications with microstructures 10-1000 nm. We demonstrate the algorithm with a few applications of industrial relevance: reactive polymer systems and emulsions. We show how one can extract viscoelastic parameters from such simulations. Then, the parameters can be fed into a traditional CE in a CFD package in a second stage. In essence, the best approach is to use the microrheology SPD simulations as a virtual rheometer, which can replace the burden of experimental rheometry. All simulations are with Siemens’ Simcenter Culgi, Simcenter StarCCM+, and gPROMS.