On the Improvement of CFD-DEM Coarse Graining Predictions Available to Purchase

Discrete Element Method (DEM) coupled to Computational Fluid Mechanics (CFD) is a powerful tool for simulating complex multiphase flows. The cost of this Eulerian-Lagrangian description, however, increases with the increase of the number of particles ∼O(N_p) which limit its use in natural and industrial scale systems. Efforts to reduce the cost of CFD-DEM capability include reducing the total number of simulated particles by lumping them in larger size representative particles (RP Model). The scaled Representative Particle simulations are less compute intensive compared to the more expensive high fidelity un-scaled/resolved simulations. The prediction accuracy of the RP model, however, decreases as larger scaling factors are used. In the current work, we study the possibility of getting improved results from RP model by using two different techniques. First attempts will be made to identify reasons for reduction in RP model prediction accuracy, then different techniques for improvement of RP model are suggested and tested. In the second part, the ability of the co-kriging surrogate model to improve CFD-DEM predictions by combining many reduced-order RP model simulations with a few high-fidelity unscaled calculations is tested. Appropriate systems are selected to evaluate each method.