AM Seminar: Two-way coupled cloud-in-cell modeling of non-isothermal particle-laden flows at a "SPARSE" cost

Speaker Name: 
Soren Henri Taverniers
Speaker Title: 
Postdoctoral Fellow
Speaker Organization: 
Stanford University
Start Time: 
Monday, April 8, 2019 - 4:00pm
End Time: 
Monday, April 8, 2019 - 5:00pm
BE 372
Daniele Venturi


Particle-laden flows are encountered in a broad range of natural problems and engineered systems. This includes geological sedimentation processes and pollutant dispersion in the atmosphere, but also high-speed gas environments such as volcanic eruptions, detonation of explosives and scramjet combustors. For problems with more than O(103) particles, particle-resolved simulations become computationally too intensive and a Particle-Source-In-Cell (PSIC) approach is the only feasible option. However, such models typically involve empirical closures for the drag that are only valid in a limited range of Mach and Reynolds numbers. When the number of particles surpasses O(109), even PSIC may become prohibitively expensive and groups of particles may then be amalgamated into computational “macro-particles”, leading to the so-called “Cloud-In-Cell” (CIC) method. CIC reduces computational cost, but typical implementations do not account for the effect of subgrid-scale (SGS) stresses, induced by the interaction between the fluid and dispersed phases, on particle dispersion patterns. In the first and main part of my talk, I will present a novel two-way coupled CIC formulation for particle-laden flows that accounts for cloud size and subgrid-scale stresses using averaging techniques, and for cloud deformation using methods from continuum mechanics. It traces a physical cloud of particles as a point and distributes its influence on the carrier flow via a multivariate Gaussian distribution function. Via two-dimensional benchmark simulations of a normal shock impinging on an initially stationary particle cloud, I will show that the two-way coupled model predicts the average cloud position and spread more accurately than a two-way coupled, first-order CIC approach. Through an appropriate initial division of the particle cloud into subclouds, predictions of the time-averaged horizontal and vertical cloud spread match those of a reference Particle-Source-In-Cell (PSIC) approach to within a few percent using up to two orders of magnitude fewer computational particles.

In the second part of my talk, I will discuss ongoing work to integrate a PSIC/CIC macro-scale approach with particle-resolved meso-scale simulations to achieve a highly accurate multi-scale modeling framework for particle-laden flows. This involves the use of meso-scale informed surrogate models built using machine learning techniques for closing the drag term in the macro-scale particle momentum equation.

Short Bio

Dr. Taverniers currently holds a postdoctoral research fellow position in the department of Energy Resources Engineering at Stanford University. He received his  PhD in Engineering Physics from the University of California, San Diego in 2016, and was a postdoctoral researcher at San Diego State University prior to joining Stanford. His expertise is in predictive modeling of multi-scale/multi-physics problems in the presence of stochastic noise and uncertainty. His current work focuses on accelerating uncertainty quantification of hot spot formation in reactive granular media.