Ansys Fluent GPU Solver for Mixing Simulations
By Ansys Senior Principal Application Engineer Hossam Metwally
The Ansys Fluent native multi-GPU solver has been shown to provide up to 8x speed up for mixing simulations. The new solver has been tested on various levels of GPU cards (from entry-level laptop cards to server level) all greatly reducing solution time. In this article, I am focusing on two examples representing mixing simulations to showcase the speed-ups when using the GPU solver.
Blend Time Prediction
In this case, I simulated a 2 million Poly mesh mixing tank. Two liquids with different properties were mixed. The first liquid had water-like properties and acted as a buffer material. The second liquid was nearly 50x more viscous and 1.08 more dense than water. The CFD simulation is transient, laminar and uses sliding mesh to represent the impeller rotation (338 rpm). The simulation took less than 37 hours to simulate blend time of nearly 450 seconds on an Nvidia A100 GPU card. Better yet, the experimentally obtained blend time at the same conditions was 420 seconds [1]. Running the same tank at a lower speed of 290 rpm yielded a blend time of 780 [1] while simulation predicts a blend time of 750 seconds and requires 83 hours of simulation time on the same Nvidia A100 GPU card.
Continuous Mixing in a Static Mixer
Static mixers (e.g. of type SMX or Kenics) are very commonly used to continuously mix materials that are highly viscous. Due to the high viscosity, the flow is typically laminar and viscous heating is very common. In addition to the geometrical complexities of these mixers, capturing both the mixing process and the viscous heating effect results in large model sizes. As an example, the model shown below has 26 million Poly cells. The steady state, laminar flow, including viscous heating effects, was run on a single A100 GPU card in less than 10 minutes.
Recommended by LinkedIn
It is important to note that:
For a complete list of Ansys Fluent GPU capabilities, please refer to Ansys Help documentation.
References
[1] Yu et al., “Mixing of Stratified Miscible Liquids in an Unbaffled Tank with Application in High Concentration of Protein Drug Product Manufacturing”, Industrial &Engineering Chemistry Research, American Chemical Society, 2018 [Ind. Eng. Chem. Res, 2018, 57, 3397-3409]
Business Development Manager CFD | CADFEM | ANSYS | Dr.-Ing. | Simulation | H2 & Fuel Cell Technology
11moReally good work! Hope to see Non-Newtonian Fluids soon in GPU Solver!