Montmorillo-Knights
Senior Design - LSU - Team #1
CFD Simulations
Recap:
Mark (team member) headed up geometry creation in October of last year. Based on the specs provided he created a geometry using Fusion 360. Two separate geometries were created one with a gas inlet injector and another without. After speaking with Daniel Barecca (Graduate Student at LSU, studying CFD) we determined that not including the inlet may prove beneficial when meshing because it will decrease the chances of having to refine the mesh at that point. Mark then began taking an ANSYS course on edX to better equip himself in using its interface. Picture to the right shows the geometry with gas inlet created in Fusion 360.
Update: January 27, 2018
Mark began the use of ICEM to develop the OGRID that will be used to mesh the geometry. There seemed to have been a problem with the blocking method used because it was about one straight week of "ogrid not successful" error message. The group met up and Daniel (team member) and Mark began work on resolving this error. This error did get resolved only leading to the realization of another error when refining the mesh around the bends. Daniel Barecca was there to save the day and updated the group on how to proceed. Dr. Tyagi (Faculty Advisor) granted permission to LSU's Supermike and after some familiarization with the High Performance Computing (HPC) platform simulations should begin next week. It was weeks like this where strapping on a boxing glove and attempting to send that cantankerous computer to the moon seemed like a good idea.
Update: February 6, 2018
Mark and Daniel developed a case file for the mesh and began running the data on a regular computer. This lead to the realization that not having the gas injector inlet on the geometry would not allow the ability to adjust incoming gas and water velocities individually. So the mesh which we spent the past few weeks working on was scrapped in favor of the geometry with the gas inlet injector port. Fortunately, the ability to O-grid the geometry came easier but, an immediate wall was hit when trying to block the geometry effectively to eliminate skewed nodes. Hours and Hours were spent on Saturday only to resolve the issue on Monday. After a week, we are back to where we were last week with the previous mesh and are hoping to actually run the case file on Super Mike this week...... hopefully. More caffeine will be required to make this week successful!
Update: February 10, 2018
Well, ladies and gents, we have now officially conquered a major obstacle. We are running our CFD simulations on Super Mike II. It is a good thing too, because we decided to see just how much of an improvement it is to run the simulation on 16 cores rather than 1 core, the improvement is amazing. The single core solving our 170000 node Bench Model simulation would have taken 4.5 days. Super Mike did it in 10 hours using only one node. Mark and Daniel are flying high this week because of this accomplishment! To get an idea of how we felt we encourage you to watch the video below. Also, I have a sneak peak picture of some of the initial simulation results. Mark is going to take over most of the CFD work from here. You should be getting some great quality animations and graphs soon!
Update: February 19, 2018
The first simulation run yielded 16 gb worth of data files, which in post processing were useful for creating nice looking still shots of the phase contour . However, we also want to include animations of the pressure contours, and Mark found out after the fact that animations need to be set up in the case file before running. We ran a second simulation with the same mesh in hopes of building these animations, only to find that the model needs to be partitioned before recording the animation if we want to see inside of the riser. The volume of files with this second try was also almost 50 gb, so hopefully recording just the views we are interested in will cut down on the additional size and computation time. Mark will run the simulation a third time this week with the new changes, as well as create some points corresponding to pressure sensors and flow meters in the model to create graphs for comparison.
Update: March 4, 2018
We finished the unrefined Bench scale simulation runs with 140,000 nodes. Below is a video of the phase animation of the run. We have since refined the mesh to 400,000 nodes and are currently trying to run that simulation. We also have created the geometry for the Pilot scale model and are hoping to start running those simulations this week.
Update: March 12, 2018
​
After 3 attempts, we successfully ran a simulation of the bench scale using the refined mesh. We attempted to change to a 3 equation transition model because the Reynolds number for our model is approaching a turbulent value, but this caused the solution to diverge. Because of this, we stuck with the laminar model and the simulation ran for 18 seconds of flow time and took 2 days to complete. The pilot scale model geometry was also refined and completed this week, and we are finished with the coarse mesh. Due to the larger size of the model, we are reading up on the process of running the simulation on an additional node in Super-Mike.
​
​