HVAC Simulation Grinding to a Halt

[GeorgeBGrant]

Hi All! I've used PyroSim to generate a fire compartment model (~ 2,500 m3) with explicit leakages (i.e. holes in the walls) totalling ~ 1 m2. The fire grows to around 11 MW and thereafter becomes ventilation controlled in agreement with previous CFAST simulations. I have since added an HVAC extract network to the model and defined the fire compartment as Zone01 but left the air leaks as described. The HVAC network is connected to several other volumes (ranging from 20 ~ 70 m3) which are defined as separate pressure zones & have leakage area ratios (Aleak/Awalls) of the order 10-6, which I appreciate is maybe pushing the model somewhat as I've seen Jason Floyd's comment previously that the HVAC model expects leakage ratios to be in the range 10-3 ~ 10-4, however I've run the model without a fire and the HVAC system seemed to run OK. Incidentally these smaller volumes also have steady air injection rates of around 5 l/s throughout the simulation. Now I've added the fire compartment to the simulation the run has stuck on timestep 250s of 850s. I'm wondering if it's my use of explicit leakage definition in the fire compartment, the very small leakage ratios elsewhere or maybe that the extraction vent in the fire compartment is struggling due to its proximity to the fluctuating flame zone on that boundary? Any help would be gratefully received!

[drjfloyd]

We would need the fds (not pyrosim) input file simplified as much as possible while still keeping the error.

[GeorgeBGrant]

Thanks Jason. I'll have a look & simplify as much as possible then post the .fds info.

[GeorgeBGrant]

I was struggling to simplify the FDS code and repeat the error but in the meantime saw another post where you suggested that the code might struggle when there are insufficient losses specified in the HVAC system. My (SS) ducts were originally input with roughness 2.5E-6 metres but I've now found better data for this and increased the roughness to 15E-6 m. The run has now progressed to 420s and failed with the "forrtl: severe (157) Program Exception - access violation" error but looking at the offending mesh & coordinates it looks like this might be due to a mesh boundary that crosses a fire source very close to the edge of the fire vent surface. I'm investigating & will close this discussion - thanks again!

[GeorgeBGrant]

I've opened this query again as I'm still having problems running this case. I reverted to 6 no. mesh partitions in the fire compartment and the runs hangs at t~413s [Timestep 82,300] with the same "forrtl: severe (157) Program Exception" error message. However when I run the same case with no fan in the HVAC system, it runs the complete 849s simulation time in around 72h (Dell Precision 5820 Workstation, Xeon Processor with 18 no. physical cores). I've cut down the file as much as I can while still retaining the error & would welcome your comments!

HVAC_Generic_Failure.txt

[drjfloyd]

I've been playing with this case trying to get it to work. First set of things I tried was:

-Your fan curve only had a couple of points which didn't include the flow that was settling out before the fire gets going. I assumed your 1500 was 95 % of the max flow and made a quadratic table (fans are generally sized to be a near the max flow so small increases in losses as ducts get dust/grime on the inside don't drastically change the volume flow).
-I simplified the HVAC inputs. I took the various ducts leaving the compartment and the ducts in the loop and combined ducts with the same diameter using WAYPOINTS to keep the same visual appearance.
-Added loss to the main compartments as 1E-4 of the wall and ceiling area and increased the losses in the smaller compartments to the same leakage fraction. (Klote and Milke's Design of Smoke Management Systems suggests 0.7E-4 as a tight value).

The case still went unstable.

I turned off the fire, and the case still went unstable but more cleanly than before. What I observed was flows reached a quasi-steady state but then Comp 6 started diverging from the other compartments and eventually went off the rails causing a crash.

I changed the fan to a constant volume flow with the quasi-steady results. That didn't resolve the issue. I'm going to see if I can remove some of the rooms and still get the crash. It is still a fairly complex network to easily be able to debug.

[GeorgeBGrant]

Jason, first thanks for all your efforts with this case. FYI, I just had to guess a fan 'curve' based on the one system operating point I was given; namely 1500 Pa & 1m3/s flowrate. The reason for the very small leakages in the non-fire enclosures is that I was trying to get the system to emulate the normal operating conditions in these volumes, where the intention is to maintain a negative pressure of between -250 to -500 Pa. I managed to get pressures consistently around -480 Pa in these spaces for the no fire case with very small leakages (around 1 cm2 or less) & simulation times up to 200 s with no instabilities. However I was aware I might be pushing it a bit as I'd seen you comments on another thread re: the valid range of leakage ratios!

[GeorgeBGrant]

Jason, quick update om my own tests here. I removed the HVAC 'loop' (Nodes 15 & 16) in case this was causing the problem but got the same crash point. Then I wondered if the proximity of the fire plume to the 'hole in the wall' in the fire compartment was just overwhelming the HVAC solver, so I moved the whole fire back along the y-axis by 7.4 m and kept my original very small leakages in the smaller process compartments. The run is currently at t= 515s after 47 hours CPU time. As the peak HRR for this case is at t~480s, I'm hoping this will now run to TEND=849s. I will confirm via this thread and then would be happy to close this issue down if you agree. Meanwhile, I'd be grateful if you could let me have the quadratic fan curve data you used in your own tests as I'd like to repeat this 2-point fan curve simulation with something a bit more realistic. Regards, George.

[drjfloyd]

I’ve continued working with this case as I have had time and available cores. We had another HVAC issue reported, and it did have a slight impact on this case but wasn’t the root cause.

I simplified your geometry and meshes to make things run faster and make it easier to play with the case.

-No OBST for exterior walls. Just VENTs where needed. This simplifies the input file a lot making it easier to work with. Grid cells outside the rooms don’t contribute to the quantities of interest and are just wasted computer resources.
-0.2 m mesh with all rooms snapped to this grid (adjusted N2 mass flux to keep mass flow the same with 0.2 x 0.2 rather than 0.1 x 0.1 inlet vents).
-Removed the fire room internal OBST. In the large room the various OBST don't have much impact on the pressure. A slight impact on rate of pressure change since they occupy volume, but they are a small fraction of the room volume and this simplification shouldn’t change the overall behavior.
-Set N2 room leakage to 1E-6 of the ceiling and wall area. This is very tight.

From your description, this case is about trying to understand the HVAC behavior which means we just need to resolve the pressure in each of the rooms. The rooms with just N2 supply you don't need much grid resolution to do this. In these rooms, we don't really care about resolving the flow field in them, and with the exterior walls at the mesh boundary there is no pressure / velocity error at the walls. Resolution can be fairly coarse and still get the pressure right. Given the size of the fire in the fire room, 0.2 m resolution should give reasonable results there especially since the main interest is the pressure caused by the fire and not the details of what is happening inside the plume.

For an HVAC system one selects the fan size to optimize efficiency at the desired operation point. A rough rule of thumb is 75 % of max pressure and 75 % of max flow. So a desired operation point of 1500 and 1 m3/s means 2000 max pressure and 1.33 m3/s max flow as a quadratic. Since this case wants to understand the system behavior in off-normal conditions, it would be best to understand the actual fan curve.

In the large room you had a bunch of OBST that you removed with a CTRL_ID to put holes in the walls. These were only a couple of grid cells across and not enough to really resolve the pressure drop. Capturing the pressure in this room is going to be key to the HVAC system performance. I don't know what these are meant to represent but I summed the area up and represented them as 8 HVAC louvers to the outside 0.6 m x 0.8 m. Representing how this room communicates with the ambient is critical to this problem. How much or how little this room pressurizes as the fire grows is going to determine whether or not smoke gets pushed into the other rooms.

Your fire eventually consumes the oxygen in the fire room. With my simplified model vitiation effects begin around 500 s which is around the time you started having issues. With the default REAC parameters (AIT=0 K), as this starts to happen the flame will start to ghost around the room looking for oxygen. This results in spikes up and down in the HRR as mixing randomly allows fuel to encounter any remaining pockets of high O2. This causes pressure spikes. The HVAC pressure solution is not fully coupled with the FDS solution. I think you may have been getting some bad feedback at this point. When significant vitiation effects occurs the bulk room temperatures are not very hot. I don’t think the default REAC parameters for reignition are going to give good results beyond that point. Post-extinction reignition is an open area of research.

Doing all this, things were stable up to and a bit beyond the start of vitiation. I think your issue was a combination of factors: the fan curve, how you were doing leakage in the large room, the fire reaching extinction, etc.

Another thing to consider in this case is you are starting with everything quiescent. It takes some time for the room pressures to settle out to the point where leakage into the small rooms equals exhaust. At first, you are just drawing a vacuum. You may want to give a short period of time at the start without of fire to let the system reach an equilibrium before starting the fire.

HVAC_Generic_Failure8.txt

[GeorgeBGrant]

Hi Jason, many thanks for your detailed observations on this case. FYI, when I moved the fire back -7.4 m in the y-direction, the model got to t=725s before it crashed yesterday. This is the furthest it's gone so far and may be well be enough for our preliminary analysis but in the background I will update the model using your above suggestions & see if I can get it to go the full way to 849s. FYI, the small leakages in the fire compartment were taken from door leakage areas used in a previous CFAST simulation of the room fire only situation. I'm happy to close this discussion down now & will do so. Best regards, George