[s.chen8]

Hello all,

I met some unexpected situation while doing some IC engine cycle simulation. Remaining all the setup unchanged(including embedding strategies, combustin model, emission model, etc,.) and change the base grid size only, everything went smooth when I switch from 4mm, 3mm and 2mm. The final outputs were linear ly converging and the difference getting smaller.

However, while I tried 1.5mm as base grid size, the output was diverged and showing some different trends. The engine is a common 1.5L NA gasoline inline-4 engine, 73mm bore, running at 3000rpm WOT, 23.77% EGR with stoichimetric condition. I was using 58 threads (29 cores), 3.3 GHz.

My question is: Does it happen before while changing base grid size? OR Does it relate with computer multi-core capability? Or does it relate with AMR distribution while changing base grid size?

Thank you!

[nitesh.attal]

Originally Posted by s.chen8

Hello all,
I met some unexpected situation while doing some IC engine cycle simulation. Remaining all the setup unchanged(including embedding strategies, combustin model, emission model, etc,.) and change the base grid size only, everything went smooth when I switch from 4mm, 3mm and 2mm. The final outputs were linear ly converging and the difference getting smaller.

However, while I tried 1.5mm as base grid size, the output was diverged and showing some different trends. The engine is a common 1.5L NA gasoline inline-4 engine, 73mm bore, running at 3000rpm WOT, 23.77% EGR with stoichimetric condition. I was using 58 threads (29 cores), 3.3 GHz.

My question is: Does it happen before while changing base grid size? OR Does it relate with computer multi-core capability? Or does it relate with AMR distribution while changing base grid size?

Thank you!

Hello,



If your engine is susceptible to Cycle-to-Cycle variability, then this is something that can be captured by the code. Slight perturbation (due to restart, number of cores, or change in random number seed) may lead to a different evolution path and hence a different result.


Few considerations below,


1) If you have parcels in the simulation, you would have to increase the number of parcels as you refine the mesh. This is because you do not want the parcels to overshadow the mass in a given cell as this will lead to incorrect momentum transfer.

2) To reduce CCV, you can use repeatable setting in the solver (1st order and low monotone tolerance)
3) If the engine exhibits CCV then you would have to run multiple cycle to really gauge the performance of the simulation, the experiments and simulations in this case should be statistically in agreement.


Thanks,

[s.chen8]

Hello Nitesh,


Thank you very much for your reply.

Actually the model is running in one cycle instead of multiple consecutive cycles. In this way CCV is not considered in this case.

My question is still the same: without changing anything except the base grid size, the final result of base-grid 1.5mm is divergent from base-grid 2mm (besides the max pressure, the max APHR, the entire combustion events have been enhanced unexpectedly compared with 2mm, 3mm, 4mm).

the max embedding level in this model is the sphere embedding which applied around spark-plug: level 5 with 2mm embedding radius and level 4 with 4mm embedding radius. What's more, a level 3 AMR embedding of velocity and temperature is also applied inside cylinder during combustion happening (temperature embedding from -17.4 to 140, velocity embedding permenant).

I notice that there was a paper issued by Converge:
'Modeling Turbulent Combustion Using a RANS Model, Detailed Chemistry, and Adaptive Mesh Refinement' (SAE 2014-01-1116). This paper indicated that while using RANS model, the smallest scale of element should be no less than 0.1mm. So, Does it relate to RANS because of the unappropriate small element?


Please check out the attachment file, which is a snip picture of PV diagram. As observed, the base-grid=1.5mm pressure trace is showing a faster combustion (a shorter combustion duration although a slightly lower max pressure than the 2mm ones) than any other case .

Thank you and all the best.

Originally Posted by nitesh.attal

Hello,



If your engine is susceptible to Cycle-to-Cycle variability, then this is something that can be captured by the code. Slight perturbation (due to restart, number of cores, or change in random number seed) may lead to a different evolution path and hence a different result.


Few considerations below,


1) If you have parcels in the simulation, you would have to increase the number of parcels as you refine the mesh. This is because you do not want the parcels to overshadow the mass in a given cell as this will lead to incorrect momentum transfer.

2) To reduce CCV, you can use repeatable setting in the solver (1st order and low monotone tolerance)
3) If the engine exhibits CCV then you would have to run multiple cycle to really gauge the performance of the simulation, the experiments and simulations in this case should be statistically in agreement.


Thanks,

[nitesh.attal]

Originally Posted by s.chen8

Hello Nitesh,


Thank you very much for your reply.

Actually the model is running in one cycle instead of multiple consecutive cycles. In this way CCV is not considered in this case.

My question is still the same: without changing anything except the base grid size, the final result of base-grid 1.5mm is divergent from base-grid 2mm (besides the max pressure, the max APHR, the entire combustion events have been enhanced unexpectedly compared with 2mm, 3mm, 4mm).

the max embedding level in this model is the sphere embedding which applied around spark-plug: level 5 with 2mm embedding radius and level 4 with 4mm embedding radius. What's more, a level 3 AMR embedding of velocity and temperature is also applied inside cylinder during combustion happening (temperature embedding from -17.4 to 140, velocity embedding permenant).

I notice that there was a paper issued by Converge:
'Modeling Turbulent Combustion Using a RANS Model, Detailed Chemistry, and Adaptive Mesh Refinement' (SAE 2014-01-1116). This paper indicated that while using RANS model, the smallest scale of element should be no less than 0.1mm. So, Does it relate to RANS because of the unappropriate small element?


Please check out the attachment file, which is a snip picture of PV diagram. As observed, the base-grid=1.5mm pressure trace is showing a faster combustion (a shorter combustion duration although a slightly lower max pressure than the 2mm ones) than any other case .

Thank you and all the best.

Hi,



What you are showing is the effect of grid resolution on your simulation. Looks like the simulation approaches grid convergence at 2mm base grid as these results are very similar to 1.5mm base grid. (assuming some CCV, so you will have to run multiple cycles to really tell).
Please take a look at length scale reported in turbulence.out (or region specific) and compare that to the grid resolution you have to see if you are indeed resolving all the scales.


Thanks,

[MFGT]

Originally Posted by s.chen8

The engine is a common 1.5L NA gasoline inline-4 engine, 73mm bore, running at 3000rpm WOT, 23.77% EGR with stoichimetric condition. I was using 58 threads (29 cores), 3.3 GHz.

just two thoughts from my side:


a) 23.77% EGR is a lot, what combustion model do you use? ECFM, G-Equation or SAGE? If SAGE, what mechanism?


b) 58 threads on 29 cores, have you tried running without HyperThreading?

[s.chen8]

Hello Tobias,

Thanks for the interest.

Actually 23.77% EGR is a lot for IC engine but not the limit of Nowadays. This engine is in production already. I am running the model with SAGE, and mechanism from Prof. Ming Jia.

I have not try running without hyperthread yet since ddl is approaching. It is pretty much sure that the calculating speed will be definitely slower without hyperthread.

Originally Posted by MFGT

just two thoughts from my side:


a) 23.77% EGR is a lot, what combustion model do you use? ECFM, G-Equation or SAGE? If SAGE, what mechanism?


b) 58 threads on 29 cores, have you tried running without HyperThreading?

[MFGT]

Hi,

a) sure thats not the limit, but I am asking since my experience was that the standard Jia PRF mechanism (48 species, 152 reactions) from the example cases failed with EGR rates higher than ~10%.
In this cases i switched to ECFM with tabulated laminar flame speeds of a more detailed mechanism with 335 species and 1610 reactions.


b) I am running my simulations on a 2x16 Dual CPU server, without HT enabled. I didnt test it with 64threads, but allways thought that it wouldn't be beneficial.

[s.chen8]

Hi Tobias,

Thanks for the reply.

I wonder the reference about stating the EGR limit of Jia PRF mechanism since I have not met any trouble of converging during calculation, including any sign of recovering.

Also, I double checked the SAE paper 2013-01-1098 'Gasoline Combustion Modeling of Direct and Port-Fuel Injected Engines using a Reduced Chemical
Mechanism'. The results showed that Jia's might not suitable for low pressure combustion but no mention of EGR or dilution.

The original paper which issued Jia's mechanism: 'Enhancement on a Skeletal Kinetic Model for Primary Reference Fuel Oxidation by Using a Semidecoupling Methodology', did not mention EGR limit either.


I may try to disable Hyperthread and run it for trial testing. It is interesting that for Intel I-series CPU(such as I9,I7), the core-speed could be higher than Intel Xeon series during multi-core running condition. The only problem is that I-series motherboard does not support dual CPU setup as same as Xeon series.

Originally Posted by MFGT

Hi,

a) sure thats not the limit, but I am asking since my experience was that the standard Jia PRF mechanism (48 species, 152 reactions) from the example cases failed with EGR rates higher than ~10%.
In this cases i switched to ECFM with tabulated laminar flame speeds of a more detailed mechanism with 335 species and 1610 reactions.


b) I am running my simulations on a 2x16 Dual CPU server, without HT enabled. I didnt test it with 64threads, but allways thought that it wouldn't be beneficial.

[MFGT]

Originally Posted by s.chen8

I wonder the reference about stating the EGR limit of Jia PRF mechanism since I have not met any trouble of converging during calculation, including any sign of recovering.

I might be wrong, so I will do a test for loads about 12bar BMEP

[s.chen8]

Hello Tobias,

Please let me know if you got any update.

I contacted Prof. Ming Jia personally about this EGR limit question. He replied that his mechanism had not considered heavy EGR condition yet, back to 2013 when he published this mechanism. What he suggested is that, since the mechanism may not perform precisely under heavy EGR condition, it is better to calibrate absolute tolerance and relative tolerance setup in SAGE based on some given experimental data.

Hope this helpful.
Cheers.

Originally Posted by MFGT

I might be wrong, so I will do a test for loads about 12bar BMEP

[MFGT]

Hey,



well, I am surprised my testcase went well.
20% external EGR burned with passive pre-chamber and default Jia mechanism, just had to adjust the spark timing a bit.


Maybe (but I dont think so) something has been improved from 2.4 to 3.0 because I definetly have old simulations that burned very poorly with 20% EGR.

[s.chen8]

Hello Tobias,

20% EGR with passive pre-chamber... that's amazing...

How much ignition energy did you apply on the spark? I was doing active PC with 40mJ, lambda=1.6 with no EGR and combustion went fine. I have not try any EGR test on either passive or active PC before, since Mahle's paper emphasized how EGR could easily ruin PC performance (especially passive ones).

Let me know if you want to share some info. We should consider open another thread about pre-chamber.

Cheers.

Originally Posted by MFGT

Hey,



well, I am surprised my testcase went well.
20% external EGR burned with passive pre-chamber and default Jia mechanism, just had to adjust the spark timing a bit.


Maybe (but I dont think so) something has been improved from 2.4 to 3.0 because I definetly have old simulations that burned very poorly with 20% EGR.

[MFGT]

I kept the original two times 30mJ.


For active Pre-Chamber I also had a similar loadpoint, Lambda 1.6. Here I also went up to Lambda > 2 and it worked quite well.


Maybe we should open a Pre-Chamber thread