Four-valve spark-ignition engine with best-practice mesh settings

This example represents a typical full-cycle engine simulation scenario. The engine has a bore diameter of 17 cm. The geometry contains large intake manifold and exhaust manifold structures. There are four moving valves. It burns gaseous fuel and employs spark ignition and flame propagation models. The chemistry mechanism contains 33 species. There is no spray. The mesh settings follow the best practices shown in various Forte engine tutorials. The average cell count in the mesh is around one million. The simulation duration covers the full engine cycle.

The recommended number of cells per core is around 5000, which corresponds to 200 cores for the average cell count of one million. Using more than 200 cores will continue to provide wall-clock time reduction, but the reduction gradually becomes marginal and insignificant.

Four-valve direct-injection spark-ignition engine with finer mesh (on the order of 10 million cells)

This example represents a scenario where a much finer mesh (compared to best practice) is used. This DISI engine has a bore diameter of 10 cm. The geometry contains both an intake and an exhaust manifold of medium-sized volume. The engine burns gasoline and uses spark ignition and flame propagation models. The chemistry mechanism contains 59 species. The spray model is used, but the spray model is active for only a short fraction of the simulation duration and the peak number of spray parcels is low (< 5000), so it is not considered as a spray-intensive calculation.

The optimal number of cells per core is around 30000, which corresponds to 333 cores for the 10-million cell mesh. The wall-clock time can keep being reduced all the way up to 500 cores, but the time reduction after 300 cores gradually becomes marginal.

Spray combustion in a constant volume chamber

This example represents a spray intensive scenario. It simulates the injection and burning of liquid fuel into a constant volume chamber, called IQT (ignition quality tester). A 249-species chemistry mechanism is used for the combustion and species transport calculations. The average cell count is 0.6 million.

The recommended number of cells per core is 3500, which corresponds to about 170 cores for the cell count of 0.6 million. The benchmark study shows consistent and considerable wall-clock time reduction up to 200 cores. However, beyond 200 cores, the scaling performance quickly deteriorates. A calculation with further refined mesh also showed the sudden deterioration of scaling performance beyond 200 cores.

Screw compressor with best-practice mesh settings

This screw compressor contains a male rotor and a female rotor, with diameters of 13 cm and 12 cm, respectively. The lengths of the rotors are 20 cm. The geometry contains a suction port and a discharge port. The total volume of these two ports roughly equals the volume of the casing that houses the two rotors. Using best-practice mesh settings as demonstrated in the positive displacement compressor tutorial, the average cell count is 0.5 million. This is a pure gas-phase simulation using the real-gas model.

The recommended number of cells per core is 5000, which corresponds to 100 cores for the cell count of 0.5 million. For this mesh setting, beyond 100 cores, the scaling performance starts to become moderate until reaching 200 cores, where the wall-clock time starts to increase.

Liquid pump with best-practice mesh settings

This is a water pump case that contains two identical claw-shaped screw rotors, with a diameter of 12 cm. The length of the rotors is 16 cm. The geometry contains a suction port and a discharge port. The total volume of these two ports is roughly half the volume of the casing that houses the two rotors. Using best-practice mesh settings, the average cell count is 2.4 million. Compared to the screw compressor case in another example, the higher cell count in this liquid pump is mainly due to more complicated rotor shapes and more fine structures that need to be properly resolved using smaller cells. This simulation uses the mixture-based two-phase flow model.

The optimal number of cells per core is 6000, which corresponds to 400 cores for the cell count of 2.4 million. Benchmark results show that this case can scale all the way to 500 cores but the reduction in wall-clock time begins to tail off around 400 cores.