This use cases was peformed in collaboration with Convergent Science. High fidelity combustion simulations, like Converge CFD, take significant computational resources. This creates challenges particularly when optimizing under uncertainties:

1) Minimize number of simulations

2) Reduce the time needed to get results

3) Generate accurate probabilistic analysis without extensive simulations

A diesel combustion process optimization comparison was done using a previously successful genetic algorithm and a novel design space reduction method based on SmartUQ’s emulators.

The SmartUQ method used adaptive DOE techniques and emulation to provide optimal sampling patterns in regions with a high probability of containing optimal points.

The Adaptive DOE based method reduced the 3 dimension design space by over 97% with 112 pts (2 iterations) and a good optima was found with 232 pts (5 iterations). The genetic algorithm identified the best optima but took 568 design points and 71 iterations (i.e. generations). This project demonstrates the promise of hybrid techniques combining the advantages in simulation efficiency and run time of the adaptive DOE based method with other methods like genetic algorithms.