Kodavasal, J., Pei, Y., Harms, K., Ciatti, S. et al., "Global Sensitivity Analysis of a Gasoline Compression Ignition Engine Simulation with Multiple Targets on an IBM Blue Gene/Q Supercomputer," SAE Technical Paper 2016-01-0602, 2016, doi:10.4271/2016-01-0602.
In internal combustion engine computational fluid dynamics (CFD) simulations, uncertainties arise from various sources, such as estimates of model parameters, experimental boundary conditions, estimates of chemical kinetic rates, etc. These uncertainties propagate through the model and may result in discrepancies compared to experimental measurements. The relative importance of the various sources of uncertainty can be quantified by performing a sensitivity analysis. In this work, global sensitivity analysis (GSA) was applied to engine CFD simulations of a low-temperature combustion concept called gasoline compression ignition, to understand the influence of experimental measurement uncertainties from various sources on specific targets of interest-spray penetration, ignition timing, combustion phasing, combustion duration, and emissions. The sensitivity of these targets was evaluated with respect to imposed uncertainties in experimental boundary conditions and fuel properties. In the present study, the sensitivity of the targets to uncertainties in CFD model parameters and chemical kinetic rates was not studied. Closed-cycle CFD simulations were performed using a 1/7th cylinder sector mesh representative of a four-cylinder, 1.9 L, multi-cylinder diesel engine modified to run on gasoline, under compression ignition. Several inputs to the CFD model (in terms of experimental boundary conditions and fuel properties) were chosen, and uncertainties were identified for each of these. The values of these inputs were then perturbed using the Monte Carlo method to generate a set of CFD simulations. These simulations were run in multiple ensembles, with multiple restarts. Each case was run on 64 cores, with 128 simulations running concurrently per ensemble on half a rack (8192 cores) of an IBM Blue Gene/Q supercomputer. Multiple 128-simulation ensembles were run simultaneously. Based on the GSA applied to the results of these simulations, critical experimental inputs to the CFD model in the context of a low temperature combustion engine simulation are identified for the relevant targets.