b'Automotive | Engineer Innovationinjection strategies for gasoline direct230Feasibleinjection engines.220 InfeasiblePareto Front Feasible Exemplarily, the goal of the first210 Pareto Front InfeasibleTurbulent Kinetic Energy (J/Kg)demonstrator study was to findSelected CandidatesReferenceinjection strategies that maximize the200 Best Feasiblecombustion efficiency of a genericBest Infeasiblegasoline engine operated under full190load stoichiometric homogeneous charge conditions at 4,000 revolutions180per minute. The derived targets to 170achieve this goal were to maximize both turbulent kinetic energy and fuel160distribution homogeneity in the combustion chamber at ignition timing,1500,8 0,85 0,9 0,95 1thereby ensuring an ignitable mixture atEquivalence Ratio Uniformity Index (-)the spark plug. Further constraints wereFigure 1: Trade-off between turbulent kinetic energy and fuel distribution homogeneity at specified for the maximum fraction ofignition timing for 80 different multi-pulse injection strategies fuel flowing back into the intake port and non-vaporized liquid fuel residual in the chamber at ignition timing3assuming they are a major contributor2.5to particle emissions. A generic fuel injection pulse train was parametrized2allowing for any number between one to four trapezium spray pulses, with a1.5Tumble (-) Turbulent Kinetic Energy (J/Kg) Equivalence Ratio Uniformity (-)variation of injection start, respective pulse durations and intervals between1any two pulses. For each of the strategies a transient CFD simulation of0.5the mixture preparation from intake 0through compression up to ignition timing was carried out using fully600automatic in-cylinder simulation technology. Covered physics include500moving piston and valves, spray injection and vaporization, droplet-wall interaction and wall wetting as well as400the turbulent flow and air-fuel mixing. 300A tool-embedded optimization algorithm was used which is200characterized by its hybrid and adaptive nature, in other words100automatically using a blend of state-of-the art optimization0algorithms combining local and global1search strategies and tuning itself to the design space during the search.0.8This is very cost- effective in both complex and simple problems. Consequently, no optimization0.6expertise was required to conduct the study and the only input that is0.4required to the optimizer is the number of injection strategies to be assessed.0.2By this means, 80 different injection strategies were studied fully0autonomously on a high performanceFigure 2: Transient evolution of averaged tumble, turbulence and equivalence ratio uniformity computational cluster. for three selected candidate injection strategies (dotted lines)47'