Diesel engine optimization with multi-objective performance characteristics by non-evolutionary Nelder-Mead algorithm: Sobol sequence and Latin hypercube sampling methods comparison in DoE process

This research deals with the diesel engine optimization by Nelder-Mead algorithm and initial points distribution done by Sobol sequence and Latin hypercube sampling methods. Nelder-Mead algorithm is a non-evolutionary algorithm which needs some initial points in order to start the optimization process and then understand the relationship between input parameters and output objective functions. In this study, these points are produced once by Sobol sequence and once by Latin Hypercube in order to make a comparison between these two sequences and investigate the effect of sequence on the results. The input parameters are Da (outer bowl diameter), Dm (bowl middle diameter), Tm (bowl center depth), nozzle hole half outer cone angle and nozzle hole outer diameter, and objective functions are combustion noise, swirl and indicated torque. Final results show that although in both cases the results are close to each other, in Sobol mode it takes just 7 RunIDs (Run Identification) to find the solution but in Latin Hypercube method the algorithm needs 27 RunIDs to find the solution. In both cases almost the combustion noise changes just 0.2%, swirl improves 20.4% and the indicated torque increases about 7.65%.