Input parameter tuning of 3D biodiesel engine simulation using parallel surrogate optimization algorithm

摘要

Successful simulation of the 3D biodiesel engine relies on the accurate input-parameter tuning of the 3D platform named KIVA4-CHEMKIN, which was time- and resource-consuming through traditional uncertainty analyses or experimental design methods. In this study, the input-parameter tuning was treated as an optimization procedure to minimize the root-mean-square error between the simulated in-cylinder pressure and the experimental in-cylinder pressure. A parallel time-varying hyperparameter surrogate algorithm with a radial basis function was proposed to achieve the goal of a favorable solution and less computation time. The key input parameters-the start of injection, injection duration, Sauter mean radius, fuel temperature, and in-cylinder temperature at the intake valve closure-were tuned within their feasible ranges. A 3D KIVA4-CHEMKIN model, involving a skeletal mechanism of 112 species and 498 reactions, was used to test the effectiveness of the 3D biodiesel engine simulation. Only seven iterations with a total of 84 cases could achieve a favorable solution under a parallel paradigm. The possible limitation of the proposed algorithm lies in the mandatory requirement of parallel computing resources. The parameter tuning had an appreciable impact on the estimation of the in-cylinder pressure. The resultant observation-a delayed start of injection with a short injection duration-helps to produce a better fitting between the simulated pressure curve and the experimental results.