3D Combustion Bowl Geometry Optimization Using Response Surface Modeling

The Diesel engine combustion system performance trade-off study and its complex combustion parameters make it difficult to analyze the prediction of each parameter in detail with traditional simulation and testing methods. These challenges lead to large computational time and cost. However, to facilitate the reduction of iterations in the simulation and testing stage, the DOE and Optimization can help at the design stage. The optimization simulation tool can be used to predict the overall system performance and maximize the utilization of computational, by automating the process integration and performing DOE optimization.

In this work, we intend to demonstrate the approach of Response Surface Modeling (RSM) to optimize the geometries of combustion systems from performance and emission perspective. The Gaussian Process RSM algorithm, supplemented by Uniform Latin Hypercube (ULH) and Incremental Space Filler (ISF) DOE designs, has been used to arrive at an optimized piston bowl geometry for a DI diesel engine, having the potential to perform well both at the rated power and maximum torque operating points. modeFRONTIER is used to carry out the optimization process. Three principal piston bowl parameters have been identified for optimizing the geometry: (a) Bowl diameter, (b) Bowl depth, and (c) Bowl angle, and the engine performance trade-off parameters are Indicated Mean Effective Pressure (IMEP) & NOx.

A sensitivity analysis shows the bowl diameter to be the dominant geometry parameter in influencing the Indicated Mean Effective Pressure (IMEP) of the engine. The IMEP increases with a reduced bowl diameter, but at the expense of increased NOx. Within our parameter range of investigation, bowl depth was observed to be less influential than the bowl diameter and the bowl angle was found to be the least influential of the three parameters in affecting the engine performance. Due to the strong nonlinearity of the combustion problem, the RSM method has generated the 3D RSM surface manifested in an intricate shape, highlighting the importance of an appropriate algorithm selection to minimize the prediction error. Finally, virtual optimization was carried out to achieve the target performance. In the end, the competency of the parallel coordinate chart has been shown to prove it as a smart and elegant tool for a multi-objective optimization problem.