Development of a reduced-order design/optimization tool for automotive engines using massively parallel computing

Design and optimization of automotive engines present unique challenges on account of the large design space and conflicting constraints. A notable example of such a problem is optimizing the fuel consumption and reducing emissions over the drive cycle of an automotive engine. There are over twenty design variables (including operating conditions and geometry) for the abovementioned problem. Conducting design, analyses, and optimization studies over such a large parametric space presents a serious computational challenge. The large design parameter space precludes the use of detailed numerical or experimental investigations. Physics-based reduced-order models can be used effectively in the design and optimization of such problems. Since a typical drive cycle is represented by 1500 to 2000 sample data points (engine cycles), it is essential to develop fast and robust computations so that the entire engine cycle computation is done close to real-time speeds (on the order of 100-150 milliseconds). Harnessing the power of high-performance computing, it is possible to perform optimization of automotive drive cycles using massively parallel computations. In this work, we discuss the development of a parallel fast and robust reduced-order modeling tool to compute integrated quantities such as fuel consumption and emissions (NO and CO) over a range of engine drive cycles. As an illustrative example, we perform a massively parallel simulation consisting of 4096 synthetic drive cycles, representative of a fleet of cars. The impact of parameters such as humidity, initial cylinder pressure, inlet air temperature, and residual gas fraction on the performance and emission are presented.