A Time-Saving Methodology for Optimizing a Compression Ignition Engine to Reduce Fuel Consumption through Machine Learning

Applying a suitable design optimization technique is a crucial task for optimizing compression ignition engines because of the time-consuming process of optimization even with advanced supercomputers. Traditional computational fluid dynamics (CFD) used in conjunction with design of experiment (DOE) methods requires executing the CFD model several times. A response surface is usually fitted to relate the inputs to the outputs, which is often created based on linear regression. This method is not well suited to capture interaction effects between inputs and nonlinearities existing during engine combustion. A combination of genetic algorithm (GA) and CFD tools usually eventuates better optimum results. However, the CFD simulations must be executed sequentially, resulting in extremely high computational times, which makes it impossible to apply an optimization study using a single desktop computer.

The current study examines a novel approach, which combines CFD, GA, and a type of machine learning approach, namely artificial neural networks (ANNs), in order to optimize a compression ignition engine to achieve its minimum indicated specific fuel consumption (ISFC). Start of injection (SOI) timing and input pressure were selected as the optimization variables in order to investigate improvement in ISFC without any hardware modifications of the engine. Maximum in-cylinder peak pressure and ringing intensity (pressure rise rate) were chosen as the optimization constraints. Conducting a reliable optimization study with a single desktop computer in a shorter time can be achieved by using the proposed methodology. The results indicate that a 97% decrease in the estimated number of days to achieve the final results was obtained, compared to the traditional CFD-GA approach. Furthermore, adopting this methodology eliminates the necessity for additional response surface fitting to GA data. Therefore, it facilitates an examination of design parameter effects on the engine outputs, doing sensitivity analysis, post-processing the optimization results, and providing a powerful tool to gain optimum designs. The final optimum point illustrates a 10% improvement in ISFC, while avoiding sensitive regions and without exceeding optimization constraints.