Engine Calibration Optimization Based on its Surrogate Models

Abstract

Diesel engines are becoming increasingly complex to control and calibrate with the desire of improving fuel economy and reducing emissions (NOx and Soot) due to global warming and energy usage. With ever increased control features, it is becoming more and more difficult to calibrate engine control parameters using the traditional engine mapping based methods due to unreasonable calibration time required. Therefore, this research focuses on the problem of performing engine calibration within a limited budget by efficiently optimizing three control parameters: namely variable geometry turbocharger (VGT) position, exhaust gas recirculation (EGR) valve position, and start of injection (SOI). Engine performance in terms of fuel consumption (BSFC) and emissions (NOX) are considered as objective function here with the constraint on boost pressure and engine load (BMEP). Since the engine calibration process requires a large number of high-fidelity evaluations, surrogate modeling methods are used to perform calibration quickly with a significantly reduced computational budget. Kriging metamodeling is used for this work with Expected Improvement (EI) as acquisition function. Results show more than 60% decrease in computational cost with results close to actual near Pareto optimal set.