Multi-Objective Stochastic Bayesian Optimization for Iterative Engine Calibration

Abstract

Engine calibration is an important step to achieve optimal engine performance with satisfactory emissions and it is an expensive process in general. In recent years, a new process called Bayesian optimization has come into picture for reducing expensive function evaluations. It efficiently performs exploration-exploitation in design space to identify optimal region. But the work is mostly focused on deterministic case. Unfortunately, practical system measurements almost always contain random noises. Therefore, for this research work, stochastic Bayesian optimization approach has been implemented for a multiobjective engine calibration problem with constraints. Three control parameters: variable geometry turbocharger (VGT) position, exhaust gas re-circulation (EGR) valve position, and start of injection(SOI) are calibrated to get a trade-off (Pareto) curve between engine fuel consumption (BSFC) and its emissions (NO x ). Simulations are performed at different noise level to validate the effectiveness of the proposed algorithm at adverse conditions. Promising results are obtained at all noise levels with optimal solutions near actual pareto front.