Optimal Calibration of high DOF Engines Considering Fuel Economy and Combustion Stability

• Published in: IFAC Proceedings Volumes Vol. 42; no. 26; pp. 224 - 234
• Main Authors: Lee, Tae-Kyung, Kramer, Denise, Ohl, Greg, Filipi, Zoran S.
• Format: Journal Article
• Language: English
• Published: 2009
• Subjects: artificial neural network; combustion variability; fuel economy; high degree-of-freedom; Internal combustion engine; multi-objective optimization; optimal calibration; virtual sensing; high degree-of-freedom; artificial neural network; optimal calibration; fuel economy; combustion variability; Internal combustion engine; multi-objective optimization; virtual sensing
• ISSN: 1474-6670
• DOI: 10.3182/20091130-3-FR-4008.00030

Engine optimal set-point calibration is critical for achieving the full potential of the high degree-of-freedom (HDOF) engine. At part load operating conditions, fuel economy is typically selected as a calibration objective. However, to guarantee vehicle drivability and smooth engine operation with low vibration and harshness, combustion variability must be considered as an additional objective. In this paper, simulation-based optimal engine calibration methodology is proposed and demonstrated on a SI engine with dual-independent variable valve timing (di-VVT) and a charge motion control valve (CMCV), designed specifically to improve combustion stability at low loads and speeds. The combustion variability can generally be tied to burn duration, and CMCVs accelerate burning by increasing the turbulence intensity when blocked. Accurate cost function evaluations are enabled with a multi-scale simulation approach. It utilizes high-fidelity, physics-based models to fully characterize engine responses, and subsequently captures the engine behaviour with artificial neural networks (ANN). In addition, statistical regression analysis of experimental data correlates the coefficient of variation in indicated mean effective pressure (COVIMEP) with one of the main combustion indices. Hence, trained ANNs and regression equation for the COVIMEP provide fast models suitable for evaluating the cost function during large number of optimization runs. Optimal engine set-points are determined by solving the constrained optimization problem with multiple objectives over part load operating conditions. Assigning weights to the fuel consumption and the combustion variability terms in the cost function allows exploring the trade-off, and developing an implementable feed-forward strategy.