Investigation about predictive accuracy of empirical engine models using design of experiments

This study focuses on improvement of the predictive accuracy of empirical engine models using the Model Base Calibration (MBC) method. This research discusses the effects of the number of measurement points on the accuracy of models for different Design of Experiments (DoE) by using a direct-injection 4-cylinder diesel engine. The results show that the predictive accuracy of the models converges on fixed values when the number of measurement points is increased in Latin Hypercube Sampling (LHS) and D-Optimal Design. This is because the probability density distribution of the measurement data has little variation as the number of measurement points increases. Comparing LHS and D-Optimal indicates that D-Optimal displays a higher level of accuracy, it is able to extend the boundary model because of its greater number of measurement points at the boundaries of the boundary model. In addition, it is possible to predict the fuel consumption when empirical engine models are used in simulations of the New European Driving Cycle (NEDC) under hot conditions and cold start conditions.