Parameter Set Reduction and Ensemble Kalman Filtering for Engine Model Calibration

Abstract

A novel methodology is presented in this paper to reduce the burden of calibrating an engine model associated with a high number of parameters and nonlinear equations. The proposed idea decreases the calibration candidate parameters by detecting the most influential ones in an engine air-charge path model and then using them as a reduced parameter set for further model calibration. Since only the most influential parameters are tuned at the final calibration stage, this approach helps to avoid over-parameterization associated with tuning highly nonlinear engine models. Detection of the influential parameters is proposed using sensitivity analysis followed by principal component analysis (PCA) as an early off-line stage in the model tuning process. Then, an ensemble Kalman filter (EnKF) is used for tuning the detected influential parameters. The Jacobian-free suboptimal filtering approach of EnKF allows tuning parameters either with off-line recorded data or during on-line engine testing. Using EnKF along with parameter set reduction presents an approach for decreasing the complexity of parameter tuning for online model calibration. Results from experiments on a heavy duty diesel engine show an average of 50% improvement of the model accuracy after calibrating the engine model using the proposed reduced parameter set tuning methodology.