A Two-Stage Model for Sequential Engine-Out and Tailpipe Emission Estimation

This paper presents models for the estimation of vehicular NO_x emissions of gasoline-powered vehicles and presents an analysis of the performance based on real driving data. The main contribution is a two-stage model for the sequential estimation of engine-out and tailpipe emissions. This structure allows on-board operation (i.e. the computations can be performed on standard automotive ECUs) and achieves an accurate estimation performance as indicated by a statistical analysis. The estimation of engine-out emissions is based on multiple linear regressions (with a low number of parameters) using training data from driving cycle data. The test data is taken from road measurements to obtain a realistic assessment of the performance of the models under real driving conditions. The accuracy is within 3% for a cumulated error index. For the second model stage, a reduced physical model of the conversion efficiency of a catalytic converter is proposed. This stage is based on physical knowledge about typical conversion behaviour of a three-way catalytic converter. We further provide a comparison with a regression-based model of the second model stage and observe that both approaches are feasible. Both achieve an accuracy within 7% for a cumulated error index. However, the physical model performs better at detecting particular emission events, while regression-based estimation tends to average out these effects.