Virtual Sensor Design of Particulate and Nitric Oxide Emissions in a DI Diesel Engine

As a physical description of the emissions of a specific engine is seldom possible, we present here a method to design an online dynamic estimator for PM and NOx based on data. The design method is based on a systematic search of function candidates performed using genetic programming after data have been pre-treated in an adequate fashion. While data and a simple data pretreatment prove enough for NOx, some basic physical understanding is necessary to preset the method and obtain the required precision in the case of PM. The method has been applied for raw emissions of a production DI diesel engine and shows a remarkable prediction performance. While the method is not able to replace the insight in the design process given by physical understanding, the authors expect substantial advantages in all those cases in which the prediction of overall behavior is required, as virtual sensors, and also expect that the suitable introduction of additional physical knowledge can strongly improve the results.