NOx Prediction in Diesel Engines for Aftertreatment Control

Abstract

Modern Diesel engines are faced with two major emission challenges in their quest to become an environmentally compatible source of motive power, Nitrogen Oxides (NOx) and Particulate Matter (PM). Advanced techniques, such as High Pressure Common Rail (HPCR) fuel injection combined with multiple injections per cycle, are commonly employed to minimize in-cylinder production of NOx and PM. However, to meet the EPA mandated standards it is essential that an aftertreatment system be used. Typical Diesel aftertreatment systems will employ some form of a NOx reducing catalyst and a particulate trap for PM removal. Lean NOx traps and Selective Catalytic Reduction (SCR) are examples of aftertreatment techniques frequently used in Diesel engine applications. Whatever the method of choice, knowledge of the feed-gas NOx concentration is essential for not only assessing the performance of the NOx reduction catalyst but also for defining the control strategy for the aftertreatment system with respect to the management of the reductant quantity to be injected. In the absence of a dynamic NOx emission model the control algorithm has to depend on either a NOx sensor upstream of the catalyst or a static map of the feedgas NOx level as some function of engine influence factors. While NOx sensors add to the overall system cost, creating an accurate and representative NOx map over the entire engine operating range can be a challenging task. A dynamic NOx model would, in theory solve, both of these problems, however it is essential that the model be simple and implementable in real time. A model that uses inputs that are not available from the standard measurement set is of little use for real time control applications as is a model that predicts the temporal and spatial NOx evolution in the engine combustion chamber as such models tend to be computationally expensive. However, it is essential that the model behave like a fast NOx sensor in predicting cycle averaged NOx emission. In this paper we present an approach to developing such a model and present results from model validation against vehicle data. The basic structure of the model relies on well-known mechanisms that describe the NOx creation and decomposition chemical kinetics. Simplifying assumptions are made to allow available measurements to be used as inputs to the model. This leads to a parametric model where the unknown parameters are estimated using Nelder Mead optimization routine available in Matlab®. Model validation against vehicle data is also presented.