Sensorless Selective Catalytic Reduction Using Artificial Neural Networks for Emissions Prediction With Fuzzy Logic Control

The transportation industry is a major contributor to the increase of greenhouse gasses present in the atmosphere. With the number of automobiles increasing every year, the U.S. government has implemented several regulations to reduce the environmental impact of the transportation industry. The most recent regulations increase the Corporate Average Fuel Economy (CAFÉ) to over 50mpg by 2025. These increased fuel economy standards will save consumers money, reduce dependence on foreign oil and cut GHG emissions in half (1). In order to comply with these regulations and reduce GHG emissions, automakers are improving powertrain efficiency and diversifying their fuel sources. One way automakers are improving fleet fuel economy is by offering more efficient Compression Iginition (CI) engines. Compression ignition engines can have a 10% improvement in peak efficiency over a Spark Ignition (SI) Engine. Although CI engines have higher efficiencies, they also have higher Nitrous Oxide (NOx) emissions. One of the most effective methods for reducing NOx emissions is a Selective Catalytic Reduction (SCR) system. Current methods for reducing NOx emissions using SCR rely on two NOx sensors for close loop control. These sensors add substantial costs to the production exhaust after treatment systems. This paper presents an intelligent control technique to achieve accurate prediction of NOx emissions and closed loop control without the use of expensive on board sensors. Simulation models were created to validate two artificial neural networks that aim to replace the upstream and downstream NOx sensors. The upstream neural network was trained using dynamometer data from a General Motors 1.3l turbo diesel engine. This neural network represented NOx emissions as a function of engine speed and throttle position. The downstream ANN was created using a nonlinear statespace plant model that simulates the catalyst NOx and nh3 reaction. To control the nh3 injection into the catalyst, a Fuzzy Logic Controller (FLC) was implemented. The FLC controller had two inputs: the error function calculated from the output NOx and a predetermined NOx target as well as the predicted surface coverage from the nh3 reaction. The results from steady state and drive cycle simulations are shown. The work presented in this paper serves as a proof of concept for the sensorless SCR system that was developed as part of ERAU’s entry in EcoCAR2: Plugging Into the Future. The simulations were conducted as part of year 1 of the EcoCAR2 competition and will be further developed during years 2 and 3 on ERAU’s Series Plug-in Hybrid Electric Vehicle.