Development and Validation of ANN Models for a SCR Catalyst

This thesis focuses on the imperative task of developing and validating an Artificial Neural Network (ANN) model for the simulation of Selective Catalytic Reduction (SCR) systems, crucial components employed to mitigate nitrogen oxide (NOx) emissions in vehicular exhaust gases. The escalating global emphasis on environmental conservation, particularly in reducing NOx emissions, necessitates innovative methodologies to comply with stringent regulatory standards. In response to this challenge, we propose a novel approach utilizing data-driven techniques, specifically neural networks, to model the intricate dynamics of SCR reactors. The modeling effort is divided into two essential categories: Thermal Modeling and Chemical Modeling. The former involves Wall Temperature and Outlet Gas Temperature ANN, while the latter encompasses Storage, NH3 Slip, and Conversion Efficiency ANN. The workflow for each category includes Design of Experiments (DoE) for data generation using the GT-SUITE plant model, followed by the training and validation of the ANN with data derived from the DoE. Subsequently, the ANN is rigorously tested on transient cycles, with results compared against the GT-SUITE model. This comprehensive approach ensures a holistic simulation, capturing both thermal and chemical aspects of the SCR system. The methodology's effectiveness is validated through systematic testing, providing insights into the ANN's performance across varying conditions. In conclusion, this thesis presents a robust framework for SCR simulation, marrying the strengths of data-driven neural networks with the intricacies of thermal and chemical dynamics. The results underscore the potential of this approach as a key enabler in developing an alternative to model-based controller algorithms.