Due to the stringent emission regulations for diesel engines, increased attention has been dedicated to after-treatment technologies. The main problem with diesel engines is the significant emissions of nitrogen oxides (NOx). Two technologies have been established as main solutions to reduce NOx emissions under lean conditions.
These are the selective catalytic reduction (SCR) catalysts and the lean NOx trap (LNT) catalysts. While SCR catalysts are becoming prevalent as a solution to the NOx problem LNT, when used in tandem with SCR, they hold great promise for decreasing running costs and improving the performance of exhaust aftertreatment solutions.
An LNT catalyst is a complex nonlinear chemical system designed for cyclic operations, which can be very difficult to model in detail, and a proper model is necessary for good control due to inability to measure key catalyst states.
In the ADELE project, we are using a simplified model, based on the main chemical and physical processes typical to LNT. We are then tuning the model parameters using multiple types of input data and constrained multi-objective genetic algorithms to generate models that will behave well on real world data while respecting expected model behaviour.
The behaviour of an LNT catalyst changes drastically during its lifecycle due to thermal aging and sulfur poisoning. These effects influence the usefulness of the system if left unmodeled. We plan to compensate these effects by using the tuning method on differently aged systems in order to deduce the optimal way to include such effects on the model parameters.
We also plan to investigate the use of deep learning techniques such as recurrent neural networks to model and improve this system.
Project partner: Ford Research And Innovation Center, Aachen, Germany