Advanced evolutionary learning based methods for optimal characterisation of non-linear aftertreatment technologies (ADELE)

Abstract:

The main focus of research will be on LNT (Lean NOx Trap) models. While widely used in European applications these systems are very sensitive to ageing and sulfur poisoning requiring models that can track these features together with a good accuracy in wide operating range to yield a satisfactory prediction on Real World Driving Cycles. Recently established Sulfur ageing SGB rig at Ford Research in Aachen can do complex characterization of thermal and sulfur ageing of LNT samples. The task is to establish methods and algorithms for optimal transfer of the measured data into the existing models or identify the need for specific model features that need to be added or enhanced in order to improve the model prediction quality. The models will further be validated by the data stemming from various engine dyno and vehicle tests for given catalyst formulations. The goal is to develop custom made algorithms based on genetic (evolutionary) algorithms to initially tune the existing models. In the second phase of the project algorithms will be developed that provide self-tuning structures that can describe effects beyond functionality of the current models. Finally the optimal test setup for lab experiments will be investigated in order to enable creation of high fidelity data which in turn will lead to further enhancement of model tuning.

Short description of the task performed by LARICS:

The main focus of research will be on LNT (Lean NOx Trap) models. While widely used in European applications these systems are very sensitive to ageing and sulfur poisoning requiring models that can track these features together with a good accuracy in wide operating range to yield a satisfactory prediction on Real World Driving Cycles. Recently established Sulfur ageing SGB rig at Ford Research in Aachen can do complex characterization of thermal and sulfur ageing of LNT samples. The task is to establish methods and algorithms for optimal transfer of the measured data into the existing models or identify the need for specific model features that need to be added or enhanced in order to improve the model prediction quality. The models will further be validated by the data stemming from various engine dyno and vehicle tests for given catalyst formulations. The goal is to develop custom made algorithms based on genetic (evolutionary) algorithms to initially tune the existing models. In the second phase of the project algorithms will be developed that provide self-tuning structures that can describe effects beyond functionality of the current models. Finally the optimal test setup for lab experiments will be investigated in order to enable creation of high fidelity data which in turn will lead to further enhancement of model tuning.

Duration:
February 2016 – January 2019

Principal investigator:
Prof.dr.sc. Zdenko Kovačić