Rapid modelling and control of exhaust after-treatment systems

Exhaust after-treatment systems (EATS) are required for modern engines to meet current emission legislation. For diesel engines, the typical EATS architecture comprises a diesel oxidation catalyst (DOC), a diesel particulate filter (DPF), an urea-based selective catalytic reduction (SCR) catalyst and an ammonia slip catalyst (AMOX/ASC) in series to treat hydrocarbons (HC), particulate matter (PM), nitrous oxides (NOx) and ammonia (NH3) emissions, respectively.

Integration of the DPF and the urea-SCR catalyst in a single block (SCR-in-DPF) is an emerging technology to meet more stringent emission standards in the future. The integrated SCR-coated DPF can control NOx and PM emissions simultaneously. Integration of the two functionalities into a single device is expected to save cost, space, weight and thermal inertia.

To demonstrate the effectiveness of the technology and facilitate industry adoption, a representative system model and an appropriate control scheme need to be developed. Modelling of the SCR-coated filter system is complicated. The main challenge is how to best capture the complexity of the physical and chemical phenomena in a simplified but adequate representation. This work is focused on the development of an SCR-in-DPF model which achieves the right balance between adequacy and complexity, and which can form the basis of a control algorithm to be implemented within an engine control unit (ECU).

A numerical high-fidelity model (HFM) of the SCR-coated filter system was developed. The main modelling approach adapted an existing catalysed DPF model with SCR functionality. Fundamental changes are made to the description of the specie material balances in the channel and wall layers to deliver faster than real time implementation of the SCR filter model. The developed model also enabled complete customisation of the numerical solution approach——an option not typically available in commercial software.

To support the control development, a less complex but still reasonably accurate controls-oriented model is desirable. Two approaches were pursued for the model order reduction (MOR): a “grey-box” approach via Proper Orthogonal Decomposition (POD); and a “black box” approach via artificial neural network (ANN) function approximation. The POD delivered a reduced order model with a higher degree of fidelity to the original high-fidelity model (HFM), but with limited improvement in simulation times. The ANNs, on the other hand, delivered a reduced order model which achieved a more significant reduction in simulation times——on the scale of three orders of magnitude——but with a poorer level of fidelity to the reference HFM.

The developed SCR-coated filter model was applied in a system control application. Two model-based urea-dosing controllers were developed for the SCR units on a diesel engine EATS: a proportional controller strategy and a switching controller strategy. The proportional controller was developed based on analysis of the data generated from a Millbrook London Transport Bus (MLTB) test cycle conducted on an ADL400 bus fed into SCR-filter model. The critical system parameters with the strongest correlation with outlet NOx and NH3 emissions were identified and configured as the proportional controller inputs. The second controller was a switching controller modelled after the air-fuel ratio (AFR) control strategy of a three-way catalyst (TWC). This controller switches between low or high ANR dosage rate based on feedback of how the ratio of outlet to inlet NOx concentration compares to a given threshold value.

The performance of both controllers over the MLTB cycle demonstrated superiority to a conventional fixed ANR approach. At the optimised gain settings, the proportional controller strategy delivered up to 11% reduction in mass of outlet NOx emission, and the switching controller reduced mass of outlet NOx emission by about 5%, over the MLTB cycle without a stiff penalty of increase in NH3 emission.