Estimation of Urea Dosage in an SCR Unit


Publication channel


Year of the publication



Sandell, L.  Estimation of Urea Dosage in an SCR Unit. 2021.



Related to:


Impositions of increasingly stringent restrictions on the harmful emissions of nitrogen oxides (NOx) from combustion engines have popularized the use of chemical exhaust gas treatment in the form of selective catalytic reduction (SCR). This method requires the addition of ammonia (NH3) as reducing agent, although the precursor urea (CO(NH2)2) is commonly used, in order to avoid the safety concerns associated with pure NH3. Accurate urea dosing is important from both an environmental and financial perspective. A common problem in urea dosing systems is decreased dosing accuracy when the urea flow demand is low, due to unreliable urea flow rate measurements.

The objective of this thesis was therefore to develop a model-based estimator capable of accurately predicting the urea flow to the SCR. A physics-based nonlinear model was derived for Wärtsilä’s urea dosing system, while a linearized version was developed as well, to explore the possibility of using a linear estimator. This option was later ruled out, as the linearized model approximated the nonlinear model rather poorly. Consequently, the applied nonlinear estimators were based on the extended Kalman filter (EKF), and augmented with the capacity to estimate unknown inputs.

To determine the unknown model parameters, linear regression was performed on process data supplied by Wärtsilä. The achieved accuracy was unsatisfactory, in large part due to non-ideal behavior, in the form of hysteresis and deadband, exhibited by the valves in the system. Although attempts at modelling these phenomena were made, no solution was found due to time constraints. As fitting the model to data proved problematic, no extensive validation on separate datasets could be performed.

The performance of the developed estimators was promising, as they produced accurate estimates under the influence of disturbances in the flow sensor and dosing valve actuator, respectively, provided an accurate model. Before the estimator performance can be validated on actual process data, the model will nonetheless have to be improved upon, by accounting for the observed non-ideal valve behavior in an appropriate manner.