Multiphase Modelling of SCR Systems: Using the Taguchi Method for Mixer Optimisation. 2017-26-0113

Selective catalytic reduction (SCR) systems have become the preferred technology to deal with NOx emissions in Diesel engines. Their efficiency is highly reliant, among other factors, on the uniformity of distribution - known as Uniformity Index (UI) - of NH₃ which is injected into the system through a urea-water solution (UWS). SCR system make use of a mixer component designed to achieve the desired UI levels. However, the great variety of exhaust systems, makes it impossible to employ a universal solution. Therefore, each SCR system requires of a tailor made mixer, capable of achieving the required UI, while preventing urea crystallisation and minimising pressure drops. Computer fluid dynamics (CFD) tools together with optimisation techniques based on the design of experiments (DoE) can be used to obtain the appropriate mixer design.

This study presents a Taguchi (L4) DoE for a series of multi-phase numerical analyses, computed with the commercial software AVL Fire 2014 v, to measure the impact of three factors - the mixer blade’s number (X1) and angle (X2), and the distance between the injection point and the mixer (X3) - in the performance of an SCR system, specifically NH₃ UI, Midcone pressure drop and Wall film formation. The analysis consisted of a completely mixed turbulent flow, solved using a two equations turbulence model (k-z). The interaction of the injected particles was solved with an Euler/Lagrange approach, the liquid phase calculation was based on the statistical Discrete Droplet Method interacting with the numerical solution of the conservation equations of the flow pattern.

It was found that blade angle had a major impact on performance, the impact of number of blades was dependent on the blades angle, while the distance from the injector had minimum impact in performance apart from Midcone pressure drop, where a mixer closer to the injector helped to reduce pressure drops. Moreover, a linear regression model was used to predict the output of three Cases (A, B, and C), finding good agreement between the predicted results and the CFD calculated performance.