Reduced Order Models For Exhaust AfterTreatment Systems - Coupling Modeling, Simulations and Chemometrics

Doktorsavhandling, 2024

The use of hydrocarbon based fuels and the high temperatures generated during combustion processes are major sources of gaseous pollutants that are detrimental to human health and the environment. Emission legislation is increasingly becoming stringent to mitigate the harmful effects of emissions. Exhaust Aftertreatment systems are a group of catalytic devices that convert these harmful emissions into products like carbon dioxide and nitrogen. Space limitation in the exhaust line creates nonuniform flow in terms of flow through bends and dead volumes. This limits the performance of the EATS. The flow from the engine does not proceed uniformly to the EATS, creating a flow maldistribution at the inlet of the EATS. Consequently, velocity, temperature, and concentrations at the exit are influenced by the flow distribution at the inlet.
Accurate models that capture the spatial and temporal variations of the flow distribution in EATS, specifically during cold start conditions and real driving emissions (RDE) tests, are essential to comply with stringent emission standards. 1D and 3D-computational fluid dynamics (3D-CFD) models are used to predict the conversion of species at the exit of EATS. While 1D models are robust, they lack accuracy, whereas, 3D-CFD models offer higher accuracy, but require significant computational resources. This study addresses these challenges primarily through Computational Fluid Dynamics (CFD) simulations and proposes a methodology for developing reduced-order models.
Firstly, characterizing and quantifying flow distribution in EATS under transient conditions with realistic geometry is performed using non-reactive simulations. Flow uniformity index is used to characterize the extent of variation of the flow parameters in any catalyst plane. In addition to the uniformity index, contours and histograms are employed to demonstrate the non-uniform flow field.
The effect of inlet pulsations on the mixed cup conversion at exit of catalyst is studied using transient reactive simulations. Four transient inlet profiles, viz., constant flowrate, sinusoidal, rectangular, and triangular pulse profiles, are chosen to describe the inlet pulsations. The results show that the fluctuations and pulsations in the incoming flow to a monolithic reactor in an aftertreatment system, affect both the transient response of the reactor as well as its time-averaged performance. The method of specifying the inlet boundary conditions also influences the solutions.
A methodology for developing a reduced-order model by combining physics-based CFD solutions with multivariate data analysis methods is proposed. This method is demonstrated by combining CFD solutions of transient reactive simulations on a diesel oxidation catalyst with chemometric techniques. Performance evaluations validate the efficacy of the multi-channel model over single-channel models. Computational efforts for creating the multichannel model are comparable to single-channel models, when utilizing available CFD data and coupling chemometrics analysis. This enables rigorous control applications with improved accuracy. The methodology can be extended to real-world emissions aftertreatment systems with complex geometries.
Further, predictions of species conversions in systems with flow maldistribution are made by performing steady state reactive 3D-CFD simulations and mapping the same with 1D-SCM. A pseudo-channel is envisaged that provides the same species conversion as the 3D-CFD, by formulating an objective function, which is the difference of species conversions of 3D-CFD and 1D-SCM. The error of the objective function is minimized by iteratively varying the velocity that will provide the same conversion in a 1D-SCM. The pseudo-channel model outputs agree closely with the CFD results in various steady-state and transient test cases.
Detached eddy simulations were carried out under nonreactive conditions on the geometry with bends to confirm the validity of RANS simulations, as RANS simulations are computationally more effective than DES. Flow uniformity indices were of the same order in both cases, however, DES showed fluctuations.
This thesis aims at developing reduced order models combining CFD simulations and regression and chemometric techniques. It also highlights the limitations of a single channel model in a realistic geometry case. The thesis also attempts to predict species conversion of transient reactive simulations, from the solution of steady state reactive simulations, as the former is computationally more expensive than the latter. The developed pseudo-channel model and multi-channel model can be used for realtime monitoring and control applications. These two methodologies require a computational load comparable to that of 1D models. Validation of these models using EATS experiments under transient conditions is recommended for future research.