| Description |
Combustion science and engineering depend heavily on the availability of effcient computational methods for reacting flows with detailed chemical kinetic models. Exa-scale computing will likely play a signi ficant role in addressing the challenges of designing next-generation, low-temperature, high-pressure engines and understanding the ignition chemistry of diesel fuels and renewable biofuels. Towards the development of scalable simulation methods capable of handling increasingly detailed chemical models, we utilize a nonlinear solver known as pseudotransient continuation (tc) and study computational flame diagnostic tools. We develop and extensibly implement fundamental algorithms for homogeneous reactors, di usion-reaction systems, and compressible, reactive flows. We study pseudotransient continuation as a nonlinear solver for implicit discretizations of homogeneous combustion problems with detailed chemical models. The sti ness, nonlinearity, and transiently-unstable modes of combustion chemistry are challenges to traditional Newton methods that are well met by tc. We develop and apply adaptive tc methods to autoignition problems and periodic stirred reactors with repeated ignition/extinction events and demonstrate the efficiency tc o ers relative to traditional solution techniques. We compare the efficiency of a range of physical time discretizations and rigorously analyze numerical convergence of our methods in computing slowly evolving solutions. For homogeneous reactors, we illuminate proper formulation of reactive ow problems and show some effects of state speci cation and Jacobian matrix formulation on the performance of high-order time integration. A major difficulty of complex flame simulation is the analysis and interpretation of results, which often demand a framework of conceptual models. The technique known as chemical explosive mode analysis (CEMA) is frequently utilized in studying ignition and flame stabilization mechanisms with turbulent flame simulations. With detailed analysis of steady and unsteady amelets we've studied the interaction of explosive chemical modes with di ffusion to improve CEMA as a diagnostic tool. Results indicate several improvements to CEMA, particularly for con gurations involving two-stage ignition at low temperatures. The tc-GESAT method is extended to di usion-reaction problems, an important step towards establishing its utility in reactive ow simulation. Additionally, as an alternative to direct computation of Jacobian eigenvalues, we explore principal component analysis (PCA) and nonlinear regression to parameterize the eigenvalues. PCA-based parameterization appears feasible for tc solvers as well as for flame analysis such as in CEMA. We have developed a `block-implicit,' or `point-implicit,' pseudotransient continuation method for compressible reacting flows and general advection-di usion-reaction equations. For the compressible system we employ acoustic preconditioning and local dual timestepping. Preliminary tests have shown the code's capability of solving compressible, multicomponent, multidimensional flows with combustion chemistry on parallel computers. We have implemented this throughout the code stack of the multiscale simulation group at the University of Utah. Implicit methods have been added to the Expression Library code, with automated assembly of task graph components from the implicit method. Expressive code for pointwise matrix assembly and automatic sensitivity calculation through the graph decomposition has been developed to enable extensibility for multiphysics simulation. With these modi cations the flexibility of the code stack designed for explicit integration methods is largely retained and implicit methods are available with minimal developer eff ort. |
