A generalized framework for multi-scale simulation of complex crystallization processes

  • Das verallgemeinerte Rahmenwerk für mehrskalige Simulation komplexer Kristallisationsprozesse

Kulikov, Viacheslav; Marquardt, Wolfgang (Thesis advisor)

Aachen : Publikationsserver der RWTH Aachen University (2011)
Dissertation / PhD Thesis

Aachen, Techn. Hochsch., Diss., 2010


The objective of the presented thesis is the development of a software-technical and algorithmic solution for the dynamic simulation of complex multi-scale problems in the field of crystallization and fluid dynamic process modeling. In the thesis, all aspects of the problem solution are considered. The proposed solution is based on the representation of the complex problem in the form of a generalized process flowsheet. This flowsheet is solved by the specialized software simulation tools coupled by means of an integration platform CHEOPS. CHEOPS supports representation of the process flowsheet and includes the algorithms for the flowsheet simulation. The units of the flowsheet (usually, the apparatuses) are represented by externally stored mathematical models and solved by the simulation software. To enable integration into the flowsheet model, interfaces to a number of external software tools such as FLUENT, Parsival, gPROMS, MATLAB and HYSYS have been implemented in CHEOPS. The modular dynamic simulation algorithm for the solution of the flowsheet problem was developed and tested first on the illustrative example, which represents a crystallization process flowsheet. The developed coupled simulation approach is further applied to the solution of the multi-scale problem, which involves a fluid dynamics subproblem and crystallization subproblem described with the population balance and the crystallization kinetics. This multi-scale problem is represented as a generalized flowsheet, where process phenomena are represented as flowsheet units. Different decomposition options and choices of the coupling variables to be transferred between the subproblems are analyzed. As the considered phenomena have different scales, discretization grids for the individual subproblems have to be chosen. The problem decomposition is performed such that for each subproblem, the best matching spatial grid is determined. The fine spatial grid is introduced for the fluid dynamics, and the coarse grid (compartments) is introduced for the crystallization subproblem. Scale integration techniques to bridge between the grids are implemented and evaluated. The error sources in the coupled simulation are discussed and the problems that arise in the error estimation are formulated. The method was successfully applied to an illustrative example, for which the validation using a reduced approach (Method of Moments) was possible, and the errors can be evaluated. It was found that the two major causes of deviation from the reference solution are inconsistencies in the problem formulation between the subproblems, which cannot always be avoided, and the choice of the coarse grid, which introduces discretization error for the quantities within the compartments. Further development of the method was done by introducing a compartment adaptation, where the compartment boundaries are adjusted according to specified criteria during runtime using the adaptation procedure developed in the study. Simulations with adaptation were performed for different choices of criteria. The adaptation method showed ambiguous results depending on the choice of criteria. In particular, the predictions improved when both the kinetics and the residence times were accounted for. The developed generalized flowsheet method was applied for the complex case study where the crystallization and the fluid dynamics models were solved for a lab-scale crystallizer and state-of-the-art models of the process kinetics, taken from the literature. The method succeeded to simulate this model as a generalized flowsheet and can be used for the other problems with similar complexity. However, due to large differences of time scales of the subproblems, the simulation time was large, thus the model solution was found to be dependent on small disturbances, and the simulation accuracy was insufficient.