Unified modeling and efficient simulation of chromatographic separation processes
- Einheitliche Modellierung und effiziente Simulation chromatographischer Trennprozesse
Leweke, Samuel; Wiechert, Wolfgang (Thesis advisor); Mitsos, Alexander (Thesis advisor)
Aachen : RWTH Aachen University (2021, 2022)
Dissertation / PhD Thesis
Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2021
Liquid-solid packed-bed column chromatography is the workhorse and main cost factor of downstream processing in biopharmaceutical industry. In this context, mechanistic chromatography modeling has increasingly gained importance due to its potential for accelerating process development and reducing operating costs. This requires efficient and robust simulation software for executing typical modeling workflows such as parameter estimation or process optimization. In this work, we derive the classical chromatography model family from a high-resolution three-dimensional model and extend it to include particle size distributions, reaction terms, and multi-state binding models. These single-column models are embedded in a system of submodels to allow the description of cyclic multi-column processes. Moreover, the reaction terms enable modeling of fixed bed reactors and crystallization units, for example. We redesign the software architecture of the single-column solver CADET version 2.0.3 to accommodate the new mathematical modeling framework. In the next step, we implement several one-dimensional rate models and the two-dimensional general rate model. To this end, we detail the finite volume discretization of the partial differential equations and discuss low-level implementation details. We apply a strongly-coupled monolithic approach with implicit time stepping to handle the system of submodels. Using nested domain decomposition, the software exploits modern multi-core CPUs by multi-level task-based parallelism. In order to verify the code, we derive closed-form analytical solutions of the models in the Hankel-Laplace domain. Reference solutions are generated by numerical inversion using arbitrary precision arithmetics. Moreover, we prove error bounds of the inversion method for a large class of models to control the accuracy of the solutions. We investigate the parallel scaling behavior and compare the performance with the previous version of CADET. Compared to version 2.0.3, the new software is 29% faster on average and has a 17 percent points higher parallel efficiency when using two threads. In a multi-column benchmark, we observe an additional speedup due to multi-level parallelism. Finally, we demonstrate the capabilities of the simulation framework in several case studies. First, we investigate the effects of spatial variations of porosity and particle size distributions in a microcolumn using a two-dimensional model. Second, we compute the dynamics and the cyclic steady state of a simulated moving bed reactor for the enzymatic hydrolysis of sucrose to fructose and glucose using invertase. Last, we consider thermodynamic buffer equilibria and optimize a detailed pH gradient model.