Process intensification and optimization based on computational fluid dynamics simulation: membrane reactor and fed-batch culture processes

Abstract

Process intensification refers to improvements of a process at operational, functional and/or phenomena levels. It is typically achieved by integrating multiple unit operations into one, integration of functions or phenomena, or targeted phenomena within a process, e.g., mixing and mass/heat transfer. To enhance mixing uniformity and mass/heat transfer, one approach in process intensification involves strengthening or regulating the flow pattern through the use of equipment add-ons such as baffles, stirrers, and other similar devices. To ensure efficient design and operation of these flow-enhanced processes, it is essential to possess the capability to model, simulate, and optimize complex flow patterns. Computational fluid dynamics (CFD) involves solving Navier-Stokes and other related equations numerically in order to visualize the flow fields and distributions of key intensive properties, e.g., concentrations, temperature, pressure. CFD is a handy tool for designing and optimizing flow-enhanced processes but its use is hindered by its high computational cost. Therefore, CFD is difficult to embed in traditional optimization methods or even to use in data-driven optimization's contexts. To face this challenge, this study investigates the approach of coupling CFD with Bayesian optimization (BO), which is a data-based optimization method that specifically addresses the exploration-exploitation trade-off. BO enables the optimization of 'expensive-to-evaluate' functions which allow only a small number of function evaluations. The main emphasis of this thesis will be to test the efficacy and identify/resolve the challenges of implementing the approach in the design and operation of advanced membrane reactors and industrial fermentors.

It is crucial to assess the scalability of both processes due to their sensitivity to local distributions such as concentration and flow when scaled up for increased production. This study first focuses on evaluating the scalability of a catalytic membrane reactor (CMR) with multiple tubes, where concentration polarization and temperature deviation are significant factors. The results of the scalability analysis demonstrate that incorporating baffles in the CMR for process intensification yields economic benefits by reducing the required amount of catalyst while maintaining comparable performance to conventional CMRs. However, the reactor's structural limitations impose restrictions on heat transfer, preventing infinite scale-up (imposing a limit on the scale up). With this motivation, the paper aims to optimize the reactor's structure within a scalable range, focusing on achieving the highest energy efficiency. Unlike previous studies that solely analyze hydrogen yield on a small scale, this research compares processes of different scales by setting energy efficiency as the objective function, which provides an overall measure of process performance while duly considering the heat transfer effect. The analysis reveals that the optimal solution involves a design that narrows the gap between tubes and reduces their diameter, enhancing the heat transfer effect despite accommodating the largest number of tubes. Additionally, by reducing the reactor length, unnecessary catalyst consumption and heat injection are prevented, as rapid chemical reactions and physical permeation phenomena primarily occur at the reactor's front. However, an excessively short reactor length hinders complete reaction due to a limited residence time. Therefore, the reactor length is optimized by considering the number of baffles and the baffle cut, aiming to ensure a complete reaction. The results not only efficiently identify a scalable optimal design for the membrane reactor through comparison of various objective functions but also provide valuable guidelines for reactor design during scale-up.

The second part of the research involves conducting simulations of the fed-batch fermentation process on a pilot scale, taking into account previous findings that a concentration gradient occurs in the medium on a larger scale. In the fermentation process, the rotation of the impeller is utilized to create a uniform medium and enhance fermentation performance. However, in larger-scale fermentations, the performance is influenced by the dispersion of gas and the distribution of medium concentration, which depend on the design and operating conditions of the reactor. Previous studies dealt with the design and operating conditions independently, and paid attention to the media environment, such as gas holdup, which can affect the growth of strains. However, solely increasing gas holdup fails to consider strain growth in regions with locally low strain concentration or in dead zones where gas delivery is limited. Motivated by this, the paper shifts its focus from gas holdup to gas distribution, aiming to optimize the fermentor's design and operating conditions for strain growth rate based on fermentation kinetics. To achieve this, CFD simulations of fermentors are combined with fermentation kinetics estimated by experimental data. Initially, 32 sets of data concerning design and operating conditions were generated through the implementation of Design of Experiment (DOE) with the aim of maximizing the strain's growth rate. The results of the DOE reveals a wide distribution of the objective function across variables, and the top 10 % of experiments mostly indicate a fast agitation speed. However, it cannot be concluded that a global optimal solution is found because other variables except for the agitation speed point to different values. The BO results exceeded the maximum strain growth rate found using the DOE with only 7 additional experiments starting with the initial number of 5 experiments. The top 10 % of experiments obtained through BO exhibit characteristics such as wider baffle widths, smaller spaces between impellers, slower gas flow rates, and higher agitation speeds. The resulting slow gas flow rate demonstrates that the gas distribution, not the gas holdup, is important for the growth rate of the strain. Furthermore, the optimal solution demonstrated that gas distribution is influenced by the design even under conditions of low gas flow rates and high agitation speeds. For instance, an experiment featuring a narrow baffle width and large space between impellers, despite similar operating conditions, reveals a larger dead zone. Hence, the optimal solution obtained by BO provides valuable insights into the design and operating conditions required for achieving perfect gas dispersion in a fermentor.