Simulation Based Optimization of Stirred Tank Reactors

The design of a stirred tank reactor is a multi‐input and multi‐output (MIMO) problem. Design parameters include geometric parameters (impeller & tank shape, size number, baffles.) as well as operating conditions (impeller RPM, fill level etc.). A common design objective is minimizing mixing time, maximizing mixing quality, to achieve a specific scale of agitation etc. Additionally, various constraints can be imposed on the performance of such a reactor, such as avoiding overflow, avoiding mixing `dead zones', imposing a maximum shear stress etc. Designing to these multiple objectives and constraints by modifying a large number of design parameters is a challenging task since the relationships of these multiple inputs and outputs can be highly non‐linear.

Due to this reason, the design and scale‐up of stirred tank reactors has traditionally been a very time consuming task. This is the primary motivation for developing a design methodology that is both autonomous, and efficient in targeting specific reactor performance. Optimization algorithms lend themselves well to being applied to solve such multi‐input, multi‐output problems. An optimization algorithm iteratively runs evaluations and understands the sensitivities between all inputs and combination of inputs to all outputs. The algorithm successively guesses a better design until a converged design is reached that optimizes all the required objectives while satisfying all constraints. This study details the implementation of this optimization methodology to design optimal stirred tank reactors for various design objectives and constraints. The use of an optimization algorithm for design studies requires two components: an analysis tool and an optimization algorithm. The optimization algorithm uses the analysis tool to generate the relationship between a set of inputs and outputs. The reactor is analyzed using a computational fluid dynamics (CFD) approach with STAR‐CCM+. The use of STAR‐ CCM+ for mixing tank performance prediction has been validated by the work of Maas et al. Results for a pareto front optimization with the goal of minimizing power and mixing intensity is presented.