(316b) Bioreactor Simulation With Cuda

Bioreactor simulation with CUDA

Christian Witz¹,
Tawan Tantikul¹, Johannes Khinast^1,2

¹Graz
University of Technology, Institute for Process and Particle Engineering

²Research
Center Pharmaceutical Engineering

Abstract

The effectiveness of a biopharmaceutical manufacturing
process depends to a large extent on the efficiency of the bioreactor,
especially in the field of generic drugs. So far, the engineering process of
the reactor design has been mostly driven by empirical knowledge, as the
simulation of this complex multiphase and multiscale process was impossible for
many years. However, despite the recent improvement of computational
capabilities, the simulation of an industrial scale reactor takes months for
only a few seconds of real operation time.

Thus, the goal of this study is to use graphic cards to speed
up this simulation. The Compute Unified Device Architecture (CUDA) technology
of nVidia has made the computational power of graphic processing units (GPUs)
available for scientific calculations [1]. In the multiphase simulation, the
large number of computing units in the GPU leads to a significant reduction of
calculation time. To archive this acceleration, an efficiently parallelizable
simulation method is needed.

The lattice Boltzmann method (LBM) which was developed based on the lattice gas
automata [2], represents an efficient way to numerically capture the liquid
phase flow dynamics on the GPU computing machine. It uses a regular grid with
evenly distributed nodes. To model the geometry without changing the regular
grid the immersed boundary method is used. The air bubbles are simulated with
the Lagrangian particle tracking (LPT) method. The sum of the forces acting on
each bubble, i.e. the drag, the buoyancy, the lift force, the history force,
the added mass effect and gravity is used to determine the acceleration of the
particle. The acceleration and the time step length give the velocity and the
position change at the end of the time step. For coalescence and breakup
stochastic models are used. The phases are coupled with a two way approach.

As the calculation results are computed on the graphic card, simulation
data can be visualized real time during the calculation. Furthermore to
simulate large reactors, the code has a multi GPU functionality, hence, it can
distribute the workload of the simulation on several graphic processors. This
multi GPU code is able to handle the large scale, memory demanding simulation
such as the simulation of the industrial scale reactor.

References

[1]        http://www.nvidia.com/cuda

[2]        Sukop,
M.: Lattice Boltzmann Modeling, Springer, Berlin, 2007.

Submission details

Session
- 26013 Mixing Issues and Scale-Up/Scale-Down Strategies for the Production of
Pharmaceutical and Biopharmaceuticals

Emails:

christian.witz@tugraz.at

tawan.tantikul@tugraz.at

khinast@tugraz.at