Dr. Paulus Lahur1, Dr. Gerald Pereira1, Dr. David Howard1, Dr. Phil Kilby1
1CSIRO, Clayton, Australia
Biography:
Paulus has been working in CSIRO for nine years. He has been contributing to various projects that require high performance computing (HPC). His background includes computational fluid dynamics (CFD) and software development.
Abstract:
Static mixer is a device that mixes fluids without any moving part. Such device is useful for many applications, including removal of chemicals within fluid.
Traditionally, this device is designed manually. In this project, leveraging the availability of computational power, we aimed to both fully automate the design process and come up with innovative designs that perform significantly better than human designed mixers.
Our optimisation process is driven by Genetic Algorithm. Industrial optimisation is generally limited to small geometric variation, as in the case of aircraft wing design, where one design is like another. For our project, to come up with innovative designs, we need to allow for very large variation in shapes. One design may have little in common with another.
This poses significant challenges to the simulation process, as any arbitrary shapes must be handled. The choice of method and its implementation is critical to the success of this effort, as we simulate flow for thousands of design candidates automatically. Having the computation stopped due to error and troubleshooting every now and then is simply not an option.
To this end, we have developed and implemented Lattice-Boltzmann method (LBM) for fluid simulation, which is well known for its robustness, coupled with automated voxel grid generation. The grid generator can take solid objects of any shapes as input and tolerate overlapping solid objects and defects.
The method has proven to be effective. In our test cases, our mixers managed to significantly outperform the human-designed mixer.