Research Fellows
Mladen Banovic
Early Stage Researcher 12 at Universität Paderborn
Computer-aided design (CAD) tools and systems are considered essential for industrial design. They construct and manipulate geometry of a certain component with an arbitrary set of design parameters. However, it is a challenging task to incorporate a parametric CAD description into a gradient-based shape optimization loop since the commercial CAD systems usually do not provide the so-called shape sensitivities with respect to design parameters of the model being optimized. Typically, these derivatives are calculated using inaccurate finite differences (FD). To get the exact derivative information, algorithmic/automatic differentiation (AD) can be applied if the CAD sources are available.
AD is a technique to modify a computer program such that it provides derivatives for its functions. This modification process involves integration of an AD software tool into the original program source code to generate its differentiated version.
This research is focused on efficient AD of the following CAD libraries:
- Open-source CAD kernel Open CASCADE Technology,
- VKI in-house CAD tool CADO,
- Rolls-Royce in-house airfoil design and blade generation tool Parablading.
The differentiated CAD libraries have been coupled with discrete adjoint CFD solvers, also produced by algorithmic differentiation. The work conducted within IODA project is the first example of developing complete differentiated design chains that are successfully applied to gradient-based shape optimization of industrial components.