Research Fellows / Open Positions

Early Stage Researcher 12 at UPB

Efficient automatic differentiation of CAD systems

The derivatives of the parametrisation are required to fully integrate the parametrisation into gradient-based optimisation workflows, but for typical CAD systems these derivatives are not available. In this project the open-source CAD kernel OpenCASCADE (OCC) is differentiated using the Automatic Differentiation (AD) software tool ADOL-C from UPB. The focus for this project is to advance with the capability of the AD tool in order to successfully differentiate the complex and large source-code of OCC.
This project is in conducted in close collaboration with ESR2 at QMUL and ESR9 at OpenCascade.

Objectives:

• M6-M18: Familiarisation with OCCT, Efficient automatic differentiation in forward mode of the open-source CAD program OCCT, demonstration on small testcase. Secondment to OCC (WP4).
• M18-M28: Extension of the methodology to the reverse mode of AD at suitable parts of the differentiated OCCT package for improved efficiency.  Secondment to QMUL. (WP4).
• M28-M36: . Automatic differentiation of mesh to CAD  algorithms. Secondment to VKI. (WP3).
• M36-M42: Integration with in-house solver for demonstration/evaluation, linking as a plugin into ParaView using the IODA API (WP6).

For further details, contact Prof. Andrea Walther, andrea.walther@uni-paderborn.de