WP 4: CAD-based parametrisation

The ultimate design requirement is an optimal design defined in the CAD modelling system. This work package will investigate three strategies to use the CAD system to represent shape during the optimisation process, with its defining parameters acting as the design variables.

NURBS based approaches offer a rich design space but their benefit can be limited by the definition of the NURBS profile available, which may not be appropriate for optimisation. An adaptive sub-division process will be investigated which will use adjoint sensitivity information to refine the NURBS definition to be more suitable for optimisation. Other work is required in enriching the vocabulary of the light-weight CAD kernel to include trimmed patches and resulting moving patch intersection following a design perturbation, as well as the formulation of constraints.

Featured CAD-based representations have the advantage that the features chosen to represent the part can be used to embed the designers intention for the underlying design, and constrain the effects of the design changes on model shape to fulfill this. The approach is limited by the fact that the CAD model is usually prepared with manufacturing in mind, and the features included may not be the best choice for optimisation. Within this work package a process to adaptively insert CAD features based on the sensitivity map, to provide parameters which can change the shape in the optimum way, will be investigated.

Automatic differentiation (AD) has the advantage that while a geometry definition for the design is available in the CAD system, the geometry itself is not used in the computation of gradients. As such AD has the potential to provide accurate gradients efficiently. Within this work package the differentiation of the CAD tool to provide derivatives with respect to model parameters will be explored.

Objectives:

• CAD derivatives through finite-differencing

• CAD derivatives through algorithmic differentiation of OCCT

• CAD derivatives through differentiation of specialised turbo-machinery CAD kernels

• NURBS patch refinement, subdivision and elevation to refine the design space

• Adaptive CAD feature insertion

Personnel and Project Deliverables

WP leader: Queen's University Belfast - Trevor Robinson, Cecil Armstrong

Rejish Jesudasan (ESR 1) "NURBS kernel with trimmed patches"

Orest Mykhaskiv (ESR 2) "Differentiated reduced OCCT" and "Differentiated OCCT with extended kernel" and "Parametric engine for OCCT"

Flavio Gagliardi (ESR 7) "NTUA parametrisation tool" and "NTUA CAD tool differentiated"

Salvatore Auriemma (ESR 9)"Differentiated reduced OCCT" and "Differentiated OCCT with extended kernel" and "Parametric engine for OCCT"

Dheeraj Agarwal (ESR 10) "Strategies for adaptive CAD feature insertion"

Mladen Banovic (ESR 12) "Differentiated reduced OCCT" and "Differentiated OCCT with extended kernel" and "Parametric engine for OCCT" and "Improved differenyiation of OCCT"

Ismael Sanchez Torreguitart (ESR 13) "Reference forward differenced evaluation chain for validation" and "Grid to CAD sensitivities"

Marc Schwalbach (ESR 14) "CAD with cold-to-hot transformation"