Research Fellows
Orest Mykhaskiv
Early Stage Researcher 2 at Queen Mary University of London
Optimal shape design with automatically differentiated CAD parameterisations
Typical engineering workflow for aerodynamic design could be considered as the three-stage process: (i) modelling of a new component in a CAD system, (ii) its detailed aerodynamic analysis with CAE/CFD simulation software on the computational grid and (iii) manufacturing of the CAD model using CAM tools. Numerical shape optimisation, conducted during the analysis stage (ii), is becoming an essential industrial method to improve aerodynamic performance (cost) of the shapes immersed in fluids. For instance, a modification of the geometry of an aircraft wing leads to changes in the airflow around it and could increase generated lift force.
In this project, for the first time, Automatic Differentiation (automatic backpropagation) was applied to the aerodynamic design process (i, ii) - providing derivatives of CFD and CAD software. These derivatives quantify the influence of CAD/shape/geometry parameters (e.g. aircraft's wing-span, height of a car) on the changes of aerodynamic cost functions (e.g. lift of the aircraft, drag of the car, etc.). The application of this data in computationally efficient gradient-based optimisation algorithms resulted in the improvement of several industrial components from automotive, turbomachinery and aerospace industries.
