Research
Adjoint-based design optimisation techniques have reached Technology Readiness Level (TRL) 6 for some examples, such as automotive ducts and turbine blades, but general industrial use is currently still in the research stage. Adjoint methods are expected to revolutionise modern gasturbine design: by applying adjoint techniques, the optimisation of the entire gasturbine system with millions of degrees of freedom is within reach of the current available computational power.
However, today’s reality is far from this prospect. Current adjoint design optimisation only considers aerodynamic performance, preventing the optimisation of complete systems which are by nature multidisciplinary. This project will develop an adjoint optimisation methodology that goes beyond only aerodynamic considerations and includes other disciplines such as structural mechanics, vibration dynamics and conjugate heat transfer concurrently, such that in the longer term optimisations of complete systems will become achievable.
The key to achieving a true multidisciplinary adjoint design optimisation is to exchange data across discipline interfaces efficiently and accurately, which implies to define a datum reference geometry to work with. Here we propose to use a master CAD geometry that is shared between all the different disciplines. This differs strongly from the current practice in adjoint techniques, which mainly considers parametrisations that are only suitable for either aerodynamic or structural optimisations. Using a master CAD geometry to define deformations under changes by design variables in gradient-based optimisation requires the differentiation of a CAD system. This has not yet been performed as CAD systems are proprietary and as such not accessible. Including the CAD system inside the design loop has however also the significant advantage that the optimal shape is produced in CAD format, ready for further analysis or manufacturing without any transcription step which would invariably impair optimality.
Approach
Turbomachinery components are typically designed in a specialized CAD environment where the blade shape is controlled by advanced engineering parameters. The obtained CAD geometry is then used as ‘master geometry’ and shared between the different analysis paths. Each analysis first discreties the shape through a grid generation process and then solves a set of partial differential equations, discretied on the grid. Finally performance parameters are extracted which serve as objectives or constraints. Our aim is to differentiate such entire design chain for several analysis, i.e. to aerodynamic analysis to include structural analysis and heat transfer.
A reverse differentiation will be performed for the entire design chain, including the CAD kernel and grid generation/deformation tools, CSM and CFD analysis tools. This will allow for obtainingproduce an efficient computation of the CAD design parameter sensitivities irrespective of their number, and allow for large system optimiszations.
Multidisciplinary design process of a radial compressor. Left: primal problem with CFD and CSM analysis. Right: sensitivity analysis in forward and reverse mode.