Towards uncertainty quantification and management in multi-disciplinary design optimisation

• Published in: 2016 IEEE Congress on Evolutionary Computation (CEC) pp. 885 - 892
• Main Authors: Loxham, J., Kipouros, T., Savill, A. M.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.07.2016
• Subjects: Algorithm design and analysis; Linear programming; Manufacturing; Optimization; Robustness; Sensitivity; Uncertainty
• DOI: 10.1109/CEC.2016.7743884

In this paper we explore uncertainty quantification and management in an industrial context. We use a multielement airfoil section at take-off conditions, the so called Airbus Test Case A, for the demonstration of our methods. We also try to identify through this case study the elements that are necessary to synthesise in order to move towards the characterisation of multi-disciplinary complexities of uncertainty. Multi-point optimisation, using a multi-objective Tabu Search, is first used to demonstrate a significant decrease in sensitivity to variations in incidence and flap deflection. Sigma-point, non-intrusive spectral projection, and non-intrusive point collocation methods of uncertainty quantification are then applied to the same test case with uncertain incidence. An additional uncertain parameter, the flap deflection, is then added for a final optimisation using non-intrusive point collocation. Though mean increases in lift to drag ratio are comparable, a systematic reduction in attainable robustness is observed when compared to the single uncertainty cases. We also emphasise on the importance of visualising the multi-dimensional data that are produced from such computational design studies, and the possibility to identify hidden characteristics and correlations between parameters and performance characteristics. We use Parallel Coordinates and interactively we can reveal such relationships. We believe a suitable synthesis of all these tools and methods can assist towards the identification of implicit correlations between uncertainties that are produced when different disciplines are considered simultaneously for industrial design.