13th European Conference on Turbomachinery Fluid dynamics & Thermodynamics
Turbulence modelling remains a challenge for the simulation of turbomachinery flows.Reynolds Averaged Navier-Stokes (RANS) equations will still be used for high-Reynoldsnumber flows for several years and so there is interest in improving their prediction capa-bility. Machine learning techniques offer several strategies which could be exploited forthis purpose.In this work, an approach to improve the Spalart-Allmaras model is investigated. In par-ticular, the model is used to predict the flow around the T106c low pressure gas turbinecascade. As a first step, an Artificial Neural Network (ANN) is trained on the data gen-erated by the original model. Then, an optimisation procedure is applied in order to findthe weights of the network which minimise the error between the predicted results andthe available experimental data. The new model is tested at different Reynolds numberson the T106c cascade and on a wind turbine airfoil in post-stall conditions. Significantimprovements are observed in the condition chosen for the optimisation. Future work willbe devoted to the generalisation of the approach by including multiple working conditionsoptimisations and adding new physical variables as inputs of the ANN.