This paper presents the computational methodology, and experimental investigations accomplished to enhance the efficiency of a turbine stage by applying non-axisymmetric profiling on the rotor hub wall. The experimental setup was a two-stage axial turbine, which was tested at “LISA” test facility at ETH Zurich. The 1st stage was considered to create the flow history for the 2nd stage, which was the target of the optimization. The hub cavity of the 2nd stage was designed with large dimensions as a requirement of a steam turbine. The goal was to optimize the interaction of the cavity leakage flow with the rotor passage flow to reduce the losses and increase efficiency.
The computational optimization was completed using a Genetic Algorithm coupled with an Artificial Neural Network on the 2nd stage of the test turbine. Unsteady time-accurate simulations were performed, using in-house developed “MULTI3” solver. Besides implementing all geometrical details (such as hub and tip cavities and fully 3D blade geometries) from the experimental setup into the computational model, it was learned that the unsteady upstream effect could not be neglected. A novel approach was introduced by using unsteady inlet boundary conditions to consider the multistage effect while reducing the computational cost to half. The importance of this implementation was tested by performing a steady simulation on the optimized geometry. The predicted efficiency gain from steady simulations was 4.5 times smaller (and negligible) compared to the unsteady approach. Excluding the cavity geometry was also assessed in a different simulation setup showing 3.9% over-prediction in the absolute efficiency value.
Comprehensive steady and unsteady measurements were performed utilizing pneumatic, Fast Response Aerodynamic (FRAP), and Fast Response Entropy (FENT) probes, on the baseline and profiled test cases. The end-wall profiling was found to be successful in weakening the strength of the hub passage vortex by a 19% reduction in the under-over turning. As a result, the blockage was reduced near the hub region leading to more uniform mass flow distribution along the span. The flow angle deviations at the higher span position were also corrected due to better control of the flow angles. Furthermore, the improvements were confirmed by reductions in entropy, Secondary Kinetic Energy, and pressure unsteadiness. The accurate computational implementations led to an excellent agreement between the predicted and measured efficiency gain.