Automated compressor blade design requires solution of a high-dimensional optimization problem involving several evaluation tools to account for both aerodynamic and structural aspects. However, the computational complexity of CFD and FEM simulations is rather high, and time scales for the evaluations may be rather different. Both problems can be overcome by using surrogate modeling where the expensive simulations are replaced by computationally cheap approximation model and design evaluation is decoupled from the optimization loop. A generic design process structure for multi-disciplinary optimization is introduced, which makes use of surrogate modeling to mitigate time restrictions and reduce unnecessary waiting times for the independent sub-processes. Partial least squares (PLS) in combination with Kriging is investigated as a measure to reduce dimensionality and multi-collinearity, which thereby reduces the time for building up the surrogate model to speed up the design process. This is demonstrated by several well-proven optimization test functions and also an application to a real compressor blade optimization for stationary gas turbines.
