Three-Dimensional Aerodynamic Optimization of a Multi-Stage Axial Compressor

An optimization method combined of a genetic algorithm, an artificial neural network, a CFD solver and a blade generator, is developed in this research and applied in the three-dimensional blading design of a newly designed highly-loaded 5-stage axial compressor. The adaptive probabilities of crossover and mutation, non-uniform mutation operator and elitism operator are employed to improve the convergence of the genetic algorithm. Considering both the optimization efficiency and effectiveness, a mixture of high-fidelity multistage CFD method and approximate surrogate model of the feed-forward ANN is used to evaluate the fitness. In particular, the database is updated dynamically and used to re-train the surrogate model of ANN for improving the accuracy for predicting. The last stator of the compressor is optimized at the near stall operating point. The tip bow with relative bow height Hb and bow angle αb are treated as design parameters. The adiabatic efficiency as well as the penalty of mass flow and total pressure ratio constitute the objective functions to be maximized. The optimum (H_b = 0.881, α_b = 14.7°) obtains 0.4% adiabatic efficiency increase for the whole compressor at the optimized operating point. The detailed aerodynamic is compared between the baseline and optimized stator, and the mechanism is analyzed. The optimized version obtains 5.1% increase in stall margin and maintains the efficiency at the design point.