Numerical optimization for stator vane settings of multi-stage compressors based on neural networks and genetic algorithms

This paper outlines the aerodynamic optimization of stator vane settings for multi-stage compressors via the combination of an artificial neural network (ANN) and a genetic algorithm (GA). The investigation is conducted on a newly developed 5-stage highly loaded axial flow compressor. A three-layer perceptron neural network is employed as surrogate model, replacing an in-house one-dimensional blade stacking computation code. The stagger angles of the four stator vanes serve as the input data of the ANN, and the compressor aerodynamic performances are the outputs of the network. The well-trained ANN is subsequently incorporated into the optimization framework, which is based on an improved real-coded GA. Various advanced strategies, including the elitism operator, blend crossover, non-uniform mutation and self-adaption parameters, are introduced to the GA to promote the searching efficiency and solution globality. The optimization is conducted on the reference operating points under both design- and part-speed conditions to achieve maximum adiabatic efficiency with restrictions on the pressure ratio. The results show that for the design speed, the original stator vane setting is good, and the room for growth in efficiency is limited based on the one-dimensional optimization. However, the optimized stator vane settings improve the adiabatic efficiency by more than 1% under part-speed conditions, and the enhanced efficiency is achieved over the entire operating range. Regardless of the assumption of quasi-one-dimensional flow, the effectiveness of the optimization framework in dealing with the stage-mismatching is demonstrated. Moreover, a new sensitivity analysis method using ANN is proposed to evaluate the relationships between the geometric parameters and aerodynamic performances of the compressor.