MACHINE LEARNING-BASED MULTI-DISCIPLINARY OPTIMIZATION OF TRANSONIC AXIAL COMPRESSOR BLADE CONSIDERING AEROELASTICITY

The optimal design of the turbomachinery field can produce the best results when considering multiple fields (e.g., structural vibration, aerodynamic, and aeroelasticity). Since multi-disciplinary phenomena are very complex and nonlinear, a higher level of optimization is needed. In this study, the transonic axial compressor blade was optimized using machine learning-based multi-disciplinary optimization techniques. The entire simulation includes important three factors to consider when designing the axial compressor blade: vibration, aerodynamic performance, and aeroelasticity. In the first step of the optimization, six variables that make up the blade design were used in a design of experiments to explore the design space. Frequency, efficiency, pressure ratio, and aerodynamic damping ratio were used as the output parameter used in the optimal design. The first to third blade modes frequency from the modal analysis were used for parameters of vibration fields. The efficiency and pressure ratio are values that can be obtained from steady CFD calculations for evaluating aerodynamic performance. The aerodynamic damping coefficient was obtained from transient CFD, which is widely used to determine the presence or absence of fluid-induced vibration and flutter. As a result of machine learning-based optimization, the optimized blade shape has a lower risk of blade resonance. Also, the aerodynamic performance of the axial compressor has been improved compared to the reference model. In addition, the aerodynamic damping coefficient, an important indicator in terms of aeroelasticity, rose by up to 15%, completing optimization considering the aeroelastic effect.