Efficient Surrogate Modeling for Turbine Blade Row Cyclic Symmetric Mode Shapes

Abstract

In this study, we consider the Modal Cyclic Symmetric (MCS) analysis to generate the modal frequencies and mode shapes of a tip-shrouded turbine blade row. In total 87 input parameters are considered here that define the geometry of the turbine blades. The objective here is to build an efficient surrogate model which maps these input parameters to the mode shapes. The main challenge is the high dimensionality of a complex mode shape at a mesh resolution of industrial standard which is almost of the order of O(107). Thus, the idea here is to efficiently reduce the high dimensional output into a lower dimensional representation using the unsupervised deep learning architecture called the Convolutional Variational AutoEncoder (CVAE). The CVAE model helps in performing efficient surrogate model building using GE’s Bayesian Hybrid Model (BHM) in a lower dimensional output space as well as produce feasible mode shapes for new geometries after training.