Performance Evaluation of Damaged Blades Using Numerical Simulation and Machine Learning

Abstract

The formulation of maintenance strategies for gas turbines requires an accurate and rapid assessment of damaged blade performance. Variety and uncertainty in the three-dimensional geometric deformation of gas turbine vanes and blades increase the difficulty of degradation analysis. In this paper, the effects on the aerodynamic performance of partial damage to turbine nozzle guide vanes are investigated experimentally and numerically. A methodology for constructing a deformation–degradation model for high-pressure gas turbine vanes/blades is proposed based on machine learning. A database consisting of 670 numerical simulation results with various deformation patterns is constructed for the vane cascade. A mapping is constructed between the matrices defining geometric deformations and the total pressure loss coefficient of the cascades based on a convolutional neural network model. The accuracy of this deformation–degradation model is verified against experimental results. We find that deformation on the suction side of the blade has the most significant influence on aerodynamic performance. The deformation–degradation model constructed for a specific cascade provides a rapid evaluation of aerodynamic performance based on rough geometrical measurements without additional experiments or numerical simulations, providing a potential tool in the precise repair of gas turbines.