Automated Data-Driven Physics Discovery of Turbine Component Damage

Abstract

We propose an automated physics discovery algorithm for turbine component damage modeling. Our algorithm utilizes operational data of a mechanical component and discovers an interpretable symbolic formula that describes the physics. We illustrate our algorithm through two numerical examples and demonstrate that the discovered formulas can predict the future damage accurately. Our framework is flexible and easily applicable to all areas of science and engineering. With cutting-edge machine learning tools, researchers can simply input the experimental data and then the physics formulas are printed out automatically. The application of this new algorithm may reduce the time spent on research and development of physics models significantly, while achieving almost the best accuracy in prediction.