A Data-Driven Modelling Approach and Uncertainty Analysis for Rotor System Health Assessment

The rotor system is an important part of rotating machinery. Real-time health assessment of the rotor system is critical to ensure the safe operation of rotating machinery. Due to the complexity of rotating machinery, it becomes increasingly difficult to establish an accurate physical model. With the availability of extensive data from simulations and experiments, data-driven modelling based on sensor data has gradually become mainstream. The previous analysis of the uncertainty of data-driven model is not accurate and comprehensive. This paper focuses on the influence of model parameter uncertainty, which are induced by noise, unexpected frequency components and excitation frequency fluctuations, on the accuracy of system health assessment. Firstly, the bilinear equation for the faulty rotor systems was used to theoretically analyze the influence factors of the model uncertainty. After that, the correlation between model uncertainty and system health assessment accuracy is analyzed. The employed health assessment approach uses Generalized Associated Linear Equations (GALEs) and Nonlinear Output Frequency Response Functions (NOFRFs). Finally, combining with the fault mechanism, the theory and approach suitable for improving the assessment accuracy are proposed.