A Machine Learning Based Hybrid Nonlinear Character Monitoring Approach For Compressor Blades Fault Diagnosis Using Blade Tip Timing Measurement

Abstract

Compressor blades which are exposed to harsh working conditions in turbine engines will suffer from varying damage and complex faults inevitably. Thus, monitoring the blade dynamic condition instantaneously and effectively is essential for detecting initial defects. As a prominent noncontacting measurement method, blade tip timing (BTT) is widely implemented in compressor blades vibration detection. The purpose of this study is to diagnose blade faults by obtaining nonlinear dynamic characters from BTT data. Firstly, nonlinear dynamic model for cracked rotating blades is built. Then blade vibration frequencies are extracted from BTT signals based on the sparse representation model. In addition, the relationship between the nonlinear dynamic response of blades and the vibration frequency is revealed. Finally, with the assistance of machine learning algorithms, blade damage degrees are classified combining with vibration frequency spectra and nonlinear characteristics. Numerical simulations are designed to verify the feasibility of the proposed method.