Multi-Stage Robust Bayesian High-Resolution Identification of Asynchronous Blade Vibrations Using Blade Tip Timing.

Area of Science:

• Mechanical Engineering
• Signal Processing
• Condition Monitoring

Background:

• Blade Tip Timing (BTT) is crucial for non-contact vibration monitoring in rotating machinery.
• Practical BTT accuracy is hindered by noise, undersampling, and spectral leakage.
• Existing methods struggle with robust performance under adverse conditions.

Purpose of the Study:

• To develop a multi-stage robust Bayesian high-resolution identification framework for BTT.
• To address challenges of noise, undersampling, and spectral leakage in BTT measurements.
• To enhance the accuracy and efficiency of vibration analysis for blade health monitoring.

Main Methods:

• Utilized Kalman filtering for noise-suppressed rotational speed estimation without probes.
• Employed an attention-enhanced dynamic convolutional autoencoder for window function generation to minimize spectral leakage.
• Implemented all-phase FFT with sub-bin interpolation for high-resolution phase extraction.
• Applied a Tukey-biweight robust aggregation for multi-channel phase fusion.
• Incorporated Bayesian priors for physically plausible vibration order estimation.
• Developed a coarse-to-fine multi-stage search for computational efficiency.

Main Results:

• Demonstrated significant noise floor reduction (approx. 60 dB) on a rotor-blade test bench.
• Achieved a maximum frequency identification error of 7.84% under constant and variable speeds.
• Accelerated the search process by approximately 48.6% compared to exhaustive methods.
• Validated the framework's effectiveness in suppressing sensor noise and spectral leakage.

Conclusions:

• The proposed multi-stage robust Bayesian framework offers a reliable and efficient solution for BTT.
• The method significantly improves vibration identification accuracy and reduces computational load.
• This advancement is critical for effective blade health monitoring in rotating machinery.