Sparse Representation Based Frequency Detection and Uncertainty Reduction in Blade Tip Timing Measurement for Multi-Mode Blade Vibration Monitoring

Minghao Pan, Yongmin Yang, Fengjiao Guan, Haifeng Hu, Hailong XuSensors 2017 · DOI

TL;DR

This paper introduces a novel method for accurately monitoring blade vibrations in turbomachinery using Blade Tip Timing (BTT) data. By leveraging sparse representation and a new de-noising technique, the method effectively identifies multiple vibration frequencies even from under-sampled and noisy BTT signals, without requiring additional sensors.

What Problem Does This Paper Address?

Accurate monitoring of blade vibrations in high-speed rotating machinery like turbines and compressors is crucial for fault diagnosis and preventing catastrophic failures. Blade Tip Timing (BTT) is a promising non-contact technique for this, but it suffers from two major challenges:

  1. Under-sampling: BTT systems typically use a limited number of probes, meaning the data collected is not frequent enough to capture all vibration frequencies according to the Nyquist theorem.
  2. Measurement Uncertainties: The BTT signals themselves are prone to noise and inaccuracies, such as probe position shifts, which can distort the timing measurements and lead to incorrect vibration frequency identification.

These issues make it difficult to accurately recover the true vibration frequencies, especially when blades exhibit multi-mode vibrations (vibrating at multiple frequencies simultaneously). Existing methods often struggle with these under-sampled and noisy conditions, or require a large number of probes, which is impractical in many real-world scenarios.

Methods

The paper proposes a four-phase approach:

  1. Sparse Representation Model: They first build a mathematical model of the BTT measurement by exploiting the fact that blade vibration signals are "sparse" in the frequency domain. This means that only a few specific frequencies are dominant, and the signal can be represented efficiently using these few frequencies.
  2. Frequency Spectrum Recovery: Using the theory of sparse representation and compressed sensing, they develop a method to recover the vibration frequency spectrum from the under-sampled BTT data. This involves formulating the problem as a mathematical optimization task.
  3. Uncertainty Analysis: The authors quantitatively analyze the typical sources of BTT measurement uncertainties, specifically focusing on signal noise and probe position shifts. They model these uncertainties as additive noise to the ideal BTT signal.
  4. Improved Recovery Method (BPDN): To address the corrupted signals due to uncertainties, they adapt the sparse representation method by incorporating a Basis Pursuit De-noising (BPDN) algorithm. This algorithm is designed to find a sparse solution while simultaneously minimizing the impact of noise, effectively de-noising the signal and improving the accuracy of frequency recovery.

A key aspect of their approach is the careful selection of probe placement to minimize the "coherence" between the sensing matrix, which is crucial for successful sparse recovery.

Datasets

The paper utilizes both simulated and experimental data for evaluation:

  • Simulated Data:
    • Multi-mode vibration signals were generated using Equation (25) and (26), incorporating both clean and noisy scenarios with varying noise-to-signal ratios (NSR).
  • These simulations were used to compare the performance of their proposed BPDN method against a standard Basis Pursuit (BP) algorithm.
  • Experimental Data:
    • Vibration-based test rig: Data was collected from a damaged rotor blade specimen subjected to harmonic excitation. This was used to validate the recovery of multi-mode vibration responses in the frequency domain.
  • Rotating blades test rig: BTT signals were acquired from a rotating bladed disk under controlled conditions. This experiment focused on demonstrating the uncertainty reduction capabilities of the proposed method in a more realistic scenario.

No new benchmark datasets were introduced; rather, the paper uses existing experimental setups to validate its proposed method.

Key Insights and Ideas

  • Sparsity is Key: The fundamental idea that blade vibration signals are sparse in the frequency domain is the bedrock of the proposed method, allowing for efficient representation and recovery.
  • Probe Placement Matters: The spatial arrangement of the BTT probes significantly impacts the success of sparse recovery. The paper emphasizes finding placements that minimize the coherence of the sensing matrix for optimal results.
  • BPDN for Robustness: The introduction of the BPDN algorithm is a critical advancement, enabling the method to effectively handle noisy and uncertain BTT data, a common issue in real-world applications. This is achieved without needing more probes.
  • Multi-mode is Achievable: The method demonstrates the ability to recover multiple vibration frequencies simultaneously, which is essential for understanding complex blade dynamics and identifying various fault conditions.
  • No Extra Probes Needed: A significant advantage highlighted is that the proposed method achieves improved accuracy and robustness without requiring an increase in the number of BTT probes, making it a practical and cost-effective solution.

Key Results

  • Simulation Results:
    • The BPDN method showed significantly better performance in recovering vibration frequencies from noisy simulated data compared to the standard BP algorithm. For instance, in one simulation, the probability of recovery (β) for BPDN was 56.03% compared to 31.29% for BP under specific noise conditions.
  • The accuracy of recovery generally improved with an increased number of probes (from 3 to 5), but the BPDN method consistently outperformed BP across all probe configurations.
  • Experimental Results:
    • The experimental validation confirmed the effectiveness of the proposed method in recovering vibration frequencies from both clean and corrupted BTT signals.
  • Figure 17b shows a clean recovered spectrum using BPDN, clearly identifying the dominant frequency (2270 Hz), which was well-accorded with the Campbell diagram, whereas the BP algorithm's result (Figure 17a) was mixed with biased frequency characteristics.
  • The method successfully identified vibration frequencies in both simulated and experimental multi-mode scenarios.

Limitations and Future Work

  • Single Synchronous Mode in Current Equipment: The authors note that due to the limitations of their current experimental setup (e.g., material stiffness, excitation method), they were primarily able to excite and analyze a single synchronous mode of the blades.
  • Future Work: The authors plan to focus on acquiring and analyzing BTT sampling data that contains multi-mode vibration components and measurement uncertainty simultaneously from compressors or jet engines. This will further validate the proposed method in more complex and realistic operational environments. They also aim to explore more advanced techniques for acquiring and analyzing such data.

Summary updated

BibTeX
@article{pan2017sparse,
  author = {Minghao Pan and Yongmin Yang and Fengjiao Guan and Haifeng Hu and Hailong Xu},
  title = {{Sparse Representation Based Frequency Detection and Uncertainty Reduction in Blade Tip Timing Measurement for Multi-Mode Blade Vibration Monitoring}},
  journal = {Sensors},
  year = {2017},
  doi = {10.3390/s17081745},
}
Figure