Enhancing Sparse Decomposition Based Blade Vibration Parameter Identification

Shuming Wu, Haifeng Hu, Zhibin Zhao, Yang Zhi-bo, Laihao Yang, Shaohua Tian, Xuefeng ChenJournal of Mechanical Engineering 2019 · DOI

Abstract: Blade Tip Timing (BTT) is an effective method for monitoring aero-engine blade health. Unlike traditional strain gauge measurements, BTT can simultaneously monitor all blades without affecting their natural vibration characteristics. However, BTT data is inherently undersampled. To address this, this paper proposes an Enhancing Sparse Decomposition (ESD) technique for blade vibration parameter identification. Sparse decomposition represents signals within a redundant dictionary to find the sparsest solution. Compared to traditional Basis Pursuit algorithms, ESD achieves higher sparsity. By establishing an enhanced sparse optimization model and solving it using the primal-dual interior-point method, the technique effectively identifies signal features. Validation through simulations and rotor test rig experiments shows that ESD avoids spectral aliasing and leakage, filtering out interference to produce clearer spectra than traditional MUSIC and Least Squares methods.

Introduction: The Challenge of Blade Monitoring

Aero-engine blades operate in extreme environments characterized by high temperatures, pressures, and centrifugal forces. Early detection of fatigue cracks is vital, as these faults alter a blade's vibration frequency. Currently, engineers rely on two primary methods: intrusive strain gauges and non-intrusive Blade Tip Timing (BTT).

While BTT is superior for monitoring every blade simultaneously without adding mass or wiring, it faces a significant signal processing hurdle: undersampling. Because sensors only capture data once per revolution, the sampling rate falls far below the Nyquist frequency. This leads to "ghost" frequencies (aliasing) and significant noise, making it difficult to extract true vibration parameters.

Schematic of BTT sensor installation and measurement signals. (a) shows the sensor layout on the casing; (b) illustrates the timing pulses generated as blades pass sensors.

Methodology: Enhancing Sparse Decomposition (ESD)

The researchers propose Enhancing Sparse Decomposition (ESD) to "fill in the blanks" of sparse BTT data. The method relies on the fact that blade vibrations are typically composed of only a few dominant frequencies, meaning the signal is "sparse" in the frequency domain.

Key Components of ESD:

  1. Sparse Representation: The signal is modeled as a combination of a few "atoms" from a massive redundant dictionary of sine and cosine waves.
  2. Reweighted L1 Minimization: Standard L1 minimization (Basis Pursuit) can sometimes produce blurred results. ESD uses an iterative reweighting strategy that penalizes smaller coefficients more heavily in each step, forcing the algorithm to converge on the true underlying frequencies.
  3. Optimization: A primal-dual interior-point method is employed to solve the weighted optimization problem, ensuring high precision even in noisy experimental conditions.
  4. Non-Uniform Sampling: The paper highlights that non-uniform sensor spacing (as seen in Figure 1a) captures more diverse signal phases, providing richer information for the ESD algorithm to reconstruct the signal accurately.

Results and Discussion

The ESD method was rigorously tested against Nonlinear Least Squares (NLS) and the MUSIC (Multiple Signal Classification) algorithm using both numerical simulations and a physical rotor test rig.

Comparative Performance

In asynchronous vibration simulations, ESD achieved a Root Mean Square Error (RMSE) of 0.0272, significantly outperforming NLS (0.0585) and MUSIC (0.0364). As shown in the spectral comparisons below, ESD produces a much cleaner output. While NLS and MUSIC suffer from spectral leakage and background noise, ESD identifies the vibration orders (0.1, 0.5, and 0.55) with high resolution and zero "floor" noise.

Comparison of vibration spectra: (a) NLS spectrum shows significant noise; (b) MUSIC spectrum improves resolution but retains some artifacts; (c) ESD spectrum provides the clearest identification of vibration orders with no leakage.

Experimental Validation

On a physical test rig with 16 blades—one of which contained a 1mm deep crack—the ESD method successfully identified the 833 Hz vibration signature using only 60 revolutions of data. This demonstrates the algorithm's robustness in real-world scenarios where data may be limited and noise levels are high.

Conclusion

The Enhancing Sparse Decomposition technique represents a significant step forward for BTT-based engine health monitoring. By leveraging the inherent sparsity of vibration signals and employing iterative reweighting, ESD overcomes the aliasing and noise issues that plague traditional undersampled data analysis. Future work will focus on optimizing sensor placement geometries and adapting the dictionary approach for even more complex, multi-mode vibration patterns found in full-scale aero-engines.

Summary updated

Abstract

摘要: 叶端定时是发动机叶片监测有效手段,相比于传统的应变片测量方式,叶端定时不仅可以同时监测所有叶片的振动状态,而且不会对叶片本身的振动造成影响。但是叶端定时采样数据存在高度欠采样的特点,针对该问题,提出基于增强稀疏分解(Enhancing sparse decomposition,ESD)的叶片振动参数辨识技术。稀疏分解是一种在冗余字典中对信号进行分解,通过求解优化问题得到信号在冗余字典下最稀疏解的信号处理方法。增强稀疏分解相比于传统的基追踪算法,可以得到更为稀疏的解。在建立了增强稀疏优化模型后采用原对偶内点法对优化问题进行求解,从而实现信号特征参数的有效辨识。将提出的基于增强稀疏分解的叶片振动参数辨识技术应用于不同类型的仿真数据和转子叶片试验台参数辨识,并与传统的MUSIC算法和最小二乘拟合相对比,提出的算法可以有效避免频谱混叠和泄露现象,并滤除其他频率成分的干扰,得到更清晰的谱图。
BibTeX
@article{wu2019enhancing,
  author = {Shuming Wu and Haifeng Hu and Zhibin Zhao and Yang Zhi-bo and Laihao Yang and Shaohua Tian and Xuefeng Chen},
  title = {{Enhancing Sparse Decomposition Based Blade Vibration Parameter Identification}},
  journal = {Journal of Mechanical Engineering},
  year = {2019},
  doi = {10.3901/jme.2019.19.019},
}
Figure