Mode Parameters Estimation of Vibration Signal Based on Aberrant Point Clustering and Elimination

Qingwei Ye, Dandan Wang, Yu ZhouResearch 2013 · DOI

TL;DR

This paper introduces a novel method for accurately estimating vibration signal parameters by combining the Levy polynomial fitting technique with k-means clustering to automatically identify and remove "outlier" data points. This improved approach significantly enhances the precision of mode parameter estimation, especially in the presence of strong noise or closely spaced vibration modes, outperforming the traditional Levy algorithm.

What Problem Does This Paper Address?

Accurate estimation of mode parameters (natural frequency, damping, and mode shape) is crucial for Structural Health Monitoring (SHM). These parameters are essential for updating finite element models, detecting damage, and evaluating structural performance. Traditional methods like the Levy algorithm, which fit a polynomial to the frequency response function (FRF), struggle when dealing with noisy data or when multiple vibration modes are very close together. In these scenarios, outliers in the data can lead to significant deviations in the estimated parameters, reducing accuracy and reliability. The paper aims to address this by developing a more robust parameter estimation technique that can effectively handle these challenging conditions.

Methods

The core of the proposed method is an iterative algorithm that refines the mode parameter estimation. It starts by fitting a Levy polynomial to the entire set of measured frequency response data. Then, it calculates the distance between each data point and the fitted curve. This set of distances is clustered using the k-means algorithm into two groups: points with small distances (likely good data) and points with large distances (potential outliers or "aberrant points"). The aberrant points are then eliminated, and the Levy polynomial is refitted to the remaining data. This process is repeated iteratively until the estimated parameters stabilize, ensuring that only reliable data points contribute to the final estimation. This outlier elimination step is the key innovation, improving upon the standard Levy algorithm.

Datasets

The paper utilizes simulated vibration signals for evaluation. Specifically, it uses a "viscous damper system simulation signal" denoted as S1 and S2. These signals are generated with controlled levels of white Gaussian noise (R) and varying parameters like natural frequencies (f1, f2, f3) and damping coefficients (d1, d2, d3). The authors specifically mention using a "fixed-two mode frequency signal S1" and a "variable frequency space S2" to test different scenarios, including dense modes. The sampling number is N=512.

Key Insights and Ideas

  • Outlier Detection is Key: The paper highlights that simply fitting a polynomial to all data points, even with a good algorithm like Levy's, is insufficient when noise or dense modes are present. Identifying and removing outliers is critical for accurate parameter estimation.
  • K-Means for Robustness: The application of k-means clustering to identify aberrant points is an elegant and effective way to automate outlier removal, making the process more practical than manual selection or experience-based thresholding.
  • Iterative Refinement: The iterative nature of the algorithm, where outliers are removed and the fit is re-evaluated, allows for a progressively more accurate estimation of the true underlying vibration modes.
  • Superiority in Challenging Conditions: The proposed method demonstrates a significant advantage over the standard Levy algorithm when dealing with strong noise and closely spaced vibration frequencies, which are common challenges in real-world structural analysis.

Key Results

The experiments show that the improved algorithm significantly reduces fitting errors compared to the standard Levy algorithm, especially at lower Signal-to-Noise Ratios (SNR). For instance, at an SNR of 1.8785 dB, the improved algorithm's fitting error was 0.1010, while the Levy algorithm's error was 0.3381. The paper also demonstrates that the algorithm performs better in identifying dense mode parameters, with average relative errors consistently lower than the Levy algorithm, particularly when modes are closely spaced. The improved algorithm maintains a fitting error within 0.2 even as signal strength decreases, indicating high stability and low noise sensitivity.

Limitations and Future Work

The authors acknowledge that the standard Levy algorithm's precision is good when there is no or very low noise. The proposed method aims to overcome the limitations of the standard Levy algorithm in noisy or dense mode scenarios. While the paper focuses on simulated data, the authors state that experiments show the algorithm can achieve good fitting effects under strong noise, dense modes, and high ratios of peak. Future work could involve applying and validating this algorithm on real-world structural vibration data to further confirm its effectiveness and robustness in practical engineering applications.

Summary updated

Abstract

A new mode parameter estimation method of vibration signal is put forward in this study. At first, the frequency response curve of vibration signal is fitted by Levy polynomial and the each distance between the fitted curve point and the frequency response curve point is calculated. Then the distance set is clustered by k-means algorithm into two classes. One class is clustered with smaller distance points and another class is clustered with larger distance points which are named aberrant point set. The class of larger distance points clustered will be eliminated and the new frequency response curve is obtained. At last, the new frequency response curve is fitted by Levy polynomial again and the new aberrant point set is eliminated again and so on. Finally, the fitting accuracy will be arrived according to the above algorithm. Plenty of simulation tests to vibration signals show that this algorithm can accurately extract mode parameters of the vibration frequency spectrum. It also confirms that in the different noise intensity and different distance between adjacent frequency cases, the precision of the algorithm proposed by this study is obviously superior to the existing Levy algorithm.
BibTeX
@inproceedings{ye2013mode,
  author = {Qingwei Ye and Dandan Wang and Yu Zhou},
  title = {{Mode Parameters Estimation of Vibration Signal Based on Aberrant Point Clustering and Elimination}},
  booktitle = {Research},
  year = {2013},
  doi = {10.19026/rjaset.6.4123},
}
Figure