Study on Optimization of Infrasound Filtering Method for Coal Sample Failure under Load

Jianguo Zhang, Man Wang, Bing Jia, Yujie WangFrontiers in Earth Science 2022 · DOI

TL;DR

Researchers have developed a hybrid "EMD-wavelet" filtering method to isolate the subtle infrasound signals generated by coal samples under stress. By combining the adaptive nature of Empirical Mode Decomposition with the precision of wavelet thresholding, this approach significantly improves signal-to-noise ratios, offering a more accurate way to predict coal and rock failure in mining environments.

What Problem Does This Paper Address?

In the high-stakes environment of underground mining, predicting when a coal seam or rock wall might fail is a critical safety challenge. One promising "early warning" indicator is infrasound—sound waves with frequencies between 0.01 and 20 Hz. Because these waves have long wavelengths and low attenuation, they can travel long distances through rock, making them ideal for monitoring internal deformation.

However, there is a major hurdle: real-world mines are incredibly noisy. Mechanical equipment, ventilation, and human activity create a "soup" of acoustic interference. Conventional filtering methods (like band-pass filters) often fail when the noise frequency overlaps with the target infrasound signal. If we can’t clean the signal effectively, we can’t accurately predict the "abrupt" failure events that lead to disasters.

Methods

The authors propose and test an EMD-Wavelet Joint Filtering method. To understand why this is novel, we have to look at the two standard techniques it combines:

  1. Empirical Mode Decomposition (EMD): This is a self-adaptive method that breaks a complex signal down into several "Intrinsic Mode Functions" (IMFs) based on the signal's own time scales. It’s great for non-stationary signals but often leaves residual noise within the individual IMFs.
  2. Wavelet Filtering: This uses mathematical "wavelets" to decompose a signal into high and low frequencies, applying a threshold to "snip off" the noise. While powerful, its success depends heavily on choosing the right "basis function" and thresholding method.

The Hybrid Approach: The researchers first use EMD to decompose the signal into IMFs. Then, instead of just recombining them, they apply wavelet thresholding to each IMF component. Finally, they reorganize the cleaned components to reconstruct the signal. This "best of both worlds" approach aims to retain the abrupt characteristics of coal failure while scrubbing away the background noise.

Datasets

The study utilized two primary data sources:

  • Experimental Data: 30 coal samples (specifically 3# coal seam anthracite) were collected from various mines, including the Jincheng Sihe and Jiaozuo Jiulishan mines. These were processed into standard 50mm x 100mm cylinders and tested to failure under axial load.
  • Simulated Data: To quantitatively test the filters, the authors constructed a synthetic "mixed signal" consisting of 10 Hz and 25 Hz sine waves buried under 6dB of white noise. This allowed for precise calculation of the Signal-to-Noise Ratio (SNR) and Correlation Coefficients.

Key Insights and Ideas

  • The "Abruptness" Challenge: Coal failure isn't a smooth, linear process; it's characterized by sudden stress releases as internal pores and cracks snap. The authors emphasize that a filter must be able to handle these "abrupt" signals without smoothing them out, as these spikes are the primary indicators of impending failure.
  • Adaptive vs. Rigid: Standard filters are often too rigid. The beauty of the EMD-wavelet approach is its adaptivity—it doesn't require the researcher to pre-define the exact frequency bands of the noise, which is vital because mine noise is constantly changing.
  • Synergy over Substitution: The paper demonstrates that EMD alone is insufficient (it's "leaky" with noise), and Wavelets alone can be too aggressive. The "aha moment" is that applying wavelets to the IMFs provides a much cleaner separation than applying either method in isolation.

Key Results

The hybrid EMD-wavelet method outperformed both standalone wavelet and standalone EMD filtering across all metrics:

  • Signal-to-Noise Ratio (SNR): The joint method achieved an SNR of 13.44 dB, a 124% improvement over the raw mixed signal. For comparison, standalone EMD only reached 8.7 dB.
  • Correlation Coefficient ($r$): The joint method maintained a correlation of 0.98 with the original signal, indicating extremely high fidelity and minimal signal distortion.
  • Frequency Domain: In the frequency domain plots, the joint method was the only one to "thoroughly" remove noise peaks while perfectly preserving the target 10 Hz and 25 Hz signals.

Limitations and Future Work

While the results are compelling, the authors acknowledge that this study was conducted in a controlled laboratory and simulation environment.

  • Real-world Complexity: Actual mine noise is rarely "white noise"; it can be impulsive or periodic. Testing the algorithm against specific industrial noise profiles (like a continuous miner or ventilation fan) would be a logical next step.
  • Real-time Processing: The paper does not discuss the computational overhead of the EMD-wavelet method. For a real-world early warning system, this algorithm would need to run in near real-time on edge computing hardware.
  • Prediction Integration: The study focuses on filtering the signal. The next phase of research would involve linking these cleaned signals to specific "failure precursors" to create a functional automated warning system.

Summary updated

Abstract

Effective filtering of the infrasound signal generated by coal samples is the basis for realizing the prediction of the infrasound of coal sample damage. Based on the infrasonic signal test of the coal samples during the loading process, a simulation method was used to construct a mixed signal containing noise signals and infrasound signals. Three methods are used to filter the mixed signal, including wavelet filtering, EMD filtering, and EMD-wavelet joint filtering. The filtering effect was compared by correlation coefficient, signal-to-noise ratio, and frequency domain waveform graph. The comparison results showed that the EMD-wavelet joint filtering method had the highest correlation coefficient and signal-to-noise ratio after noise filtering, and the noise signal in the frequency domain waveform diagram was the most thorough. It provides a new method for filtering infrasound signals in the process of coal sample loading, which is greatly significant for improving the accuracy of infrasound prediction of coal sample damage.
BibTeX
@inproceedings{zhang2022study,
  author = {Jianguo Zhang and Man Wang and Bing Jia and Yujie Wang},
  title = {{Study on Optimization of Infrasound Filtering Method for Coal Sample Failure under Load}},
  booktitle = {Frontiers in Earth Science},
  year = {2022},
  doi = {10.3389/feart.2022.834234},
}
Figure