A team of researchers in China has now shown that a remarkably small artificial neural network can cut through that noise and pinpoint the onset of microseismic waves with unprecedented reliability for such a compact model. Writing in Frontiers in Signal Processing, Haichen Yin of the China University of Mining and Technology and Shandong Energy Group, together with colleagues from the China Coal Research Institute and other institutions, describe a lightweight convolutional neural network (CNN) that picks the first arrival of microseismic P-waves directly from noisy single-component waveforms recorded at a deep coal mine in Shandong Province.

The problem they tackled is a classic one in seismology. Every attempt to locate a microseismic event, and therefore to map where the rock is failing and where stress is concentrating, begins with picking the precise instant at which the first seismic energy reaches each sensor. Traditional algorithms, most famously the short-term average to long-term average ratio (STA/LTA) method and the Akaike Information Criterion (AIC), detect arrivals by scanning for jumps in sliding-window energy ratios or by matching waveform statistics. These methods work reasonably well in quiet conditions, but in a working coal mine they are easily fooled. When background noise itself produces energy spikes, threshold-based pickers can trigger early, latching onto noise bursts rather than true rock fracturing, and their accuracy depends heavily on hand-tuned thresholds.

Deep learning has transformed phase picking in global seismology. Networks such as PhaseNet, developed by Weiqiang Zhu and Gregory Beroza, and EQTransformer, introduced by S. Mostafa Mousavi and colleagues, now routinely detect earthquake arrivals with superhuman speed and impressive accuracy. But those models were designed for three-component seismic recordings sampled at around 100 hertz, and they typically contain millions of parameters. Coal mine microseismic data are a different beast entirely: usually recorded on single channels, sampled at much higher rates, and drowned in industrial noise. Running a heavyweight earthquake model on the resource-constrained monitoring hardware installed at a mine face is often impractical, and transferring a model trained on teleseismic data to mine waveforms suffers from severe domain mismatch.

The new network was designed specifically for this gap. Each 1,600-sample waveform trace, recorded at 500 hertz, is first converted into a two-dimensional time-frequency image using the short-time Fourier transform, with a 64-sample window and an 8-sample hop, and the logarithmic magnitude is normalized to zero mean and unit variance. The result is a 33 by 193 spectrogram in which a P-wave onset appears simultaneously as a jump in time and a rearrangement of energy across frequencies. This dimensionality-elevation strategy is the trick behind the model’s robustness: instead of relying only on amplitude changes in the time domain, which noise can mimic, the CNN can learn the characteristic spectral texture of a genuine rock-fracture onset.

The architecture itself is strikingly economical. Three convolutional blocks, with 16, 32 and 64 channels and kernel sizes of 5 by 5, 3 by 3 and 3 by 3, each followed by batch normalization, a rectified linear activation and max pooling, extract features hierarchically. The first, larger kernel acts as a smoothing filter that suppresses high-frequency random noise and captures broad waveform contours; the later, smaller kernels zoom in on the fine-grained local mutation that marks the arrival itself. An adaptive average pooling layer and three fully connected layers then output a single continuous number: the predicted first-arrival time. That regression strategy, rather than the per-sample classification used by networks like PhaseNet, eliminates the quantization error that classification methods introduce whenever an arrival is snapped to a discrete time window. The entire network contains just 58,529 trainable parameters, roughly 0.22 megabytes in 32-bit floating point, and requires only about 4.7 million multiply-accumulate operations per trace, making it genuinely suitable for real-time deployment on the edge hardware found in mine monitoring stations.

The training data came from an extraordinary natural laboratory. The Xinjulong Coal Mine lies in the deep Luxi Coalfield, where the No. 3 coal seam is buried 800 to 1,200 metres below the surface. It is a so-called three-soft seam, with a soft roof, soft floor and soft coal body, dipping at 15 to 30 degrees, and it is cut by the F311 fault, the largest in the mine’s history, with a throw of up to 16 metres. Where the working face approaches this fault, stress concentration coefficients reach 2.8 to 3.5 and microseismic event energy runs three to five times higher than in normal areas, making rockburst risk acute. Between June and October 2023, twelve underground triaxial accelerometers deployed around the 6305 working face recorded 3,963 microseismic events. From these, the researchers compiled 597 events with reliable manual P-wave labels, yielding 1,791 single-component waveform records that were split 60/20/20 into training, validation and test sets.

Trained with the Adam optimizer under a mean squared error loss and early stopping, the model achieved a mean absolute error of 45.61 sampling points on the test set, where one sample corresponds to just 2 milliseconds. The proportion of picks with errors not exceeding 100 milliseconds reached 80.27 percent. Against the traditional STA/LTA method, evaluated on 1,189 test samples, the CNN reduced overall picking error by about 15.7 percent, and the average relative error decreased by roughly 50 milliseconds, a systematic improvement visible as a distinct left-skewed shift in the distribution of error differences. The team also retrained a PhaseNet-style U-Net on the same data for a fair comparison; on the identical test set it and the classical AIC picker both lagged well behind, with accuracies within the 100-millisecond tolerance of only 66.3 and 65.2 percent respectively, compared with 78.0 percent in that analysis.

Perhaps most compelling is how the network behaves at the extremes of signal quality. In high signal-to-noise recordings, where traditional pickers drift slightly around the true onset due to local energy fluctuations, the CNN locks onto the beginning of the emergent waveform with near-perfect alignment to human labels, demonstrating its potential to replace subjective, fatigue-prone manual picking. In low signal-to-noise recordings, where shearer noise and conveyor vibration submerge the onset so thoroughly that the human eye struggles to find it, STA/LTA frequently fires early on noise bursts, while the CNN still produces predictions close to ground truth, apparently filtering the specific waveform patterns of P-wave arrivals out of the background clutter. The labelled dataset’s median signal-to-noise ratio of 17.5 decibels confirms that much of the data the model must handle is far from clean.

The implications reach well beyond one mine in Shandong. Accurate first-arrival picking is the foundation on which every downstream product of microseismic monitoring depends, from high-precision source localization to the identification of stress concentration zones and the early warning of rockbursts and outbursts. A picker this small can run continuously on distributed monitoring hardware, flagging hazardous rock fracturing in real time without bulky computing infrastructure. The authors caution that the residuals in their tests show no systematic early or late bias, which means the symmetric loss function does not skew located hypocentres, though coupling picking and location objectives in a single asymmetric loss remains future work. They plan to expand the dataset across diverse geological settings and further optimize the model for extreme operating conditions. As mining pushes ever deeper into stressed, fault-ridden ground, this study suggests that the quietest voices of the rock, once almost impossible to hear above the industrial din, can now be picked out by a neural network small enough to live underground alongside them.

Subject of Research: A lightweight convolutional neural network for automatic first-arrival picking of noisy coal mine microseismic data

Article Title: Seismic phase picking of coal mine microseismic data based on lightweight CNN

Article References: Yin, H., Zhang, X., Li, A., Zhu, W., Hou, Y., Chen, Y., & Lu, C. (2026). Seismic phase picking of coal mine microseismic data based on lightweight CNN. Frontiers in Signal Processing, 6, Article 1884114. https://doi.org/10.3389/frsip.2026.1884114

Image Credits: AI Generated

DOI: 10.3389/frsip.2026.1884114

Keywords: microseismic monitoring, convolutional neural network, coal mine safety, seismic phase picking, rockburst, deep learning, P-wave arrival, STA/LTA, Xinjulong Coal Mine, signal processing, regression prediction, edge deployment

The research team, led by Hao Yue and Xinghai Lei with colleagues Qiang Li and Fengqi Liu, tackled a long-standing blind spot in mining engineering. For decades, engineers have modelled the rock layers above a longwall working face using the Euler–Bernoulli beam theory, which assumes a beam’s cross-section stays planar and perpendicular to its neutral axis as it bends. That assumption works well for thin, slender roof strata. But it quietly ignores shear deformation, and as roof thickness grows, the overburden stops behaving like a slender beam and starts behaving like a short, stubby one. The researchers show that with a periodic weighting interval of 30 metres, once roof thickness exceeds roughly 3.75 metres, the slenderness ratio falls outside the range where Euler–Bernoulli theory remains valid.

Their solution was to rebuild the roof model on Timoshenko beam theory, which allows the cross-section to rotate independently of the beam axis and explicitly accounts for shear deformation and rotational inertia. Crucially, the team did not stop at a single two-dimensional slice. Previous studies typically analysed fracture either along the direction of face advance or along the face length, but rarely both at once. By establishing fracture models in both the advancing and lengthwise directions, the researchers could reconstruct the full three-dimensional fracture morphology of the overlying strata, characterised by two parameters for each layer: the advanced fracture distance, where the break occurs ahead of the face, and the fracture angle, which governs how the broken blocks interlock.

The mechanical setup couples the Timoshenko beam with a Winkler elastic foundation, treating the interaction between the main roof and the softer immediate roof beneath it as a bed of linear springs whose reaction is proportional to deflection. This is a deliberate improvement over the rigid-foundation assumptions of earlier models, because it better reflects the real stress state of the rock mass, even though rock is ultimately elastoplastic rather than purely elastic. During the first weighting, when the roof spans an unsupported opening for the first time, the roof above the coal is modelled as a beam on an elastic foundation while the span above the goaf is treated separately, with boundary and continuity conditions stitching the two solutions together. During periodic weighting, the goaf side loses its lateral constraint entirely, so the roof is treated as a plate with three elastically supported edges and one free edge.

Solving the resulting differential equations yields closed-form expressions for the bending moment along the roof, and the fracture position follows from the point of maximum moment. The sensitivity analysis reveals trends with direct operational consequences. The advanced fracture distance decreases as the elastic foundation coefficient and the roof’s elastic modulus increase, but grows with roof thickness, meaning a thicker roof breaks farther ahead of the working face and demands a longer advance support distance to protect the coal wall. The fracture angle tells the opposite story in places: it decreases with increasing weighting interval and foundation coefficient, but increases with roof thickness and elastic modulus, while Poisson’s ratio barely matters. Because the fracture angle controls whether the broken roof forms a stable voussoir beam or fails by sliding or rotation, these trends translate directly into hazard prediction. A larger fracture angle favours sliding instability; a smaller one favours rotational collapse.

Stage differences matter too. Both the advanced fracture distance and the fracture angle during periodic weighting come out smaller than during first weighting, a consequence of the goaf side’s lost constraint giving the roof more rotational freedom. The practical implication is that advance support ranges should be larger for the first weighting than for subsequent ones. The model also echoes field observations from China’s Shendong Mining Area, where weighting intervals under 20 metres correspond to stepped subsidence and sliding instability with severe dynamic loading, exactly what the theory predicts for short intervals and large fracture angles.

To test the theory against physical reality, the team built a two-dimensional similarity model on a 180 by 140 by 16 centimetre platform, scaled at 1:150 geometrically, with materials of sand, cement, and gypsum proportioned to reproduce the strength hierarchy of the Kouzidong strata. A digital image correlation system with two five-megapixel CCD cameras filming at 75 frames per second tracked the principal strain field as the model face advanced. The results were strikingly quantised: the immediate roof failed at a principal strain threshold of about 2 percent, the main roof at about 6 percent, and the key stratum at about 8 percent. Higher thresholds correspond to stronger layers, which means the key stratum accumulates the most strain energy before failing and releases the most energy when it does, a finding that flags key-stratum fracture as the moment demanding pre-emptive reinforcement.

The experiment reproduced the full failure sequence in miniature. The immediate roof collapsed first at 30 centimetres of excavation, the main roof broke for the first time at 45 centimetres and then fractured periodically at roughly 15-centimetre intervals, and at 125 centimetres the key stratum failed with an audible crack, dragging extensive overlying failure with it. When the researchers imported images of the final fracture morphology into MATLAB and binarised them using Otsu’s method, the overlying strata formed an approximately trapezoidal fracture pattern, with the open-off cut side fracturing at about 75 degrees and the advancing side at about 70 degrees, a five-degree difference attributed to the hydraulic supports altering the effective foundation stiffness rather than to image-processing error, which a threshold sensitivity test bounded at plus or minus 2 degrees.

The agreement between theory and experiment was close enough for engineering use. For the initial weighting, the model predicted an advanced fracture distance of 2.1 metres and a fracture angle of 76 degrees, against experimental values of 2.25 metres and 72 degrees. For the overall fracture angle, the theoretical 76 degrees versus the measured 70 degrees represents a relative error of about 8.6 percent, comfortably within the 10 to 20 percent natural fluctuation of field weighting intervals, and small enough that both values point to the same instability mode and the same control strategy.

The study closes the loop from prediction to control. At Kouzidong, the team designed a pressure-relief blasting scheme with fan-patterned boreholes in the roadways, inclined at 45, 60, and 75 degrees with depths exceeding 18 metres, and numerical simulation showed the peak abutment stress coefficient dropping from 2.1 to 1.92 after blasting. They further proposed dividing the 266-metre-long face into three zones: reinforced end control zones within 10 metres of each boundary using special supports with 40 to 70 percent more support area, a high-resistance support zone where intact roof demands high initial support forces to prevent bed separation, and a central pressure-stabilised zone covering the middle half of the face where fragmented roof is protected by moving supports in coordinated groups of two or three, orchestrated through a remote intelligent control platform. Together, the Timoshenko-based fracture model and the zonal control scheme offer mines facing thick, hard roofs a quantitative route to safer, more predictable extraction.

Subject of Research: Fracture mechanics and stability control of thick hard roof strata over longwall coal mining faces using Timoshenko beam theory and physical simulation

Article Title: Study on fracture mechanism and stability control of longwall working face with thick hard roof in coal mine

Article References: Yue, H., Lei, X., Li, Q., & Liu, F. (2026). Study on fracture mechanism and stability control of longwall working face with thick hard roof in coal mine. Discover Geoscience, 4(1), Article 393. https://doi.org/10.1007/s44288-026-00763-4

Image Credits: AI Generated

DOI: 10.1007/s44288-026-00763-4

Keywords: longwall mining, thick hard roof, Timoshenko beam theory, Winkler elastic foundation, roof fracture angle, first weighting, periodic weighting, digital image correlation, voussoir beam, pressure-relief blasting, hydraulic supports, coal mine safety