Boundary delineation and stability assessment of post-inrush water-immersed coal pillars based on integrated microseismic monitoring and field analysis

Abstract

The stability of flooded coal pillars has long been a challenging issue in coal pillar research, especially during the recovery of production after water inrush disasters. Due to the harsh on-site conditions, it is difficult to directly collect samples for analysis, which brings numerous challenges in both design and production. Based on the engineering background of the 1314 working face of Xiaoyun Coal Mine, this paper proposes a clustering optimization algorithm and successfully uses microseismic data collected on-site to identify the boundary of the flooded coal pillar, validating the results through simulation comparisons. The study found that the flooded state within the coal pillar can be classified into saturated flooded zones, unsaturated flooded zones, and dry zones. The characteristics of the flooded coal pillar during the early stage of mining are more complex, with irregular variations in the flood boundary and local phenomena of sudden changes. Through the analysis of stress and delamination data, the primary controlling factors of this phenomenon are identified and the causes are explained. The research not only demonstrates the feasibility of using microseismic data to identify the flooding status of coal pillars but also provides valuable insights for analyzing the flooded state of coal pillars during the recovery of production after water inrush incidents. This study, particularly regarding coal pillar monitoring and safety control, presents new challenges.

Introduction

Microseismic monitoring has become a vital tool in coal mining operations, playing an essential role in various aspects of modern coal production. Applications of microseismic technology include minequake early warning and disaster prediction, monitoring the stability of surrounding rock, fault and structural interface identification, dynamic load monitoring from rockbursts, and water inrush hazard detection. Given the complex geological conditions, extensive spatial distribution of coal seams, and the numerous unknown factors inherent in underground mining, many researchers have adopted microseismic techniques as a primary method for studying key production challenges in coal mines.

Numerous studies have been conducted in this field. For example, Cai et al.¹. investigated the relationship between mining layouts and fault activation using microseismic monitoring to identify and verify fault reactivation, providing real-time insights into microcrack propagation within coal and surrounding rocks. Zhou et al.². developed a 3D microseismic data-driven jointed rock mass damage model that integrates seepage, damage, and stress fields to evaluate the stability of roof pillars during the transition from open-pit to underground mining. Liangping Hu et al.³. applied hydraulic slotting techniques at the Hujiamao Coal Mine and proposed mechanisms for dynamic load resistance and static load pressure relief. Leake et al.⁴. analyzed seismic activity characteristics induced by the retreat of two longwall panels. Wang et al.⁵. studied the spatial distribution of high-energy microseismic events related to coal and gas outbursts in high tectonic stress zones. Long et al.⁶. investigated the development of water-conducting fracture zones under deep burial and high-pressure aquifers through combined microseismic monitoring and physical modeling. Han et al.⁷. examined the mechanism of dynamic ground stress events using theoretical analysis, MS monitoring, and numerical simulation at the Tengdong Coal Mine. Wang et al.⁸. analyzed the rupture mechanisms of MS events and the spatial distribution of localized intense energy transfer and concentration structures based on rockburst events during roadway development. Zhang et al.⁹. proposed a multi-step method (CNN-KA) that integrates convolutional neural networks (CNNs), K-means clustering, and the Akaike information criterion. These studies represent only a portion of microseismic applications across different fields of coal mining research, highlighting the vast potential and promising prospects for further exploration. The use of microseismic data to validate simulation results has become an essential component in many engineering studies^{10,11,12,13,14,15,16}.

In the context of water-rock interactions, extensive research has also been conducted on microseismic characteristics. For instance, Yang et al.¹⁷. performed microseismic energy density analysis to identify water-conducting pathways and detect fracture points within rock layers. Chen et al.¹⁸. utilized a self-adaptive loading system and interpreted the water inrush process using microseismic signatures. Bian et al.¹⁹. integrated microseismic monitoring with transient electromagnetic methods to evaluate and verify the effectiveness of hydraulic fracturing in controlling coal and gas outbursts. Lian et al.²⁰. investigated the feasibility of using microseismic anomalies to predict water inrush events based on changes in water yield. Sun et al.²¹. analyzed the formation of roof water inrush pathways at working face 18,401 of the Xiqu Coal Mine. Shang et al.²². proposed a microseismic data clustering approach for early warning of floor water inrushes. Yu et al.²³. identified fluid-related hybrid-frequency microseismic signals at the Wangjialing Coal Mine, highlighting their potential as precursors of floor water inrush. Zhang et al.²⁴. conducted uniaxial compression experiments on coal samples with varying moisture content to study the effects of water on physical-mechanical properties and the generation of electromagnetic and microseismic signals. Jia et al.²⁵. compared pre- and post-mining microseismic activity and assessed the influence of large-scale weak zones on coal mining under pressurized aquifers. Gai et al.²⁶. proposed a new microseismic-based method to assess spatial damage characteristics of the floor and associated water inrush risks. Additional relevant studies can be found in²⁷and others.

Moreover, microseismic monitoring facilitates the detection of small-scale fractures triggered by stress concentration or variation within the rock mass. By analyzing source location, energy release, frequency content, and integrating geological context, it is possible to identify rock failure signals or precursory stress accumulation, thus providing critical data for disaster forecasting and engineering decision-making. This capability makes microseismic monitoring indispensable for both macro- and micro-scale investigations.

The subject of this study is the sectional coal pillar of a longwall panel. Following a water inrush accident, the coal body in the working face was softened due to prolonged inundation, creating several technical challenges prior to retreat mining. For instance, water-immersed coal pillars exhibit significantly different structural behavior compared to untreated ones. Water exposure alters the physical and mechanical properties of the coal-rock mass, resulting in non-equilibrium stress states and complex loading responses. Determining the inundation characteristics and spatial extent of such pillars remains a longstanding engineering difficulty, introducing considerable uncertainty into mine design, retreat planning, pillar reservation, and personnel safety.

Taking 1314 working face as a case study, this research addresses the challenges posed by complex goaf conditions and numerous uncontrollable factors following post-inrush recovery, which make direct sampling infeasible. As a result, remote sensing and spatial monitoring techniques become crucial for investigating the mechanical evolution of inundated coal pillars. Based on microseismic monitoring data collected during mining, this study utilizes statistical analysis and algorithmic identification methods to characterize the actual condition of the pillars, reveal relevant patterns, and provide valuable insights for the design and safe operation of post-inrush recovery projects.

Engineering background

This study examines the coal pillar in the 1314 working face of the Xiaoyun Coal Mine, which has been affected by water inrush. The Xiaoyun Coal Mine is situated in Jining City, Shandong Province, China. The 1312 goaf is located to the north of the working face, the village protection coal pillar lies to the south, the 1314 belt trough contact lane is to the west, and the boundary coal pillar of the mining area is to the east. The working face has a strike length of 955 m, a dip length of 219 m, an average coal seam thickness of 3.9 m, and an average dip angle of 15°. The coal seam structure is relatively simple. A basic overview of the 1314 working face is shown in Fig. 1.

Fig. 1

Basic overview of the 1314 working face. (A map of China showing the location of Jining City, Shandong Province (highlighted in red). The map was created using Python with Matplotlib v3.10.3 (https://matplotlib.org/), Cartopy v0.24.1 (https://scitools.org.uk/cartopy/), and NumPy v2.3.1 (https://numpy.org/). Geographical features and administrative boundaries are based on Natural Earth data accessed via Cartopy’s feature interface).

On September 10, 2018, the Xiaoyun Coal Mine first experienced slight water seepage at the 1313 working face, and within a few hours, the water inflow increased to 40–50 m³/h. Due to the failure to implement effective emergency measures, the water inflow surged to 900 m³/h by 2:00 PM on September 11. By 10:00 AM on September 13, the estimated water inflow had reached approximately 3673 m³/h. Eventually, the mine’s central pump house failed, leading to a water inrush and flooding incident.

Numerical simulation of water inrush impact on coal pillar stability

The research on the impact of water immersion on coal pillars is crucial for determining the stability and failure modes of coal pillars under water immersion. However, in coal mine production practice, due to the harsh environment of water-inundated goaf areas, it is nearly impossible to conduct on-site sampling and analysis. As a result, methods for analyzing coal pillars are limited to theoretical calculations, numerical simulations, and the analysis of field monitoring data. This section aims to simulate the effect of water influx on coal pillars, with the goal of providing a theoretical basis for the microseismic analysis in the next chapter. The parameters and model information in the simulation are referenced from²⁸.

The fluid dynamics within the 1312 goaf are described by the Navier-Stokes equations, which serve as the fundamental governing equations for momentum and mass conservation in incompressible flow regimes. These equations capture the complex interactions between pressure, viscous forces, and external influences in the airflow field of the goaf:

$$\rho \frac{{\partial u}}{{\partial t}}+\rho (u \cdot \Delta )u=\Delta \cdot [ - {p_2}I+K]+F$$

(1)

$$\rho \Delta \cdot u=0.$$

(2)

Equation (1) defines the momentum balance of the fluid system. The left-hand side comprises the transient and convective terms, which together characterize the evolution of fluid momentum in both time and space. The right-hand side includes the pressure gradient, viscous stress components, and body forces, representing the internal and external influences acting on the fluid within the domain. In contrast, Eq. (2) governs the conservation of mass for the fluid phase, ensuring continuity by accounting for any changes in fluid density or volume during transport and deformation.

In the non-elastic region, fluid behavior is described by the Brinkman flow equation, which extends Darcy’s law by incorporating viscous shear effects, thereby enabling the modeling of flow in porous media with moderate permeability. This formulation provides a more accurate representation of the flow field where both permeability and viscous interactions significantly influence the velocity distribution.

$$\frac{1}{\varepsilon }\rho \frac{{\partial {u_2}}}{{\partial t}}+\frac{{\rho ({u_2} \cdot \nabla ){u_2}}}{{{\varepsilon ^2}}}=\nabla \cdot [ - {p_3}I+K] - \left( {\mu {{{k}}^{ - 1}}+\beta \rho \left| {{u_2}} \right|+\frac{{{Q_m}}}{{{\varepsilon ^2}}}} \right){u_2}+F$$

(3)

$$\rho \nabla \cdot {u_2}={Q_m}.$$

(4)

Equation (3) characterizes the conservation of momentum in porous media systems. On the left-hand side, the terms account for both temporal variations and inertial effects, reflecting changes in fluid velocity over time as well as the momentum shifts induced by velocity gradients. The right-hand side includes multiple force contributions: pressure and viscous stress components, resistance terms, and external body forces acting on the fluid F. The parameter β denotes the nonlinear resistance coefficient, while the term − p₃I represents the pressure-induced driving force that propels fluid flow. The viscous stress is described by the tensor K, and the resistance term μk⁻¹ captures the combined influence of fluid viscosity and medium permeability. Additionally, the nonlinear resistance component $\beta \rho \left| {{u_2}} \right|$ accounts for inertial resistance effects inherent to flow through porous structures. The porosity-related resistance term $\frac{{{Q_m}}}{{{\varepsilon ^2}}}$ reflects how porosity variations modulate flow resistance, while Q_m denotes the mass source term within the system. Equation (4) provides a generalized expression for mass conservation within the porous medium.

The simulation framework adopts different governing equations tailored to the physical characteristics of each region. Specifically, fluid flow within the elastic region is described by Darcy’s law, which assumes a linear relationship between flow velocity and pressure gradient, and is appropriate for low-velocity flow in porous media. This formulation captures the fundamental seepage behavior under elastic deformation conditions, where inertial and viscous effects are negligible:

$$\frac{\partial }{{\partial t}}(\varepsilon \rho )+\nabla \cdot \left( {\rho u} \right)={Q_m}$$

(5)

$$u= - \frac{k}{\mu }\nabla p.$$

(6)

Equation (5) incorporates several key physical parameters, including the porosity ε, the fluid velocity field u, and the fluid (water) density ρ. The term ∇⋅(ρu) represents the divergence of mass flux. The permeability of the porous medium is indicated by ${k}$, and μ corresponds to the dynamic viscosity of the fluid. This equation serves as the mass conservation equation for fluid flow in porous media. On the left-hand side, it includes the temporal change term and the convection term, which together capture the time-dependent variation of fluid density within the pore structure, as well as the changes induced by advective transport. The right-hand side represents any external or internal mass sources. Following this, Eq. (6) provides the mathematical formulation of Darcy’s law, governing flow behavior under conditions of low Reynolds number and linear pressure-velocity relationships.

The model divides the coal pillar into two regions: the damaged zone and the undamaged zone²⁹which are described using the Brinkman equation and the Darcy equation, respectively. The Brinkman equation is used to simulate the response of the coal body in the damaged zone under the effect of water inrush, while the Darcy equation describes the water inrush process in the undamaged zone. The Navier-Stokes equations are employed to simulate the movement of water flow during the water inrush process. The model adopts a unidirectional coupling approach, establishing fluid-solid coupling for the entire system at the interface between adjacent regions, based on equal velocity and pressure conditions. By analyzing the fluid flow characteristics, such as velocity, direction, pressure, and viscosity, derived from the Navier-Stokes equations in the water flow region, the pressure and velocity fields exerted by the water flow on the coal pillar surface can be calculated. Additionally, by integrating fluid dynamics theory, the shear effects of the water flow on the coal pillar surface and the impact of water-induced softening can be further derived. This process relies on an interaction model at the fluid-solid interface to quantify the mechanical effects of water flow on the coal pillar.

In this study, numerical simulations are performed using the finite element method (FEM) for both spatial and temporal discretization, which is a widely used approach in geotechnical and hydro-mechanical modeling. The governing partial differential equations, specifically Eq. (1) through (6), are discretized using an implicit time integration scheme based on the Backward Differentiation Formula (BDF), which provides numerical stability for nonlinear and coupled processes. To address the system’s inherent nonlinearities, the Newton-Raphson iterative method is applied to simultaneously solve key variables such as seepage velocity and pore water pressure within the coupled hydro-mechanical context. The simulation results are presented in Fig. 2.

Fig. 2

Evolution of internal stress in a coal body during time-dependent water immersion of a coal pillar.

The left panel of Fig. 2 illustrates the time-dependent distribution of pore water pressure within the coal body during the water immersion process of the coal pillar. This simulation adopts a transient state approach and, in conjunction with on-site data from the 1314 working face, sets the water pressure for the water inrush²⁸. During the simulation, it is assumed that the water pressure remains constant. Based on the actual conditions of on-site water inrush disaster response, the calculation time is set to 36 h, with a time step of 1 h. For visualization, results are displayed every 6 h. To prevent convergence issues in the early stages of the transient simulation due to excessively high initial water pressure, a linear increase in water pressure is applied during the first time step until it stabilizes at the set water inrush pressure. To improve the clarity of the visualization, the displayed water pressure changes are binarized, and dynamic results are shown. The results vary according to the changes in the pore pressure of the coal body due to the water inrush, with the criterion being the presence or absence of pore water pressure in the transient state.

As can be seen from the figure, after a water inrush incident, the pore water pressure inside the coal body increases as the flooding time progresses. Initially, under the influence of the inrush pressure, water flows through the goaf’s internal pathways and gradually saturates the void space. Upon reaching the stop line of mining, the flow direction changes, and water begins to seek alternative discharge routes. At this stage, turbulent flow dominates the goaf, introducing hydrodynamic complexity that may affect the accuracy of numerical simulations.

Once saturation is achieved, the combined effects of hydraulic pressure, inertia, and viscous forces amplify the pore pressure exerted on the coal mass. The simulation results reveal that longer inrush durations lead to more pronounced and persistent pore pressure responses even after the water source dissipates. The peak pore pressure is observed during the active inrush phase and increases as the event duration extends.

This process exhibits a clear directional characteristic, with pressure transferring from the front to the rear of the coal pillar while gradually attenuating. In the final three simulated time intervals, the rate of pressure change decreases significantly, suggesting increasing resistance to pressure propagation and nonlinear transmission behavior. Eventually, the zones with minimal or no water influence form an arc-shaped boundary in the spatial domain.

Identification of coal pillar water immersion boundaries

Microseismic monitoring is essential for identifying and analyzing water-affected zones in coal mines. These zones are typically characterized by fracture or deformation of coal pillars and surrounding strata caused by water infiltration or changes in coal seam moisture. Such conditions pose considerable safety threats, including roof collapses, water inrushes, and dynamic ground pressure events. Microseismic technology enables the real-time detection of subtle seismic signals within these zones. When water-induced stress alterations or incipient fractures occur in coal or surrounding rock, microseismic signals provide early indications of instability. These signals support accurate localization of affected areas and facilitate timely hazard mitigation. In addition to tracking the evolution of inundated zones, microseismic monitoring allows for the quantitative assessment of their mechanical stability, thereby helping to prevent large-scale accidents during mining operations. The integrated application of microseismic data significantly enhances the precision and responsiveness of mine safety management, offering strong support for safe production.

The microseismic dataset used in this section was obtained from panel 1314 at the Xiaoyun Coal Mine. Data were continuously recorded over a 14-month period, covering the mining process from initial retreat through approximately 860 m of advancement. The dataset is presented in Fig. 3.

Fig. 3

Heatmap of energy distribution within the coal pillar derived from microseismic monitoring data.

Figure 3 presents a heatmap of energy distribution within the coal pillar, generated from microseismic monitoring data. The energy pattern displays pronounced spatial asymmetry, with concentrated clusters in specific regions and sparse activity elsewhere. The energy distribution of microseismic events within the coal pillar is the highest near the retreat end. Several distinct high-energy zones indicate a preference for microseismic activity in localized areas. These clusters are mainly situated in the southeast and diminish toward the northwest, suggesting a directional trend.

Disregarding minor geological structures, this pattern is attributed primarily to the effects of water immersion. Based on the observed energy distribution, three hypotheses are proposed: (1) water infiltration induces microseismic clustering in three main zones, likely related to roof fracturing and hydromechanical processes; (2) Based on the simulation results and the energy heatmap, it can be inferred that the southeastern cluster corresponds to the driest region; and (3) within each cluster, lower-energy events dominate the center, while higher-energy events occur at the periphery, possibly reflecting moisture gradients.

As reported in³⁰coal softening under saturation is gradual. As water progressively fills pore spaces, the mechanical properties of coal degrade (Fig. 4). Water presence also reduces fracture energy release via lubrication, chemical interactions, and S-wave attenuation³¹. Consequently, saturated zones emit the least seismic energy, followed by partially saturated zones, while dry coal exhibits the highest release. Overall, saturated regions maintain weak but stable energy levels; dry regions are similarly stable but strong; and partially saturated zones show energy levels that increase either linearly or nonlinearly with water content.

Based on these findings, a clustering analysis is conducted to classify regions into saturated, partially saturated, and non-immersed zones, thereby delineating the boundary between water-affected and unaffected coal.

Cluster analysis is an unsupervised learning approach used to partition a dataset into distinct groups based on measures of similarity or distance. The underlying principle is to group data points such that intra-cluster similarity is maximized while inter-cluster similarity is minimized. In this study, the K-means clustering algorithm is adopted. The algorithm begins by predefining the number of clusters (K), selecting K initial centroids at random, and assigning each data point to the nearest centroid. The centroids are then recalculated for each cluster, and the process is repeated iteratively until convergence is achieved, i.e., when no further changes in cluster assignment occur²². The objective of K-means is to minimize the following cost function(7):

$$J=\sum\limits_{{{\text{i}}=1}}^{K} {{{\sum\limits_{{{{\text{x}}_{\text{j}}} \in {C_{\text{i}}}}} {\left\| {{x_j} - {\mu _i}} \right\|} }^2}}$$

(7)

Where K is the number of clusters, C_i denotes the i-th cluster, μ_i is the centroid of cluster C_i, and x_j represents the data points assigned to cluster.

Fig. 4

Variation in the physical and mechanical properties of coal before and after water immersion.

The choice of initial centroids has a significant impact on the final clustering results when using the K-means algorithm. If the initial centroids are chosen poorly, it may lead to a series of problems, such as falling into a local optimum, having cluster centroids that are not representative, unstable clustering results, and susceptibility to noise and outliers. Therefore, properly addressing the sensitivity of the initial centroid selection is crucial.

In the previous section’s simulation, we gained a preliminary understanding of the flooding evolution process of the coal pillar under the influence of water inrush. Through the simulation, we can obtain the general range and characteristics of the flooding boundary. To better handle the issue of selecting initial centroids, we use the simulation results as constraints in the clustering process. We then perform multiple computations on the randomly generated centroids and assign an overall score based on the results they produce. The parameters for the composite score include energy weight, spatial weight, spatial-energy ratio, and parameter space distance.

The selection of energy and spatial weight parameters was determined through a parameter space scan and performance evaluation, using the theoretical range and the simulated flooding boundary evolution results as constraints. The results, as shown in the Fig. 4, indicate that the model selected in this paper has the highest composite score. After obtaining the optimal parameters, we further calculated the parameter space distance and analyzed the correlation between all parameters and model performance. As shown in the Fig. 5, the correlation with distance is the highest, reaching 0.789, indicating that it has the greatest impact on model performance. In contrast, the correlation with spatial weight is relatively low, at only 0.369, suggesting that it has a smaller effect on model performance (Fig. 6).

Fig. 5

Results of parameter weight selection.

Fig. 6

Correlation between model performances.

The results also validate the rationality of the parameter selection in this model. Since the clustering in this study is based on energy data, parameters related to energy are assigned higher weights. Finally, we present the comprehensive scoring of the random parameters after multiple iterations of the K-means clustering method, as shown in the Fig. 7. The star markers in the figure represent the weight parameters chosen by the model, and the corresponding cluster centroids are also determined. The K-means clustering method, iterated using this approach, effectively avoids the sensitivity issue associated with the initial centroids.

Fig. 7

The combined score of random parameters after multiple clustering iterations.

Since the K-means algorithm typically assumes that clusters in a dataset are spherical and that the clusters’ sizes and densities are roughly equal, this assumption can lead to poor clustering performance when dealing with nonlinear data. Therefore, this paper strengthens K-means clustering by incorporating a Random Forest machine learning approach. Random Forest can handle classification and regression tasks by aggregating multiple decision trees, making it effective at dealing with complex and nonlinear data. Combining Random Forest with K-means improves the clustering process in the following ways: (1) It performs better on nonlinear decision boundaries, allowing the clustering process to no longer be limited by linear separations, thus helping Basic K-means identify more diverse clusters; (2) By selecting and weighting important features, it allows for greater emphasis on features that are difficult to separate, thereby improving the clustering performance of K-means; (3) Through multiple random selections of features and data points, it helps reduce overfitting and enhances the stability of the clustering process.

The Fig. 8 demonstrates the changes in clustering performance after incorporating Random Forest, showing a significant improvement in identification efficiency. Many regions that were previously unrecognized due to the limitations of the K-means algorithm have been significantly updated.

Fig. 8

Comparison of results between basic K-means and K-means + random forest.

Similarly, to improve the model’s performance and accuracy, morphological operations were further incorporated to address issues related to isolated points and discontinuous regions, while smoothing was applied to eliminate noise³¹. Physical consistency constraints were also applied to ensure that the classification results align with geophysical principles. With these improvements, the model’s efficiency was significantly enhanced. The overall optimization effects of these improvements are shown in Fig. 9.

Fig. 9

Display of model optimization effects under global enhancement.

An optimization algorithm was applied to perform computations for the recognition task in this study. Three types of data were identified: saturated water immersion, unsaturated water immersion, and dry zones. The resulting detection and recognition outcomes are shown in Fig. 10.

Fig. 10

Spatial distribution of water-immersed regions after clustering optimization.

Figure 10 shows the distribution of water immersion areas after clustering optimization. As depicted in the figure, the dry areas are more concentrated at the tail end of the coal pillar compared to other regions. In contrast, the middle section of the coal pillar exhibits a larger distribution of unsaturated immersion areas, while the front end of the coal pillar shows a more extensive distribution of saturated immersion areas.

From the clustering result diagram, it can be seen that the clustering boundaries are limited. To identify the macro-boundaries of the saturated water immersion zone, unsaturated water immersion zone, and dry zones, and provide useful information for the working face production design, the result shown in Fig. 10 is not sufficient. To better delineate different regions and identify the water immersion boundaries of the coal pillar, an algorithm related to boundary point detection was added. As a result, the outcomes shown in Fig. 11 were obtained.

Figure 11 illustrates the water immersion boundaries of the coal pillar obtained through boundary point detection. While the results are similar to those derived from clustering, the boundary delineation here is more scientifically rigorous. As shown in the figure, several localized dry zones are observed within the coal pillar. Except for the relatively dense dry zone at the coal pillar tail end, other areas exhibit a mix of three regions: saturated immersion zones, partially saturated immersion zones, and dry zones.

In the front section of the coal pillar, the saturated immersion zone covers the largest area, indicating that this region is most significantly affected by water, which is consistent with the simulation results. However, the mixing of the three regions is more pronounced, suggesting that the stress-strain variations within the coal pillar during the retreat mining process at the excavation area are more complex.

Based on these results and in reference to studies^32,33it is believed that this phenomenon may be attributed to: the saturation of the front section of the coal pillar, which, when combined with additional disturbances from roof pressure, dynamic loading, and other factors during the mining process, leads to increased plastic deformation of the water-softened coal. As a result, drainage becomes inevitable. This, in turn, causes a recovery in the pillar’s strength, an increase in brittleness, and a reduction in lubrication between fractures. Ultimately, these processes explain the observed mixing of the three regions in the front section of the coal pillar, as seen in Fig. 11.

Fig. 11

Coal pillar water immersion boundary identified through boundary point detection.

Furthermore, by setting the weights for different regions within a local range, macro-level detection of the coal pillar water immersion boundary can be performed. Through computation, the detection results for the water immersion boundary from the macro perspective of the coal pillar were obtained. To demonstrate the reliability of the results, a comparison between the simulation results and the detection outcomes was made, as shown in Fig. 12. The features of the coal pillar water immersion area obtained through the above algorithm are as follows:

• 1) The tail end of the coal pillar is least affected by water immersion and is mostly a dry zone.

• 2) The middle section of the coal pillar is predominantly an unsaturated water immersion zone, with dry zones and saturated water immersion zones each distributed, showing a transitional characteristic from water immersion to dry zones.

• 3) During the early recovery phase, the saturated water immersion zone dominates within the coal pillar, but the dry zone seems to outnumber the unsaturated water immersion zone. This does not align with the theoretical drainage process of the immersed coal body, indicating a certain degree of abrupt change. This suggests that the characteristics of the coal pillar after water immersion in the early recovery phase are more difficult to extract and understand.

Consequently, the boundary recognition process based on the clustering optimization algorithm has been completed. The complete workflow for the boundary recognition algorithm is shown in Fig. 13. Based on the identified features of the coal pillar water immersion area, it is necessary to further analyze other parameter features of the coal pillar to better understand the reasons behind these characteristics. This will also provide better conclusions for the relationship between the coal pillar water immersion boundary and coal pillar stability, as proposed in this article. Therefore, further spatiotemporal analysis of the coal pillar will be conducted in the next section.

Fig. 12

Comparison between simulation results and boundary detection results.

Fig. 13

Overall workflow for identifying water immersion boundaries in coal pillars using microseismic data.

Spatiotemporal characteristics of water-immersed coal pillars

Water-immersed coal pillars tend to exhibit anomalous behavior during the initial stages of retreat mining, introducing added complexity to early-stage operational planning and water hazard mitigation. These irregularities may adversely impact production safety. To better elucidate the spatiotemporal evolution of water infiltration within the coal pillar and to support safer mining practices, this section adopts a combined approach based on field monitoring and microseismic data analysis.

Building upon the energy distribution heatmap and immersion boundary delineation presented in the previous section, spatial patterns in the distribution of microseismic events can be initially identified. In line with the clustering-based framework, further statistical analysis of these energy events is conducted. The results are illustrated in Fig. 14.

Fig. 14

Statistical analysis of microseismic events in panel 1314.

From Fig. 14, it can be observed that the frequency of microseismic events, as shown in the upper-left graph, follows a pattern of first decreasing, then increasing, and finally decreasing again. The region with the lowest frequency of events is located in the middle section of the coal pillar, while the highest frequency occurs at the tail end of the coal pillar. The upper-right graph illustrates the relationship between mining depth and average energy, where the values are smallest in the middle section of the coal pillar and largest at the tail end. The lower-left graph reveals that both the frequency and energy are highest at the tail end of the coal pillar. However, energy peaks in the 700–800 m range, while the highest frequency is observed between 600 and 700 m. A higher frequency indicates that fractures in the coal body in this area are more actively developed, while a higher energy accumulation suggests that this region stores relatively more energy than other areas. The region between 300 and 400 m, located in the middle of the coal pillar, exhibits the lowest energy, with event frequency ranging from moderate to weak. This suggests that microseismic events are less frequent in the middle section, and the energy of each detected event is weaker. This is an unusual characteristic and is speculated to be closely related to water intrusion. The lower-right graph shows that moderate-energy events are most frequent during the early stages of mining, small-energy events are most frequent in the middle section, and large-energy events are most frequent at the tail end. Based on the analysis of factors that could influence the coal pillar, the indicators during the early mining stage are generally at intermediate levels, while those in the middle section are the lowest. Since water significantly weakens wave propagation and energy release, and considering the microseismic characteristics of other regions, it can be inferred that the middle section of the coal pillar is heavily affected by water intrusion, while the tail end exhibits the highest statistical values. After water intrusion, the number and energy of microseismic events generally decrease. However, based on the results from early simulations and the identification of the water intrusion boundaries, it is apparent that the region most affected by water during the early mining stages does not exhibit the lowest energy release. This is also an unusual phenomenon, which aligns with the results from statistical and clustering analysis. As mining progresses, understanding the mechanical properties of the coal-rock mass in the front section of the water-impacted coal pillar becomes crucial for ensuring safety during production. To better analyze the state of the coal pillar during the early mining stage, further analysis combining stress and displacement monitoring data is required.

Based on the statistical characteristics of microseismic energy presented above, stress and displacement data during the mining face recovery period will also be statistically analyzed. The goal is to verify the characteristics of the stress, displacement, and microseismic data and make reasonable inferences to explore the specific state of the coal pillar during the recovery process. The statistical data for stress and displacement are shown in Fig. 15.

Fig. 15

The relationship between stress, displacement data, and mining depth at the 1314 working face.

Figure 15 illustrates the relationship between stress, displacement data, and mining depth. As shown in the figure, the stress in the middle section of the coal pillar remains the lowest, while the displacement data is the highest. This result confirms our hypothesis: due to the impact of water intrusion, the plastic deformation of the coal pillar in the middle section is more significant, leading to greater roof displacement. The stress distribution at the tail end of the coal pillar is highly dispersed, showing no clear pattern. However, the displacement data indicates relatively small values. This suggests that the roof displacement at the tail end of the coal pillar is minimal, indicating that the coal pillar provides strong support.

Based on the above, it can be inferred that the impact of water intrusion weakening on the tail end of the coal pillar is relatively small. Over time, most of the water within the coal pillar has evaporated, causing a significant decrease in moisture content, and the strength of the coal body has essentially returned to normal. In combination with the boundary detection results, it is believed that the drainage and moisture recovery process in the front section of the coal pillar is discontinuous and irregular. The concentrated appearance of the saturated zone confirms that this region was indeed most severely affected by early water intrusion. Additionally, the results from stress monitoring and displacement data suggest that early mining in the working face has a significant impact on the water intrusion characteristics of the coal pillar. The load applied during the early mining stage is a key parameter for the drainage effect of the coal pillar in the saturated water intrusion zone. As mining progresses and the working face enters the unsaturated water intrusion zone, the load generated by the mining face starts to have a diminished effect on further drainage, making it more difficult for the internal moisture of the coal pillar to be expelled through compressive deformation.

Conclusion

Based on the analysis above, it is concluded that the use of microseismic data in water-intruded coal pillars is feasible. Microseismic events within the water-intruded coal pillar can be categorized into three types: the saturated water intrusion zone, characterized by stable low energy; the dry zone, characterized by stable high energy; and the unsaturated water intrusion zone, where energy changes either linearly or nonlinearly.

Building on this, a clustering approach was employed to identify the water intrusion state in different regions of the coal pillar, which also led to the identification of the water intrusion boundaries within the coal pillar. This result can be applied to address the engineering challenge of determining the water intrusion state within the coal pillar when production is resumed after a water inrush. It also provides a valuable reference for future production designs in areas affected by similar geological disasters.

Furthermore, through a comparative analysis of the microseismic energy distribution map and water intrusion boundary identification results, it was found that the front section of the coal pillar in the working face exhibits distinct characteristics. In this region, saturated water intrusion, transitional stages, and dry conditions alternate and mix. In some locations, abrupt transitions between the saturated water intrusion zone and the dry zone are observed. By comparing the microseismic energy, stress, displacement, and simulation data within the coal pillar, it was concluded that the cause of this phenomenon is the load effect induced by early mining activities, which leads to drainage in the coal pillar. However, the drainage process is complex. The front section of the coal pillar is most severely impacted by water intrusion, and the interaction between coal and water is both complex and uncertain. Once the coal body reaches its saturated moisture content, subsequent drainage is influenced by a variety of factors, such as the direction of the load, the specific location, the applied forces, and the opening or closing of water channels within the rock mass. As a result, some areas may experience rapid drainage, while others may become sealed off within certain zones, making it difficult or even impossible for water to drain out. This leads to the mixing of the three different regions at the front, which in turn presents significant challenges for monitoring and ensuring the safety of the coal pillar during the early stages of mining.

Although the method of clustering analysis based on the k-means algorithm offers several advantages, such as simplicity, ease of handling large datasets, and maturity, the coal pillars studied in this paper, though significantly affected by water hazards, are located in a region with simple geological conditions. The structure of the coal pillars is uniform, and the water immersion process is relatively predictable. Therefore, using the k-means-based clustering optimization algorithm is feasible. However, this method has certain limitations when faced with more complex geological conditions. For instance, when identifying water immersion boundaries in irregular coal pillars or in areas with alternating coal layers, issues such as sensitivity to noise and outliers may arise. Large variations in geological conditions could also lead to difficulties in controlling recognition errors. Additionally, the selection of initial cluster centroids can be challenging. Therefore, when confronted with more complex geological conditions in the future, further optimization or iteration of the algorithm may be required. More advanced algorithms, such as DBSCAN, spectral clustering, or hierarchical clustering, should be considered for further research.