Permeability evolution and modeling analysis of anisotropic coal under different stress paths Authors: Baoyu Wei, Hongbao Zhao, Lu Gao Published: 2025 ABSTRACT AbstractUnderstanding the permeability characteristics of anisotropic coal under various complex stress environments is the key to improving gas extraction efficiency. The true-triaxial permeability tests under four stress paths and three flow directions are carried out using a multifunctional true triaxial geophysical device. The results show that unloading the minimum principal stress while increasing the maximum and intermediate principal stress induces significant fluctuations in coal permeability except along butt cleat direction. The permeability along the face cleat direction (Cf) is 9.9%–182.9% higher than in other directions, indicating greater compressibility of the face cleats. The Biot’s coefficient in the cleat direction is 0.728 ~ 0.841, which is higher than that in the other two flow directions, indicating that natural fractures are more abundant along the face cleats. Under the true triaxial stress path, the Cf value along the end cleat direction is 53.1% to 171.6% higher than that along the face cleat direction, and 9.9% to 182.9% higher than that perpendicular to the bedding direction, indicating that the face cleat has stronger compressibility. Under reduced minimum principal stress, increasing the intermediate principal stress suppresses fracture formation. The permeability model comparison results show that the minimum fitting value 0.9190 of Yu’s permeability model is higher than the minimum fitting value 0.8945 of Chen’s permeability model, and it better fits the fluctuation of permeability evolution, indicating that the permeability model considering cleat characteristics is better in gas drainage prediction. This study provides insights into the seepage characteristics of coal under diverse mining conditions and its permeability anisotropy.Keywords: Permeability evolution, Stress path, Permeability prediction, Fitting analysisSubject terms: Energy science and technology, Engineering, Solid Earth sciences Introduction Coalbed methane (CBM), as unconventional energy resource stored in coal seams, has gained increasing significance in alternative energy production and chemical manufacturing1–4. As shallow resources become progressively depleted, CBM extraction operations have expanded into deeper coal reserves5,6. However, increasingly complex stress environments at depth present significant challenges for methane recovery5,7–9. As mining depth increases, the stress environment of the coal seam becomes more complex10,11. After being disturbed by mining, the stress shows different changing trends, with different risk of coal and gas outburst accidents. Researches show that the exploitation of coalbed methane can alleviate coal and gas outburst disasters12–14. Permeability, a critical parameter governing CBM recovery efficiency, renders the study of coal seam permeability both scientifically valuable and practically essential. Natural cleats networks serve as the primary conduits for gas migration through coal matrices, exerting fundamental control over permeability15,16. The cleat and bedding plane structure of coal is an important factor causing the anisotropy of permeability. The closure and expansion of fractures are different under different stress paths, which in turn affects the permeability of coal seams. Therefore, the study of anisotropic permeability of coal seams under complex stress environments has high research value and practical significance. Coal anisotropy, characterized by bedding planes and orthogonal fracture sets (face cleat and butt cleat), governs its mechanical properties, strength, and seepage characteristics17,18. Scholars such as Li19 and Yu20 et al. conducted seepage tests perpendicular to the bedding plane, along the face cleat and along the end cleat to study the anisotropic characteristics of coal permeability. The cleat and fissure structure of coal constitutes its unique anisotropic seepage characteristics. These microscopic fissures are significantly affected by stress changes. Stress disturbances during coal mining promote the closure or expansion of fissures, thereby controlling the permeability of the coal seam. The increase in coal seam permeability is mainly due to the expansion and penetration of coal body fissures. In particular, the macroscopic fissures formed after coal body destruction will lead to a rapid increase in permeability, which is also strongly affected by the pre-existing cleat and bedding structure12. Identifying high-permeability flow directions is thus critical for optimizing gas extraction efficiency. Mining-induced coal failure primarily results from stress concentrations caused by operational disturbances, roof fracturing, and microseismic activity21–23. Stress variations often exhibit distinct dynamic trends along three principal directions24–26, with spatial evolution patterns dependent on mining locations25,27. To accurately replicate in situ stress conditions, permeability tests must account for these diverse stress paths. Investigations have shown that the stress environment of coal masses usually includes a true triaxial stress state (σ1 > σ2 > σ3), especially near tectonic areas, where horizontal stresses demonstrate significant directional variation28–30. Therefore, it is necessary to carry out the permeability test of coal containing cleats under the true triaxial conditions and different stress paths. Permeability prediction is of vital importance to improve gas extraction efficiency and prevent coal and gas outburst disasters. Establishing a permeability model that conforms to reality is the key to permeability prediction. Yu20 developed a strain-dependent permeability model based on the cubic law that achieved excellent fitting results. Additionally, experimental studies have found that permeability is strongly dependent on effective stress19, and the exponential relationship between permeability and effective stress was modified by Chen31. These scholars have established full-process permeability models from the perspectives of effective stress and strain, and both have been well applied. Integrating such models enables comprehensive investigation of stress path and flow direction effects on coal permeability. In this study, based on different mining conditions, four stress paths are introduced. Based on the bedding and cleat plane of coal, three different gas flow directions are designed. Permeability experiments are carried out under four stress paths and three flow directions. The permeability data obtained by testing are fitted by the permeability model of Chen and Yu, respectively, and the fitting parameters are obtained. The characteristics of permeability evolution under different stress paths and flow directions are analyzed. Thus, the effect of stress paths and flow directions on permeability evolution characteristics are obtained. This study provides support for understanding the seepage characteristics under different stress paths and the permeability anisotropy of coal. Testing methodology Sample preparation In this study, the coal samples are selected from the same coal block at the same depth in the B2 coal seam of the Baijiao Coal Mine in Yibin, Sichuan Province, China. The geographical location of the coal mine is shown in Fig. 1(a). The coal block is processed into a cube sample with a length of 100 ± 0.5 mm. The end faces are polish to obtain three sets of parallel faces. The prepared sample is placed in a drying oven for the drying step. The schematic diagram of prepared sample is shown in Fig. 1(b). It can be found that the sample has obvious bedding planes and cleats. X-ray diffraction results show that the anthracite sample is mainly composed of 5.58% kaolinite, 11.74% quartz, 5.04% calcite, and 77.64% amorphous solids. The coal sample contains 9.76% volatile matter, 63.77% fixed carbon, 27.01% ash, and 1.01% moisture12,32. The uniaxial compressive strength is 25.28 MPa perpendicular to the bedding plane and 20.03 MPa parallel to bedding plane12. (a) The geographical location of the sampling points and (b) the photo of prepared sample (The map was generated using ArcGIS 10.8 (https://www.esri.com). The map data came from the Geospatial Data Cloud, Computer Network Information Center, Chinese Academy of Sciences (http://www.gscloud.cn)). Testing system and procedures In this study, the mechanical and permeability tests subjected to true triaxial stress were carried out using a multifunctional true triaxial geophysical (TTG) instrument. The photograph and schematic diagram of the TTG are shown in Fig. 2. The specific parameters of the device are shown in Table 1. The adsorption test was carried out on the sample for 48 h before the permeability test. The specific steps of the test are as follows: (a) Each adsorbed sample is evenly glued to the four horizontal planes with silicone rubber to prevent gas leakage in the test. (b) Then put the specimen into the loading center and wrap it around with a heat shrinkable tube to isolate the confining pressure hydraulic oil. (c) The confining pressure and three-dimensional loading stress were systematically applied to the sample, and 6 linear variable differential transformers were used to measure the strain of samples. The permeability test is carried out with CO2. The gas pressure is maintained at 1 MPa, and the confining pressure is 3 MPa. Considering that temperature can affect the test results, the permeability test is carried out at 20 ± 2 °C. The photograph and schematic diagram of the TTG (a) photo of the TTG, (b) schematic diagram of the TTG, (c) true triaxial cavity schematic diagram and (d) photos of the inside of the cavity. Table 1The specific parameters of the TTG.Parameters of the TTGValuesX-axis loadMaximum: 6000 kN (accuracy: ±1.0%)Y-axis loadMaximum: 4000 kN (accuracy: ±1.0%)Z-axis loadMaximum: 6000 kN (accuracy: ±1.0%)Seepage pressure: fluid pressureMaximum: 60 MPa (accuracy: ±0.5%)Pressure chamber pressure (oil pressure)Maximum: 60 MPaAxial deformationMaximum: 10 mm (accuracy: ±0.5%)Lateral deformationMaximum: 10 mm (accuracy: ±0.5%)Mass flow meterRange: 5-5000 mL/min (accuracy: ±1.0%) The specific parameters of the TTG. In this study, four stress paths are introduced to carry out permeability tests. The gas adsorption of coal under all stress paths is carried out under hydrostatic stress of σ1 = σ2 = σ3 = 10 MPa. In order to simulate the stress change in the mining process of the Baijiao coal mine, the stress loading is controlled with a rate of 0.025 MPa/s so that the coal sample is in a stress state of σ1 = 20 MPa, σ2 = 15 MPa and σ3 = 10 MPa. The axial stress of 20 MPa is equivalent to the stress level of 800 m depth in Baijiao coal mine, which is within the mining depth range. Coal samples are then loaded with different stress paths (see Table 2), respectively. In the four stress paths, displacement control is used for loading σ1, and the loading rate is 0.002 mm/s (Fig. 3). Table 2Designed stress paths.Stress pathσ1σ2σ3Excavation engineeringStress path 1IncreasedRemainedconstantRemainedconstantComparative test groupStress path 2Decreased(0.02 MPa/s)Decreased(0.02 MPa/s)Room and pillar miningStress path 3RemainedconstantDecreased(0.02 MPa/s)Roadway tunnelingStress path 4Increased(0.06 MPa/s)Decreased(0.02 MPa/s)Long-way mining Designed stress paths. Remained constant Remained constant Decreased (0.02 MPa/s) Decreased (0.02 MPa/s) Remained constant Decreased (0.02 MPa/s) Increased (0.06 MPa/s) Decreased (0.02 MPa/s) The schematic diagram of flow directions. Q is the gas flow rate, cm3/s; v is the gas viscosity, MPa·s; l is the length of the sample, mm; A is the cross-sectional area of the sample, cm2; Pi and Po are the gas pressure levels measured at the inlet and outlet, respectively, MPa. Permeability testing and fitting results and analysis The stress-strain permeability curves under different flow directions and stress paths are shown in the Fig. 4. The characteristic values of permeability for different stress paths and flow directions are shown in the Table 3. The stress-strain permeability curves under different flow directions (a1) ~ (a3) stress path 1, (b1) ~ (b3) stress path 2, (c1) ~ (c3) stress path 3, and (d1) ~ (d3) stress path 4. Table 3The characteristic values of permeability for different stress paths and flow directions.Samples numberInitial permeability(10− 16 m2)Minimum permeability(10− 16 m2)Peak permeability(10− 16 m2)Peak stress(MPa)1-be0.4020.0221.08062.682-be1.0280.8124.85049.223-be3.5070.9333.88056.374-be0.3920.2004.32058.131-fa1.3650.0950.73061.012-fa3.6351.90220.65048.523-fa5.6292.6456.12056.714-fa1.7481.3399.87062.771-bu0.4840.1044.25657.112-bu3.2001.8656.85044.453-bu0.8040.4153.77649.174-bu1.4190.1324.80055.391, 2, 3, and 4 denote the stress path, be denotes the direction perpendicular to bedding plane, fa denotes the direction along the face cleat, bu denotes the direction along the butt cleat. The characteristic values of permeability for different stress paths and flow directions. 1, 2, 3, and 4 denote the stress path, be denotes the direction perpendicular to bedding plane, fa denotes the direction along the face cleat, bu denotes the direction along the butt cleat. As shown in Fig. 4, as the stress increases, the permeability decreases gradually. Then the permeability increases with increasing stress. This non-monotonic trend primarily stems from two competing mechanisms in the gas flow channels under stress: compaction dominance and damage propagation dominance. In the initial stage of gradually increasing stress, the natural fractures close. At this point, the cohesion and shear strength of the coal matrix are sufficient to withstand relatively small shear stresses, and no significant damage has yet occurred. This stage is dominated by fracture compaction, leading to a continuous decrease in permeability. As stress further increases, the compaction effect gradually approaches its limit, while the coal begins to experience internal damage under high stress, forming new flow channels or expanding existing fractures. This causes the damage propagation mechanism to gradually become dominant, and permeability increases. As shown in Fig. 4, the strain corresponding to the minimum permeability is often lower than the strain at the stress peak, indicating that at the point of lowest permeability, fracture compaction and damage propagation reach a dynamic equilibrium. Subsequently, as stress continues to increase, permeability gradually recovers, marking a shift in the evolution mechanism from compaction dominance to damage propagation dominance. As presented in Table 3, under initial stress conditions (σ1 = 20 MPa, σ2 = 15 MPa, σ3 = 10 MPa), the permeability along the face cleat direction is 1.6–4.5 times higher than in the perpendicular-to-bedding-plane direction and 1.1-7.0 times greater than in the butt cleat direction, exhibiting strong anisotropy. This anisotropy can be attributed to the well-developed fracture connectivity along the face cleat direction, which facilitates gas flow, combined with lower stress acting on the original fracture network33,34. Comparing different flow directions, the minimum permeability during the evolution process along the face cleat direction is higher than that in the perpendicular bedding plane direction and the butt cleat direction, resulting from the difference in the coal permeability under the initial stress condition. After the coal failure, the permeability increases significantly compare to the minimum permeability, reaching 2.31 ~ 49.09 times the minimum value. The peak permeability during the evolution process under different stress paths and flow directions vary greatly, resulting from the size of the macro-fracture formed by the coal failure and the residual stress. As illustrated in Fig. 4, the post-peak behavior of stress-strain curves exhibits distinct characteristics depending on the stress path. For stress path 1, the curve demonstrates a relatively gradual stress reduction accompanied by a noticeable stress plateau after peak stress is attained. In contrast, other stress paths show more abrupt stress drops following peak stress. The maintenance of constant minimum principal stress appears to enhance the sample’s load-bearing capacity, as evidenced by the generally higher peak stresses observed for stress path 1 compared to paths 2 and 3 (Table 3). While maintaining reduced minimum principal stress conditions, increasing the intermediate principal stress also contributes to strength enhancement. In summary, the permeability curves under the four stress paths and three flow directions have similar changing trends, which generally show a trend of first decreasing and then increasing, showing an S-type. The minimum principal stress unloading reduces the maximum strain of the coal, and the minimum permeability increases during the loading process. Parameters in Yu’s model are shown in Table 4. Based on the above two different permeability models, the permeability data under different stress paths and flow directions are fitted, respectively. The fitting results of the permeability model are shown in the Figs. 5 and 6, and the detailed fitting parameters are shown in the Table 5. The fitting parameters are the parameters of the permeability model shown in the Appendix, and R2 is the fitting degree. It can be seen that the two permeability models have good fitting effects on the permeability data under different stress paths and flow directions, and the minimum fitting degree of Chen’s permeability model is 0.8945 (see Fig. 5), and minimum fitting degree of Yu’s permeability model is 0.9190 (see Fig. 6). Table 4Parameters used in yu’s model20.ParameterValuePorosity (φ0)6.98%Elastic modulus (E)/GPa2.95Bulk modulus (K)/GPa1.76Bulk modulus of matrix (Km)/GPa12.797Bulk modulus of pores (Kp)/GPa0.421Matrix length (s1, s2, s3)/m0.096, 0.095, 0.092Fracture length (b1, b2, b3)/m0.004, 0.005, 0.008Specimen length (l)/m0.1 Parameters used in yu’s model20. Fitting results of permeability data based on Chen’s permeability model (a) stress path 1, (b) stress path 2, (c) stress path 3 and (d) stress path 4. The be denotes the gas flow direction perpendicular to bedding plane, the bu denotes the gas flow direction along the butt cleat plane, and the fa denotes the gas flow direction along face cleat. Fitting results of permeability data based on Yu’s permeability model (a) stress path 1, (b) stress path 2, (c) stress path 3 and (d) stress path 4. The be denotes the gas flow direction perpendicular to bedding plane, the bu denotes the gas flow direction along the butt cleat plane, and the fa denotes the gas flow direction along face cleat. Table 5The detailed results of fitting parameters.Samples numberChen’s modelYu’ modelCfαβγ R 2 αh R 2 1-be0.08130.28151.7315.4800.97290.7660.99582-be0.05670.38351.4012.3960.96270.7460.98873-be0.01931.98842.4071.6820.94370.7160.98434-be0.04190.07838.6615.6200.89750.6260.91901-fa0.07970.19755.7194.7260.98550.8410.98842-fa0.04070.15957.3812.1770.97640.7760.98903-fa0.02010.66644.1022.3770.97070.7860.94184-fa0.02390.07148.5783.9520.89450.7280.93361-bu0.12650.12944.2159.6840.99840.7880.99742-bu0.06230.29841.2292.8300.98340.7760.97443-bu0.05460.11351.8264.1280.98320.7720.99734-bu0.05830.17752.2055.3070.95940.6500.9753 The detailed results of fitting parameters. Discussion Effects of natural cleats The permeability of the sample under different stress paths and flow directions first decreases and then increases. To quantitatively describe the magnitude of the permeability change, two permeability ratios are introduced12, namely: rd and ri represent the magnitude of permeability reduction and increase, respectively. kmin and kp are the minimum and peak permeability, respectively, 10− 16 m2. The calculation results of the two permeability ratios are shown in Table 6. Table 6Two permeability ratios and brittleness index of coal samples.Samples numberrd (%) r i B1-be94.5048.092.602-be73.403.161.203-be21.014.971.154-be48.9820.603.501-fa93.046.683.242-fa53.011.311.773-fa47.689.861.704-fa23.406.371.851-bu78.5139.925.452-bu48.388.101.743-bu41.722.671.764-bu90.7035.361.66 Two permeability ratios and brittleness index of coal samples. Compared with other stress paths, the permeability reduction of stress path 1 is in the higher range, which is 94.5%, 93.04%, and 78.51%, respectively. Notably, the change in permeability in the butt cleat direction is significantly lower than that in the other two directions. The butt cleat itself has poor connectivity and small opening, and the space that can be compressed is limited, making it relatively insensitive to compressive stress35,36. On stress path 2, the permeability variation in the face cleat direction is 20.39% lower than that in the direction perpendicular to the bedding plane. This difference stems from the different nature of the seepage paths. As the main fracture with good continuity and long extension, the face cleat is significantly affected by the compaction effect, resulting in a sharp drop in permeability37,38. Stress path 3 yields the smallest permeability reduction (21.01%) in the perpendicular-to-bedding-plane direction. The diminished compaction effect on face cleats due to minimum principal stress reduction is most pronounced in this orientation. The increase in maximum principal stress generates tensile or shear stress in the direction perpendicular to the bedding. At the same time, the decrease in minimum principal stress relieves the compaction effect on the opposite cleat. The opening or expansion of the cleat in this direction increases, thereby significantly enhancing connectivity and effectively offsetting the decrease in permeability. On stress path 4, the permeability changes along the butt cleat direction and perpendicular to the bedding plane are low, and the permeability fluctuates. During the loading process, shear stress causes cracks to expand and penetrate rapidly, and the permeability increases, but no macroscopic damage occurs inside the coal. These newly formed cracks are compacted under the action of increased stress, and the permeability decreases, which ultimately leads to fluctuations in permeability. The fitting results of permeability are obtained by using the permeability models of Chen and Yu, as shown in Table 5. Combined with the fitting results, the effect of cleats on coal permeability can be quantitatively analyzed. Chen’s model incorporates four key parameters (Cf, α, β, and γ), with Cf specifically representing cleat compressibility31. Analysis reveals pronounced anisotropy in cleat compressibility, showing distinct directional variations. The highest Cf values occur along the butt cleat direction, followed by values perpendicular to bedding planes, while the lowest values appear along face cleats. Comparative measurements demonstrate that Cf values along the butt cleat direction exceed those along face cleats by 53.1 ~ 171.6% and surpass perpendicular-to-bedding-plane values by 9.9 ~ 182.9%. This differences indicates that face cleats exhibit greater compressibility, whereas permeability performance proves poorest along bedding-parallel directions due to reduced direct compression effects39. For β, Chen believes that it is related to the shear resistance of the coal body. In the Mogi-coulomb criterion40,41, the octahedral shear stress has a wide range of application prospects and applicability in true triaxial, which can be expressed as: Figure 7 shows the relationship between β and octahedral shear stress. For the flow directions perpendicular to the bedding plane and along the butt cleat plane, although there is a certain relationship between shear stress and β, with R2 > 0.665. the coal cleat has a huge impact on the results, especially in the flow direction along face cleat plane. For the flow directions along the butt cleat and perpendicular to the bedding plane, the poorly connected butt cleat gradually expands under stress to form macroscopic cracks, having a promoting and guiding effect on crack propagation42. The relationship between β and octahedral shear stress. The be denotes the gas flow direction perpendicular to bedding plane, the bu denotes the gas flow direction along the butt cleat plane, and the fa denotes the gas flow direction along face cleat. Model comparison reveals Yu’s permeability model, with its minimum R² value of 0.9190, demonstrates superior fitting performance to Chen’s model (minimum R² = 0.8945) for this stress path. The parameter γ, which correlates with post-failure fracture density, shows particularly high values 2.830 ~ 9.684 along butt cleat directions. Although initial permeability remains low in this orientation due to poor cleat continuity, subsequent fracture propagation leads to proportionally greater permeability increases, suggesting that new fractures preferentially form along butt cleats. Yu’s model introduces the anisotropic Biot’s coefficient (αh), which exhibits strong correlation with fracture evolution characteristics20. Larger αh values correspond to greater permeability variations and higher natural fracture densities. Spatial analysis of αh shows maximum values along face cleat directions, intermediate values along butt cleats, and minimum values perpendicular to bedding planes. This distribution confirms that natural fractures concentrate most densely along face cleats and occur least frequently perpendicular to bedding plane orientations. The effect of stress paths To accurately represent stress changes during coal seam mining, permeability tests were conducted under different stress paths. To further investigate the impact of stress conditions on permeability in true triaxial tests, an intermediate stress parameter43 is introduced to evaluate the three-dimensional stress changes during loading. This parameter is calculated using the following formula: The permeability under different intermediate stress parameters b is shown in Fig. 8. As the stress path changes, the intermediate stress parameter gradually increases, with the peak permeability being the highest on stress path 2. At this point, the intermediate stress parameters range from 0.120 to 0.128. Under this stress path, the stress in both horizontal directions is simultaneously unloaded, weakening the inhibitory effect of lateral pressure on coal seam fractures and leading to more significant coal seam damage and a significant increase in permeability. The relationship between peak permeability and intermediate stress parameter under different flow directions. Analysis of Chen’s model parameters reveals consistent trends across flow directions, with Cf values following the order: stress path 1 > 2 > 4 > 3. Notably, Cf values for path 1 exceed other paths by 43.4 ~ 321.2%. This pattern demonstrates that unidirectional stress unloading (characteristic of paths 3 and 4) reduces cleat compressibility. Confining pressure maintenance helps sustain lower permeability levels, while lateral pressure reduction enhances permeability during progressive deformation. The parameter α shows correlation with coal brittleness, particularly in directions perpendicular to bedding planes (excluding stress path 4). In addition, α also has a certain relationship with the direction of crack expansion, which is slightly different from Chen’s point of view31. To describe the brittleness of the specimen, a brittleness index B is introduced44, which can be express as follows: M is post-peak modulus, MPa. The calculated brittleness index is shown in Table 6. The smaller the indicator B, the more brittle the coal. The brittleness index B indicates generally low brittleness for coal under stress path 1, suggesting that confining pressure maintenance reduces brittleness while lower confining pressures enhance it. Figure 9 shows the relationship between the coal brittleness index B and the fitting parameter α. According to the characteristic of B > 1, the distribution of α and B follows the characteristics of logarithmic functions. Therefore, logarithmic functions were used for fitting and good results were obtained. The stronger the brittleness of the sample (the lower the value of B), the larger the value of α. The α has intensive effect to the coal brittleness and weak influence of the anisotropy of the coal structure. In the future, it is necessary to combine CT scanning technology to further study whether the direction of crack expansion has an impact on the parameters. Parameter γ, related to post-failure fracture density, shows significantly lower values (29.2 ~ 77.8% of other paths) for stress paths 2 and 3. Confining pressure maintenance promotes more extensive fracture generation during coal failure, while pressure reduction limits crack formation. Interestingly, unidirectional lateral pressure unloading combined with pressure increase on the opposite side enhances destructive crack formation. The relationship between the coal brittleness index B and the fitting parameter α. In Yu’s model, the anisotropic Biot’s coefficient (αh) follows the trend: stress path 1 > 2 > 3 > 4, indicating that confining pressure maintenance inhibits crack generation and reduces permeability. Minimum principal stress unloading slows permeability reduction, resulting in more fractures at minimum permeability compared to path 1. The lower αh values for path 4 relative to paths 2 and 3 suggest that intermediate principal stress increase suppresses fracture formation. Implications for field application Underground coal mining occurs under high three-dimensional in-situ stresses, which undergo redistribution due to mining-induced disturbances29,45. The stress changes in coal seams vary in different excavation engineering, exhibiting different stress paths. Coal seam stress paths have a significant impact on permeability evolution46,47. Therefore, in actual excavation engineering, effective gas drainage must consider the difference of stress changes. This study investigates the evolution of anisotropic permeability in coal under true triaxial stress conditions through seepage experiments along four distinct stress paths. Figure 10 shows the stress distribution and permeability evolution of coal near the working face during mining. Under stress path 2, the minimum and maximum permeability values during evolution are higher than those under other stress paths, stemming from the unloading of horizontal stress in the coal seam during actual excavation. Under stress path 3, although the minimum permeability value during evolution is relatively high, the maximum permeability value is relatively low, which is unfavorable for gas drainage. Under stress path 4, the maximum permeability falls between the other two stress paths. In this case, drilling pressure relief can be used to unload the internal horizontal stress, thereby enhancing the permeability of the coal seam48. Under unloading stress paths 2, 3, and 4, the coal seam exhibits strong brittleness. Compressive loads can cause severe damage, easily inducing strong outburst disasters49. To mitigate this problem, it is recommended to strengthen confining pressure and stress monitoring. Furthermore, the structure of bedding plane and cleat causes permeability to exhibit anisotropy, with the highest maximum permeability along the face cleat direction, which can be considered a preferred gas extraction direction. The permeability after failure along the butt cleat direction and perpendicular to the bedding plane is relatively similar. However, the bearing capacity of coal along the butt cleat direction is lower, increasing the risk of instability during gas extraction. In addition, comparative analysis of permeability models shows that the Yu’s model is superior to the Chen model, with a minimum R² value of 0.9190, compared to 0.8945 for the Chen model. The reason for the excellent performance of the Yu model is that it introduces anisotropic Biot’s coefficients, which can more accurately describe the permeability evolution of coal under real triaxial stress conditions. It can also better fit the permeability fluctuations observed under stress path 4. Therefore, a permeability model that explicitly considers fracture characteristics is more suitable for predicting gas extraction. As stress increases, coal seams will experience stages such as elasticity, damage and final failure, accompanied by corresponding permeability changes, including the permeability decreasing stage and the permeability increasing stage (pre-peak region and post-peak region). Although existing permeability models can effectively track the overall trend of permeability evolution with stress or strain, they often fail to capture the local fluctuations caused by internal damage during loading. This limitation leads to deviations in prediction accuracy. To solve this problem, external detection methods such as seismic monitoring can be used to identify in-situ fracture conditions50. In future work, we plan to combine these methods with the observation results of local fracture evolution during loading to better characterize permeability fluctuations and develop a more robust model that can reflect this dynamic behavior. Schematic diagram of stress distribution and permeability evolution during the excavation process at the working face. Conclusion In this study, four different stress paths are introduced according to different mining conditions; According to the bedding plane and cleat structure of the coal, three different flow directions are designed. Permeability tests are carried out under different stress paths and flow directions. Chen’s and Yu’s permeability models are used to fit the permeability data to obtain the parameters under different conditions. Combined with the fitting parameters, the characteristics of permeability evolution under different paths and flow directions are analyzed. The main conclusions are as follows: When the minimum principal stress is unloaded, the increase of the maximum principal stress and the intermediate principal stress cause the permeability to fluctuate when the flow direction is perpendicular to the bedding plane and along the cleat, which increases the error of the permeability model prediction. Among them, the permeability model that introduces the anisotropic Biot coefficient is relatively better, and its minimum fit value is 0.9190. When the minimum principal stress is unloaded, the increase of the maximum principal stress and the intermediate principal stress cause the permeability to fluctuate when the flow direction is perpendicular to the bedding plane and along the cleat, which increases the error of the permeability model prediction. Among them, the permeability model that introduces the anisotropic Biot coefficient is relatively better, and its minimum fit value is 0.9190. Quantitative characterization of coal fabric further elucidates the underlying anisotropy. The cleat compressibility Cf along the butt cleat direction exceeds that along the face cleat direction by 53.1%–171.6%, and surpasses the value perpendicular to bedding plane by 9.9%–182.9%, highlighting the dominant compressibility of face cleats. The parameter γ shows higher values 2.830 ~ 9.684 along butt cleats, where initial permeability remains relatively low due to cleat discontinuity, yet demonstrates greater susceptibility to new fracture formation. The anisotropic Biot’s coefficient αh follows the order of face cleat direction > butt cleat direction > perpendicular to bedding planes, confirming that natural fractures predominantly develop along face cleats and are least abundant perpendicular to bedding plane. Quantitative characterization of coal fabric further elucidates the underlying anisotropy. The cleat compressibility Cf along the butt cleat direction exceeds that along the face cleat direction by 53.1%–171.6%, and surpasses the value perpendicular to bedding plane by 9.9%–182.9%, highlighting the dominant compressibility of face cleats. The parameter γ shows higher values 2.830 ~ 9.684 along butt cleats, where initial permeability remains relatively low due to cleat discontinuity, yet demonstrates greater susceptibility to new fracture formation. The anisotropic Biot’s coefficient αh follows the order of face cleat direction > butt cleat direction > perpendicular to bedding planes, confirming that natural fractures predominantly develop along face cleats and are least abundant perpendicular to bedding plane. Unloading the minimum principal stress increases the brittleness of the coal, causing the permeability to change more rapidly during coal failure, increasing the risk of outbursts. The permeability model parameter α has a logarithmic relationship with coal brittleness: the larger α, the greater the coal brittleness. Unloading the minimum principal stress increases the brittleness of the coal, causing the permeability to change more rapidly during coal failure, increasing the risk of outbursts. The permeability model parameter α has a logarithmic relationship with coal brittleness: the larger α, the greater the coal brittleness. Appendix: Chen’s and Yu’s permeability models used for fitting The permeability model31 of Chen can be expressed as: k0 is initial permeability of coal, 10− 16 m2; Cf is cleat compressibility of coal, MPa− 1; σm is mean effective stress, MPa; σm0 is initial mean effective stress, MPa; α, β, and γ are fitting parameters; σd is deviator stress, MPa; σp is peak stress of the coal sample during the testing process, MPa. The mean effective stress and deviator stress31 can be expressed by: σi denotes principal stress (i = 1, 2, and 3), MPa. The 1, 2, and 3 are the direction of maximum, intermediate, and minimum principal stress, respectively. The permeability model of Yu et al.20 can be expressed as: αh is Biot’s efficient; K is bulk modulus, MPa; φ0 is coal porosity; ε1, ε2, and ε3 are strain in the 1, 2, and 3 directions; s1, s2, and s3 are the matrix length of coal in the 1, 2, and 3 directions, respectively, mm; b1, b2, and b3 are the fracture length in the 1, 2, and 3 directions, respectively, mm; Km is matrix modulus, MPa; Kp is the fracture modulus, MPa; l is the length of coal samples, mm; E is elastic modulus, MPa; D1, D2, and D3 are damage variables in the 1, 2, and 3 directions, respectively. Author contributions Writing—Original Draft: B.W., L.G.; Methodology: H.Z.; Writing—Review & Editing: B.W., H.Z. All authors reviewed the manuscript. Funding This study was supported by the Deep Earth Probe and Mineral Resources Exploration-National Science and Technology Major Project (2024ZD1004503). Data availability Data is provided within the manuscript or supplementary information files. Declarations Competing interests The authors declare no competing interests. Conflict of interest The authors declare no conflicts of interest. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement Data is provided within the manuscript or supplementary information files.