Analysis of Surrounding Rock Stability Based on Refined Geological and Mechanical Parameter Modeling—A Case Study

Abstract

Metallic ore deposits are generally formed through magmatic intrusions, followed by metamorphism. The geological structures in such regions are often complex, with mechanical parameters exhibiting significant variability. These characteristics dictate the need for refined geological modeling and heterogeneous mechanical parameters for rock mass stability analysis to ensure reliability. Therefore, this paper proposes a novel method for rock mass stability analysis. The method fully leverages high-density drilling data from the mine and introduces an intelligent rock quality designation (RQD) identification technique, facilitating characterization of the spatial heterogeneity of rock mass RQD. Building on this, laboratory experiment data and in situ measurements are integrated, and the Hoek–Brown criterion is employed to achieve a refined characterization of heterogeneous rock mass mechanical parameters. This method allows for a realistic inversion of in situ rock mass mechanical conditions, overcoming the limitations inherent in assigning uniform parameters. Finally, the computed rock mass mechanical parameters are assigned to the refined computational model to conduct rock mass stability analysis. Taking the Jiangfeng Iron Mine, with its complex geological conditions, as an example, this method enables the accurate evaluation of the rock mass stability, determines the feasibility of joint mining, and calculates the appropriate thickness of the isolation pillars, effectively mitigating safety risks in mining operations. This method provides a valuable reference for the rock mass stability analysis of underground joint mining operations for similar mines.

1. Introduction

Metallic ore deposits are generally formed through magmatic intrusions, followed by metamorphism [1]. The regions where these deposits are located often exhibit complex geological structures with well-developed joints and fractures, as well as common fault fracture zones. The mechanical parameters of the surrounding rock show significant variability. This variability necessitates a thorough consideration of the influence of complex geological structures, such as faults and weak planes, as well as the heterogeneity of rock mass mechanical parameters during rock mass stability analysis [2,3].

Currently, a variety of methods have been developed by both domestic and international scholars to study the stability of the surrounding rock in mining projects. The most traditional and widely used approach is the in situ monitoring method. This technique involves installing various sensors, such as acoustic emission monitors [4], displacement meters [5], and stress meters [6], to collect real-time monitoring data for the dynamic analysis of rock stability. While this method is rooted in actual field conditions and provides reliable data, it requires sophisticated equipment, involves complex data analysis, and makes it difficult to integrate data collected over different time periods. As a result, it often fails to establish consistent and meaningful patterns and may produce contradictory results that are hard to interpret [7]. Another method is the physical modeling approach, which involves constructing scaled or proportional models under laboratory conditions to simulate the mechanical behavior and mechanisms of real engineering structures or geological conditions. This method is particularly suitable for studying the interactions within complex systems, such as changes in rock mass strength under multi-field coupling [8] and the cooperation between the anchor support and the rock mass [9]. It provides scientific evidence for optimizing on-site structural designs, as well as for predicting and improving rock mass stability.

With the rapid advancement of computer hardware and software technologies, numerical simulation methods have been increasingly applied in mining engineering to address the growing complexity of mining safety challenges. These methods now cover key issues such as rock instability, seepage failure, and surface deformation [10,11,12]. In mining projects, numerical simulation has become one of the primary methods for solving complex problems due to its flexibility and versatility. By selecting appropriate constitutive models, numerical simulations can accurately reflect the complex physical properties of rock masses. Additionally, using techniques such as the finite element method (FEM) [13] and the discrete element method (DEM) [14] provides results that are closely aligned with real-world conditions.

Despite the advantages of numerical simulation methods, such as being cost-effective and efficient, they are prone to inaccuracies in mining projects with complex geological conditions. These inaccuracies can lead to distorted simulation outcomes, affecting the reliability of the results. In actual mining projects, faults and heterogeneous rock masses significantly influence the stability of the surrounding rock. To improve accuracy, key geological units, such as ore bodies and fault zones, must be modeled with high precision. Increasing the mesh density and the number of elements, while accounting for complex geological structures such as faults and weak planes, is essential for improving the accuracy of numerical simulations [15].

To make simulation results more representative of actual engineering conditions, it is crucial to fully consider the heterogeneous nature of rock masses. The mechanical properties of rock masses vary significantly from the microscopic to the macroscopic scale, and the presence of fractures, joints, faults, and different lithification processes can lead to substantial variation in the strength parameters of rock in different spatial locations [16]. Therefore, combining heterogeneous rock strength parameters with refined modeling techniques is of great importance for analyzing the stability of surrounding rock in mining operations.

This paper proposes a method for analyzing the surrounding rock stability based on refined geological and mechanical parameter modeling. The workflow of the method is shown in Figure 1. The proposed method is applied to the Jangfeng Iron Mine in Jinghong, Yunnan Province, as a case study, with detailed descriptions provided in Section 2. The reliability of the method is validated, and it offers valuable guidance for ensuring the safety of dual-level joint mining at the Jangfeng Iron Mine. Additionally, the method serves as a reference for the analysis of rock mass stability in other underground joint mining projects.

2. Methodology

2.1. Construction of a Refined 3D Computational Model

The 3D surface of the mine was constructed based on the current topographic survey map of the mine. By extracting contour lines, elevation points, and other surface markers, a digital terrain model (DTM) of the mine surface was generated using the constraint-based line function and DTM functionality. Subsequently, the “curtain” function was used to generate the 3D surface entity of the initial model [17].

The 3D solid models of ore bodies and faults were constructed using drilling data and exploration profile maps from the mine’s exploration report. The boundary lines between different ore bodies were extracted to determine the spatial position and orientation of the ore groups.

Based on the completed 3D solid model, the Griddle plugin was employed to segment the model into pillars, sub-levels, and mining areas according to different simulation scenarios, and a computational grid recognizable by FLAC3D was generated. The detailed steps are as follows:

Step 1: Select the closed non-manifold multi-surface of the mine to be meshed and choose the equidistant division method in the Griddle plugin.

Step 2: After selecting the meshing method, set the grid parameters, including grid density and boundary type. Choose appropriate grid parameters based on the shape of the surface and the requirements.

Step 3: After setting the grid parameters, click the “Generate Grid” button to generate the grid. The time required to generate the grid will vary depending on the complexity of the surface and the density of the grid. Once the grid is generated, its shape and details can be viewed in the Griddle plugin interface.

Step 4: Errors may occur when using the Griddle plugin for meshing. First, review the error message to understand the cause of the error. Errors may be due to invalid input surfaces or multi-surfaces. Check to see if the surface is closed, if there are overlaps, etc.

Step 5: Check to see whether the grid parameter settings are correct, such as whether the grid density is too high or too low, or if the boundary type is correct. Try changing the grid parameter settings, e.g., adjusting the grid density to an appropriate size. If errors persist after checking the grid parameter settings, further adjustments, such as altering the grid density or boundary type, may be needed.

Upon completion of these steps, the refined 3D solid model and mesh model of the mine, including faults, ore bodies, and surrounding rock, were constructed. This refined 3D computational model serves as the basis for the subsequent refined zoning of rock mass quality and numerical simulation. The 3D solid model of the mine can be found in Section 3, while the 3D grid model of the mine and the detailed computational solid model are presented in Section 6.1. The detailed model construction workflow is illustrated in Figure 2.

2.2. Calculation Method of Rock Mass Evaluation Indices

First, laboratory experiments, such as direct shear experiments [18], uniaxial compression experiments [19], and Brazilian splitting experiments [20], are conducted to obtain the physical and mechanical parameters of the rock, including uniaxial compressive strength, cohesion, etc. Then, structural surface parameters, such as joint orientations, joint combinations, joint spacing, joint density, and the number of joints, are acquired using digital photogrammetry techniques. More details on these parameters are provided in Section 4. These data provide the foundation for subsequent rock mass quality classification and the determination of refined mechanical parameters for the rock mass.

Rock quality designation (RQD) is one of the widely used methods for classifying engineering rock masses. It is also used to calculate the Geological Strength Index (GSI). However, traditional manual logging methods for RQD in drilling cores are inefficient and prone to subjective bias. This paper introduces an intelligent RQD recognition method based on the mask region-based convolutional neural network (Mask R-CNN) deep learning model [21], which enhances data processing efficiency and allows for refined rock mass quality characterization in mining projects. Mask R-CNN is an improvement on the Faster R-CNN model [22] and is classified as a two-stage detector. The basic structure of the Mask R-CNN instance segmentation model includes the backbone structure, feature pyramid networks (FPN), region proposal network (RPN), region of interest align (RoI Align), and output components (bounding box for target location, predicted target class, and mask based on the bounding box), as shown in Figure 3.

The backbone structure of Mask R-CNN is a standard convolutional neural network (CNN) designed to extract information from images. In addition to this, Mask R-CNN employs a feature pyramid network (FPN) to improve the network’s performance in extracting features at different scales. For large objects, deep features are used, while shallow features are applied for smaller objects. The region proposal network (RPN) is primarily responsible for generating object detection boxes. Its core operation involves sliding windows across feature maps to generate a series of anchor boxes, according to specific rules. Once these anchor boxes are obtained, the RPN determines whether each anchor box contains an object of interest and then adjusts the anchor box to better align with the actual object. The RPN classifies and regresses the bounding boxes, extracting high-scoring regions of interest (ROIs) and applying non-maximum suppression (NMS) [23]. These ROIs are then input into the ROI align layer for feature extraction.

From the architecture of Mask R-CNN, it can be observed that the ROI layer has two inputs. One is the feature map generated by the backbone structure, and the other is the detection box generated by the RPN. The primary responsibility of the ROI layer is to map the detection boxes to the corresponding regions on the feature map. In the ROI align layer, the selected ROIs from the RPN are used to extract features from the full image feature map generated by the backbone structure. Nonlinear interpolation is employed to obtain values at floating-point positions, producing a fixed-size feature map for each ROI. The mask branch and box regression operate simultaneously in the Mask R-CNN architecture. The mask branch processes each region of interest (ROI) and generates a pixel-level segmentation mask for each candidate region. The segmentation mask for each ROI is predicted pixel-by-pixel, according to the category of the object, producing a binary mask that marks the pixel areas belonging to that object. Meanwhile, box regression optimizes the bounding boxes further through fully connected layers to obtain more precise detection box locations. Mask R-CNN combines the advantages of object detection and instance segmentation, and with its efficient feature extraction and precise region alignment, it surpasses most models in terms of accuracy and scalability.

The automatic RQD logging process based on drilling image recognition involves several steps, including cropping the drilling image, converting it to grayscale, and applying filtering. Threshold segmentation is used to separate the target from the background, followed by edge detection and labeling. Overlapping and adjacent connected regions are merged based on a set threshold, and valid target information is marked. The structural planes and fracture zones within the targets are identified. Depth data for the central line of the structural plane, the width of the structural plane, the start depth of the fracture zone, and the stop depth of the fracture zone are extracted to calculate the RQD value of the core.

A digital camera is used to photograph each core box, ensuring that the lens is positioned perpendicular to the core box during the process. An affine transformation is applied to correct for perspective distortion and any skew in the core images. For each photo, the corresponding drilling number, burial depth, and core placement order are recorded.

The core images are input into the pre-trained Mask R-CNN model, where points are connected to form closed polygons, and the edge contours of the single-row cores are outlined. The RPN generates N predicted bounding boxes within the core image, and the probability of each predicted bounding box belonging to the target class is calculated. These probability values are ranked from highest to lowest, and the bounding box with the highest probability is selected as the reference. The intersection-over-union (IoU) between the remaining bounding boxes and the reference is calculated. Boxes with IoU values above the set threshold are removed. This process is repeated until all objects are detected, resulting in a set of predicted bounding boxes, or anchor boxes. When a core segment with a length ≥ 10 cm is identified, it is manually labeled using LabelMe, as shown in Figure 4a. Mask R-CNN further refines the boundaries of the detected objects and fills each region with color to accurately localize the target, producing a mask, as shown in Figure 4b. Due to variations in rock quality and fracture distribution at different locations, during the core sampling process, rocks in areas with poor quality and higher fracture distribution tend to fracture, whereas rocks in areas with better quality and fewer fractures maintain better integrity. Therefore, the ratio of the cumulative length of the core segments greater than or equal to 10 cm to the advancement per drilling cycle is used as an indicator of rock quality. Each area delineated by the mask represents a core segment that is greater than or equal to 10 cm in length. The curve of validation loss and prediction accuracy versus the iteration steps is shown in Figure 5, with a prediction accuracy of 99.1% and a validation loss of 0.065.

As shown in Figure 6, according to engineering conventions, the number of pixel points n_i along the centerline of each core segment (mask) with a length greater than or equal to 10 cm is calculated for each drilling depth. Simultaneously, the number of pixel points N along the centerline of the entire drilling depth is obtained. The RQD can then be calculated using Equation (1).

$$RQD={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^{\mathrm{m}}(l_i)}}{L}}\times 100\%={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^{\mathrm{m}}(n_i)}}{N}}\times 100\%$$

(1)

where m represents the number of core segments with lengths greater than or equal to 10 cm in the drilling depth, l_i is the corresponding length of each core segment, and L is the drilling depth length, which is related to the degree of rock mass fragmentation and can be obtained from field record logs. It is important to note that L refers to the drilling depth length rather than the total length of the recovered core. The above calculation process is performed for each drilling depth, and the calculated RQD is then correlated with the drilling number, starting depth, and ending depth, thereby achieving RQD logging for all core images. If the core has not been marked with drilling depth intervals or if field logs are unavailable, Equation (2) can be used to calculate and log the RQD for each single-row core or each core image.

$$RQD={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^m(l_i)}}{L_c}}\times \eta \times 100\%={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^m(n_i)}}{N_c}}\times \eta \times 100\%$$

(2)

where L_c and N_c represent the length of a single-row core and the corresponding number of pixel points along the centerline, both of which are constant values. η is the core recovery rate, defined as the ratio of the length of the recovered core to the actual drilling depth. The core recovery rate is recorded for each core image.

However, since the RQD classification method is a single-factor classification approach, the parameters it considers are limited, and it often fails to fully reflect the overall quality of the rock mass. Therefore, a comprehensive classification method using multiple rock mass indices is required. GSI is not only a comprehensive index method for evaluating rock mass quality but also an essential parameter in the Hoek–Brown criterion [24]. GSI is primarily determined based on the structure of the rock mass and the characteristics of its discontinuities. Information about the rock mass and joints can be obtained through digital photogrammetry of the structural surfaces. Further details of this process are given in Section 4. The structure of the rock mass and the characteristics of the discontinuities are represented by the structure rating (SR) and the surface condition rating (SCR), respectively [25]. SR is related to the joint volume (J_v), while SCR is influenced by factors such as joint roughness (R_r), weathering (R_w), and filling (R_f). The values of these parameters can be found in the relevant literature [26]. The values for SR and SCR can be calculated using the following equations:

$$SR=-17.5\ln J_v+79.8$$

(3)

$$SCR=R_r+R_w+R_f$$

(4)

It is sometimes difficult to obtain data on the quality of the rock mass and joint information from within the mine. In such cases, the rock mass structure is represented by RQD. The characteristics of the discontinuities are quantified using the joint roughness number (J_r) and the joint alteration number (J_a) [27]; these are determined as follows:

$$GSI={\displaystyle \frac{52J_r}{J_a}}/(1+{\displaystyle \frac{J_r}{J_a}})+{\displaystyle \frac{RQD}{2}}$$

(5)

As a result, we obtained several GSI values from different locations within the study area. Using these data, we performed rock mass quality classification for the mine and provided data support for calculating rock mass strength parameters. Further details of this process are given in Section 2.3.

2.3. Calculation of Refined Rock Mass Strength Parameters

To better assess the stability of the surrounding rock from the perspective of rock mass mechanical properties and to meet the requirements of refined numerical simulations, it is necessary to calculate the refined rock mass strength parameters. First, drilling sampling is conducted at the mine, with rock samples taken from different lithologies and locations. Then, laboratory testing is used to obtain the rock’s mechanical properties, followed by using the method described in Section 2.2 to derive the GSI values, thus completing the construction of the basic database. In this paper, the mechanical parameters of rocks with the same lithology are averaged for demonstration purposes. The laboratory parameters of the rock are listed in

Table 1. Laboratory experiment parameters of rock samples.

Rock Lithology	Density (kg/m3)	Uniaxial Compressive Strength (MPa)	Cohesion (MPa)	Friction Angle (°)	Elastic Modulus (GPa)	Poisson’s Ratio	GSI	mi
Non-ore skarn (surrounding rock)	2890	86.08	6.83	48.16	15.1	0.26	55	9
Ore-bearing skarn (ore body)	4020	115.89	12.02	53.13	21.1	0.27	62	19
Diabase	2690	85.2	10.95	44.1	32.5	0.29	56	19
Schist	2820	58.3	4.69	43.78	23.8	0.28	50	6
Granulite	2720	91.3	7.83	45.28	25.5	0.26	52	33
Syenite	2700	63.86	2.35	44.60	18.8	0.26	55	28
Amphibolite	2880	122.9	3.40	47.60	3.59	0.2	64	28
Granite	2580	74	8.1	43.40	3.78	0.26	60	33

. Cohesion refers to the mutual attractive force between adjacent parts within the same substance. From another perspective, it represents the shear strength of a material in the absence of any normal stress on the failure surface. The friction angle reflects the magnitude of frictional interaction between particles and is an important parameter for shear strength. The Poisson’s ratio represents the ratio of transverse strain to axial strain in a material under deformation due to applied stress [28].

In 1980, E. Hoek and E.T. Brown [29,30] proposed the original narrow-scope Hoek–Brown strength criterion based on an analysis of Griffith’s theory and the modified Griffith theory. This criterion primarily consolidated the factors affecting rock mass properties into two parameters, m_b and s. Since then, the Hoek–Brown criterion has been continuously expanded and refined based on practical experience from its application, thus overcoming its initial limitations. In 2002, Hoek et al. [31] introduced the generalized Hoek–Brown failure criterion for estimating rock mass deformation and strength. This generalized criterion consolidates the complex factors influencing rock mass properties into three parameters, m_b, s, and a, facilitating its application in engineering. The formula is as follows:

$${\sigma }_1={\sigma }_3+{\sigma }_c{(m_b{\displaystyle \frac{{\sigma }_3}{{\sigma }_c}}+s)}^a$$

(6)

In the equation, σ₁ represents the maximum principal stress at the time of rock mass failure, σ₃ is the minimum principal stress at the time of rock mass failure, and σ_c denotes the uniaxial compressive strength of the intact rock blocks. The other factors, m_b, s, and a, are material constants of the rock mass. Hoek et al. introduced a disturbance parameter D, which accounts for the effects of blasting and stress relief. The value of D ranges from 0.0 for undisturbed rock masses to 1.0 for highly disturbed rock masses. The introduction of the disturbance parameter D provided a new method for determining the values of m_b, s, and a. This revision not only refined the narrow-scope Hoek–Brown criterion but also expanded the applicability of the empirical strength criterion, making it more comprehensive and specific. The corresponding expression is as follows:

$$\left\{\begin{array}{l}{m}_b=m_i\exp \left({\displaystyle \frac{GSI-100}{28-14D}}\right)\\ s=\exp \left({\displaystyle \frac{GSI-100}{9-3D}}\right)\\ a={\displaystyle \frac{1}{2}}+{\displaystyle \frac{1}{6}}\left(e^{-{\displaystyle \frac{GSI}{15}}}-e^{-{\displaystyle \frac{20}{3}}}\right)\end{array}\right.$$

(7)

In the equation, m_i is the Hoek–Brown constant for the intact rock blocks. By substituting the GSI, m_i, and disturbance factor D values, the parameters m_b, s, and a can be calculated. With these values, the rock mass strength parameters can be determined using rock mechanics principles. The friction angle and cohesion of the rock mass can be determined using Equations (8) and (9), respectively. The specific results of the rock mass strength parameter calculations are shown in

Table 2. Calculation results of rock mechanical parameters.

Rock Mass Lithology	Tensile Strength (MPa)	Elastic Modulus (GPa)	Friction Angle (°)	Cohesion (MPa)	Poisson’s Ratio	Rock Mass Quality Classification
Non-ore skarn (surrounding rock)	0.88	12.74	32.23	0.79	0.26	III
Ore-bearing skarn (ore body)	1.36	19.21	39.55	1.44	0.27	III
Diabase	0.89	13.04	39.11	0.94	0.29	III
Schist	0.65	7.64	23.55	0.49	0.28	IV
Granulite	0.92	10.72	42.25	1.04	0.26	III
Syenite	0.87	10.66	39.59	0.93	0.26	III
Amphibolite	1.16	24.82	48.5	1.53	0.2	III
Granite	0.93	15.30	44.94	1.12	0.26	II
Fault breccia zone	0.31	2.95	22.75	0.23	0.26	/

.

$$\varphi ={\sin }^{-1}\left[{\displaystyle \frac{6a{m}_b{\left(s+m_b{\sigma }_{3\mathrm{n}}\right)}^{a-1}}{2\left(1+a\right)\left(2+a\right)+6a{m}_b{\left(s+m_b{\sigma }_{3\mathrm{n}}\right)}^{a-1}}}\right]$$

(8)

$$c={\displaystyle \frac{{\sigma }_c\left[\left(1+2a\right)s+\left(1-a\right)m_b{\sigma }_{3\mathrm{n}}\right]{\left(s+m_b{\sigma }_{3\mathrm{n}}\right)}^{a-1}}{\left(1+a\right)\left(2+a\right)\sqrt{1+\frac{6a{m}_b{\left(s+m_b{\sigma }_{3\mathrm{n}}\right)}^{a-1}}{\left(1+a\right)\left(2+a\right)}}}}$$

(9)

$$\left\{\begin{array}{l}{\mathsf{\sigma }}_{3\mathrm{n}}={\displaystyle \frac{{\mathsf{\sigma }}_{3\max }}{{\mathsf{\sigma }}_{\mathrm{c}}}}\\ {\mathsf{\sigma }}_{3\max }=0.72{\displaystyle \frac{{\mathsf{\sigma }}_{\mathrm{cm}}}{\mathsf{\rho }\mathrm{g}\mathrm{H}}}-0.91\end{array}\right.$$

(10)

where σ_3max is the upper limit of the stress constrained by the relationship between the Hoek–Brown criterion and the Mohr–Coulomb criterion, $\mathsf{\rho }$ represents the density of the rock mass, g is the gravitational acceleration, H is the burial depth, and σ_cm denotes the overall strength of the rock mass. The specific results of the rock mass strength parameter calculations are shown in Table 2. The rock mass quality grades in Table 2 are determined using the RMR method [32].

In this study, to meet the requirements for the refined calculations, the refined mechanical parameters must also be computed. Using the aforementioned basic database and the demonstrated calculation method, rock mass strength calculations are performed on samples from different spatial locations and lithologies. This results in the establishment of a rock mass strength database. A drilling database is then constructed in 3Dmine, and the rock mass strength parameters are applied to the model using the inverse distance weighted (IDW) method [33]. The IDW method is a deterministic mathematical approach used to estimate values at unsampled locations. This method employs a linearly weighted combination of known sample points to estimate the values at unsampled locations, where the weights are determined by the inverse distance between the sample points and the estimated point. It assumes that the similarity of values for a spatially distributed variable z decreases as the distance d and the power p of the distance increase. The equation for IDW interpolation is as follows:

$$\widehat{z}(x_0)={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^n\frac{z(x_i)}{d_i^p}}}{{\displaystyle {\sum }_{i=1}^n\frac{1}{d_i^p}}}}$$

(11)

where $\widehat{z}(x_0)$ is the estimated value at unsampled location x₀, x_i is the known locations, i = 1 to n, d is the distance, and p is the power parameter. The default power value is 2.

Through this process, the refined mechanical parameters of the model are obtained. As an example, in Figure 7, the refined cohesion distribution contour map for the 115 m sublevel cross-section of the model is presented and analyzed.

In the figure, the slanted lines represent the distribution of exploration lines, and the pink bands indicate the ore veins. The maximum cohesion occurs in the rock mass on the northeastern side, and the rock mass on the southeastern side of the ore vein generally shows better cohesion than does the northwestern side. Therefore, special attention should be paid to the southeastern side of the ore vein, and timely support measures should be implemented. It is evident that refined rock mass strength parameters provide valuable insights for analyzing the stability of the surrounding rock in mining operations.

3. Engineering Background

The Jangfeng Iron Mine is located in the southwestern border of Yunnan Province, under the jurisdiction of Menglong Town, Jinghong City, in the Xishuangbanna Dai Autonomous Prefecture. The mine is situated 195° from Jinghong City, with a linear distance of 59 km and a road distance of approximately 75 km. It lies 9 km southwest at 211° from Menglong Town. The central geographical coordinates of the mine are 100°38′38″ E and 21°30′06″ N. The mine area extends in a northeast-southwest direction, stretching from Manfei in the northeast to 1.5 km northeast of Guofang in the southwest. The total length of the mine is 3 km, with a width of 0.6 km, covering an area of 2.0314 km², and it is conveniently accessible.

The Jangfeng Iron Mine is divided into three phases: Phase I includes areas above the 340 m elevation, Phase II includes areas between the 100 m and 340 m elevations, and Phase III includes areas below the 100 m elevation. Phase II is further subdivided into the 100 m and 220 m levels. The mine is segmented into three ore sections, containing a total of 13 ore bodies. In sections I, II, and III, the ore bodies dip at angles ranging from 105° to 160°, with local reverse dips. The dip angles range from 75° to 89°, with an average dip angle of 83°. The longest strike length of a single ore body is 1914 m, while the shortest is 44 m. The maximum dip extension reaches 968 m, with a minimum of 18 m. The ore-bearing elevations range from 739 m to 390 m, with a total vertical height of 1129 m. The ore bodies have an average thickness of 1.75 to 28.76 m and consist of uniformly composed steeply dipping thin-to-thick ore bodies. Geological data for the Jangfeng Iron Mine are shown in Figure 8. The basis for the hierarchical selection in this paper comes from the mine production design data, as the mine is currently planning to mine the 100 m and 220 m levels. The topographic map and mining phase diagram of the Jiangfeng Iron Mine are shown in Figure 9.

Due to the need for increased production capacity, joint mining of the 100 m and 220 m levels can effectively meet production expansion requirements. However, its safety must be specifically studied. Simultaneous mining of these two levels will inevitably intensify the disturbance to the surrounding rock, potentially affecting its stability. Although the surrounding rock of the Jangfeng Iron Mine consists of relatively hard rock formations, the presence of complex faults has resulted in highly fractured rock masses. Additionally, the development of alteration zones, weathered contact zones, and numerous joints and fractures further weakens the stability of the surrounding rock. This makes the tunnels prone to geological engineering issues, such as roof falls [34], sidewall spalling [35], and floor heaving [36], which can negatively impact the stability of the ore bodies. The mining area has developed more than 14 longitudinal and transverse faults, interwoven with each other. Among them, longitudinal faults dominate, with the largest ones being the F4 and F10 faults. The main ore bodies of the mining area are hosted between these two faults. Both faults extend discontinuously for over 3000 m and are roughly parallel to the fold axis. On the surface, the two faults are spaced 55–160 m apart, and their structural planes are classified as Grade II structural planes. The other faults are smaller in scale, generally extending 300–500 m, and are considered secondary fractures within the mining area, classified as Grade III structural planes. However, faults such as F1 and F3, which formed later, have significant transverse destructive effects on folds and ore-forming structural zones. Along fault fracture zones and the contact fracture zones of intrusive rock bodies, rock structural fractures are well-developed, alteration is intense, and the rock mass exhibits uneven integrity and strength. These conditions directly impact the stability of the surrounding rock of the ore bodies. Therefore, to prevent potential mining hazards that joint mining operations might trigger, a stability analysis of the surrounding rock is essential.

4. Acquisition of Structural Surface Parameters

The weak structural planes of engineered rock masses have a significant impact on rock mass stability. Their shape, mechanical properties, and spatial configurations largely control ground pressure activity and rock mass collapse processes in mines. Investigating the properties and characteristics of rock mass structural planes through field surveys and providing both qualitative and quantitative descriptions is of critical importance in the classification of rock mass stability [37].

The ShapeMetriX 3D system [38], developed in Austria, along with a self-developed recognition program, were applied to conduct a structural plane investigation and analysis in the Phase II mining area of the Jangfeng Iron Mine. The working principle is illustrated in Figure 10, and the structural plane identification results are shown in Figure 11. The system’s accuracy is at the centimeter level.

In this structural surface parameter survey, 14 measurement points were selected in the 115, 135, and 220 sublevels of Phase II. Due to underground environmental issues, 10 effective measurement points were synthesized. Six measurement points were synthesized in the 135 sublevel, and four measurement points were synthesized in the 220 sublevel. The distribution of the measurement points is shown in Figure 12.

Due to space limitations, the structural surface orientation information for each measurement point will not be listed. The data from the 10 valid measurement points were summarized, and the stereographic projection diagrams of all structural planes for sublevels 135 and 220 were obtained, as shown in Figure 13. From Figure 13, it can be seen that there are three dominant structural surface sets in the 135 sublevel: the first set has a dip of 57° and a dip direction of 213°; the second set has a dip of 78° and a dip direction of 120°; the third set has a dip of 71° and a dip direction of 57°. In the 220 sublevel, there are two dominant structural surface sets: the first set has a dip of 59° and a dip direction of 50°; the second set has a dip of 75° and a dip direction of 228°. The average trace length of the joints and fissures in the surrounding rock is 0.66 m for the 135 sublevel and 0.48 m for the 220 sublevel, indicating dense spacing. Additionally, the on-site structural surface investigation revealed that the structural surfaces in Phase II of the Jiangfeng Iron Mine are relatively rough, unopen, and highly weathered, with some structural surfaces showing obvious cementation when exposed to water.

5. Rock Mass Quality Evaluation

5.1. Refined RQD Evaluation Zoning

The specific introduction of RQD has already been mentioned in Section 2.2, so it will not be repeated here. A borehole database is established in 3Dmine, which includes parameters such as hole coordinates, maximum hole depth, trajectory type, azimuth, dip angle, RQD, GSI, and rock mass strength. Then, the RQD parameters are used to assign values to the model using the inverse power law method, based on distance. A horizontal section analysis is performed on the model, and the plane RQD distribution contour maps for the 115, 135, and 220 sublevels are drawn, as shown in Figure 14.

As illustrated in Figure 14, the distribution of RQD values between the 115 m and 220 m sublevels shows minimal variation. The regions enclosed by red boundaries are classified as Grade IV, indicating poor rock mass quality. The areas outlined in pink represent Grade III rock masses, which are of average quality. The zones surrounded by black boundaries are rated as Grade II, reflecting good rock mass quality. The rock mass quality is generally poorer on the northeastern and southwestern sides, with a Grade IV classification. As one moves closer to the central area, the rock mass quality improves. The dashed lines in the figure represent the exploration lines. The distribution intervals of rock mass quality assessed by RQD for each section, divided by the comprehensive exploration lines, are shown in

Table 3. Rock mass quality distribution intervals for each sublevel, as assessed by RQD.

Sublevel	Grade II	Grade III	Grade IV
115 Sublevel	Exploration lines 19~22 Exploration lines 25~26	Exploration lines 26~27	Exploration lines 13~19 Exploration lines 22~25 Exploration lines 27~31
135 Sublevel	Exploration lines 18~22 Exploration lines 27~31	Exploration lines 22~27	Exploration lines 13~18 Exploration lines 31~36
220 Sublevel	Exploration lines 16~26 Exploration lines 29~32	Exploration lines 26~29	Exploration lines 13~16 Exploration lines 32~36

:

5.2. Refined GSI Evaluation Zoning

The detailed introduction of GSI has already been provided in Section 2.2 and will not be repeated here. Using the same method described in Section 5.1, GSI distribution contour maps were generated for the 115, 135, and 220 sublevels, as shown in Figure 15.

As illustrated in Figure 15, the rock mass quality on the northeastern and southwestern sides between the 115 m and 220 m levels is suboptimal, with quality grades evaluated as III and IV, respectively. In contrast, the central portion of the rock mass within this range is predominantly classified as Grade II, reflecting comparatively better rock mass quality. The distribution intervals of rock mass quality for each segment, as assessed by GSI and delineated by the exploration lines, are presented in

Table 4. Rock mass quality distribution intervals for each sublevel, as assessed by GSI.

Sublevel	Grade II	Grade III	Grade IV
115 Sublevel	Exploration lines 19~21 Exploration lines 24~31	Exploration lines 13~19 Exploration lines 21~24	Exploration lines 31~36
135 Sublevel	Exploration lines 19~23 Exploration lines 24~31	Exploration lines 13~19 Exploration lines 23~24	Exploration lines 31~36
220 Sublevel	Exploration lines 15~20 Exploration lines 24~27 Exploration lines 28~31	Exploration lines 13~15 Exploration lines 20~24 Exploration lines 27~28	Exploration lines 31~36

.

6. Stability Analysis of Surrounding Rock Based on Refined Modeling

To analyze the impact of the isolation pillar thickness between the 100 m and 220 m sublevels on the ore body, faults, surface, and surrounding rock stability in Phase II of the Jiangfeng Iron Mine, three working conditions were studied, as shown in

Table 5. Summary of working conditions for different isolation pillar thicknesses.

Plan	Operating Condition	Mining Sequence	Plan
1	No isolation pillar	Unidirectional mining	1
2	Isolation pillar thickness of 10 m	Unidirectional mining	2
3	Isolation pillar thickness of 20 m	Unidirectional mining	3

. The mining sequence of each sublevel and the extraction and filling sequence within the sublevels follow the requirements of the mine design specifications.

6.1. Refined Computational Model

Based on the 3D refined numerical model established in Section 2.1, the rock mass strength parameters calculated in Section 2.3 are assigned to the model. During the calculation, normal constraints are applied to the model’s sides, and full constraints are applied to the model’s bottom. The Mohr–Coulomb constitutive model is used. In this simulation, the goaf is set as empty cells. Empty cells refer to cells in the numerical model that exist but do not contain material and are used to represent underground voids, fractures, or other empty areas. These empty cells do not contribute any physical or mechanical properties (such as density, stiffness, etc.) during the simulation process, but their geometric shape is retained in the model to maintain the continuity of the calculations. Taking the Jiangfeng Iron Mine as an example, the size of the mine model is 1890 m × 1826 m × 1205 m (length × width × height, with the z-direction varying according to the topographical undulations). The model consists of 1,086,000 computational units and 374,000 nodes, as shown in Figure 16.

During the numerical simulation calculation of the Jiangfeng Iron Mine, the stope recovery is carried out layer by layer and segment by segment, according to the investigation report. The calculation proceeds until the unbalanced force of the 3D numerical model system reaches 1 × 10⁻⁵, which is considered balanced. Then, filling is performed, and the calculation proceeds again until the unbalanced force reaches 1 × 10⁻⁵. This process is repeated until the 3D numerical simulation calculation is completed. To further analyze in detail the stability of the surrounding rock during combined mining, two typical profiles, KH1 and KZ1, were selected along the strike of the ore body for stress evolution analysis. The locations of these profiles are shown in Figure 17a.

6.2. Analysis of Plastic Zone Evolution

The analysis of the plastic zone evolution of the ore body is shown in Figure 18a,c,e. Under the combined mining conditions of the 100 m and 220 m sublevels, the fill bodies, stope pillars, and inter-stope pillars in the areas near the ore body faults F10, F2-1, and F18 undergo plastic deformation. However, with the increase in the thickness of the isolation pillars, the extent of this plastic deformation is significantly reduced. When the isolation pillar thickness is set to 20 m, it effectively suppresses the initiation and development of the plastic zone within the stopes and fill bodies of the 100 m and 220 m sublevels. Nevertheless, certain plastic zones still exist, and special attention should be paid to the stability of the ore–rock contact zones and mining operations near faults. High-intensity pre-support may be necessary, when required.

The analysis of the plastic zone evolution of the faults is shown in Figure 18b,d,f. Under the combined mining conditions of the two sublevels, significant plastic deformation occurs at the junction of the ore body fault F2-1 and the near-ore fault F10 only when the isolation pillar thickness is 0 m. As the thickness of the isolation pillar increases, the tendency for the plastic zone to develop towards this junction weakens. However, the area of the plastic zone at the junction of the ore body fault F2-1 does not significantly improve with increased isolation pillar thickness. Nonetheless, it helps to some extent in controlling the stability of the rock mass near the faults, particularly when the isolation pillar thickness is set to 20 m, further improving the plastic deformation area of the faults.

6.3. Analysis of Deformation Evolution

The analysis of ore body deformation evolution is shown in Figure 19a,c,e. Under the combined mining conditions of the two sublevels, failure to set isolation pillars will result in significant deformation in the fill bodies and ore body at the junctions of the ore body with faults F2-1 and F18, particularly in the red dashed area between the 180 m and 260 m sublevels, which is detrimental to the safe conduct of mining activities. However, as the thickness of the isolation pillars increases, the displacement and deformation in the ore body and fill bodies in the mentioned areas gradually improve. Additionally, the local large deformation areas at the junctions of faults F2-1 and F18 with the ore body also improve. When the isolation pillar thickness is 20 m, the stability of the surrounding rock is relatively good. However, even with a 20 m isolation pillar, mining through faults still requires careful attention.

The analysis of fault plastic zone evolution is shown in Figure 19b,d,f. Under the combined mining conditions of the two sublevels, without isolation pillars, mining activity impacts faults F2-1, F18, and the near-ore fault F10. This results in large deformation areas at the junctions of fault F2-1 with the ore body and adjacent locations and small deformation areas at the junctions of faults F18 and F10 with the ore body and adjacent locations. When the isolation pillar thickness is 10 m, it can somewhat improve the displacement deformation of critical faults in these areas, but the small deformation areas remain large. With a 20 m isolation pillar, the displacement deformation of critical faults in these areas is further improved, significantly reducing the small deformation areas. This more effectively controls the displacement deformation of the near-ore regions of the three critical faults, improving the displacement deformation at the junctions of the faults with the ore body.

6.4. Analysis of Stress Evolution

6.4.1. Stress Analysis of KZ1 Profile

As shown in Figure 20, under the combined mining conditions of the two sublevels, an unloading zone is formed, and the compressive stress in the fill bodies is relatively low, indicating that the stability of the fill bodies in the first and second stopes of the 220 m Sublevel is relatively good. The pillars, inter-pillar areas, and isolation pillars are primarily compression zones, with the isolation pillars bearing greater compressive stress. When no isolation pillars are set, the 340 m Sublevel isolation pillar becomes the main compression zone, dividing the unloading zones above and below 340 m. When the isolation pillars are set, and as their thickness increases, the average compressive stress of the isolation pillars between Phase I and Phase II gradually decreases, and stability gradually improves. However, stress concentration zones are likely to form at the junctions of the isolation pillars and the ore body faults, causing plastic deformation and reducing the stability of the isolation pillars. It is necessary to strengthen stress and deformation monitoring around stopes, pillars, and faults.

The minimum and maximum principal stress ranges for the KZ1 profile were summarized, and it was found that with the increase in the thickness of the isolation pillars, the overall range of the minimum and maximum principal stresses decreases, and the maximum values of compressive and tensile stresses also decrease. Additionally, from Figure 20, it can be observed that the region of tensile stress distribution is reduced, and the area of high-stress distribution also shrinks. From the perspective of stress evolution, the stability of the mine rock mass has significantly improved. The summarized results are shown in

Table 6. Range of maximum and minimum principal stresses for different isolation pillar thicknesses in profile KZ1.

Isolation Pillar Thickness (m)	Range of Minimum Principal Stress (Pa)	Range of Maximum Principal Stress (Pa)
0	−6.6922 × 107~1.4813 × 104	−2.3786 × 107~1.0002 × 106
10	−6.0441 × 107~9.3152 × 104	−2.1936 × 107~9.2852 × 105
20	−4.5129 × 107~2.6060 × 104	−2.1475 × 107~8.4438 × 105

.

6.4.2. Stress Analysis of KH1 Profile

As shown in Figure 21, under the combined mining conditions of the two sublevels, an unloading zone is formed, and the maximum tensile stresses in the inter-pillars are 0.87 MPa, 0.86 MPa, and 0.81 MPa, respectively, all below 1 MPa. As the thickness of the isolation pillar in the 220 m Sublevel increases, the maximum tensile stress in the inter-pillars gradually decreases, indicating better stability. This shows that increasing the thickness of the isolation pillars helps improve the stability of the inter-pillars. The pillars, inter-pillars, and isolation pillars are primarily compression zones, with the isolation pillars bearing greater compressive stress. When no isolation pillars are set in Phase II, localized areas of excessive compressive stress occur in the 340 m Sublevel isolation pillar, and excessive compressive stress areas appear at the junction of the bottom of the 100 m Sublevel and the ore body faults. This indicates that the stability in these areas is poor, and plastic deformation is likely to occur. When the isolation pillars are set, and as their thickness increases, the average compressive stress of the isolation pillars between Phase I and Phase II gradually decreases, and stability gradually improves. However, stress concentration zones are still likely to form at the junctions of the isolation pillars and ore body faults, reducing the stability of the isolation pillars and affecting the stability of the surrounding rock near the stopes, pillars, and faults.

The minimum and maximum principal stress ranges for the KH1 profile were summarized, and the variation pattern is the same as that of the KZ1 profile. The summarized results are shown in

Table 7. Range of maximum and minimum principal stresses for different isolation pillar thicknesses in profile KH1.

Isolation Pillar Thickness (m)	Range of Minimum Principal Stress (Pa)	Range of Maximum Principal Stress (Pa)
0	−1.5899 × 105~−5.3325 × 107	−2.0863 × 107~8.7776 × 105
10	−1.4940 × 105~−5.1028 × 107	−2.1413 × 107~8.6008 × 105
20	−1.4426 × 105~−4.8914 × 107	−2.1055 × 107~8.0903 × 105

.

7. Conclusions

In this study, a method for surrounding rock stability analysis based on refined geological and mechanical parameter modeling was proposed. This method was successfully applied to the Jiangfeng Iron Mine, a case study involving multiple faults and joint mining in two levels. The stability of the mine rock mass was analyzed from three perspectives: plastic zone distribution, deformation evolution, and stress evolution. The method utilizes RQD intelligent identification technology, significantly improving the speed and accuracy of core data processing, which facilitates the calculation of rock mass parameters. Moreover, the analytical results of this method provide guidance for optimizing mining structures and offer reference value for ensuring mine safety. The main conclusions of this paper are as follows:

(1) This method establishes a 3D refined calculation model, integrating rock mechanics parameters, borehole data, and structural plane orientation information. It provides a refined evaluation and zoning of rock mass quality using RQD and GSI. By employing the modified Hoek–Brown strength criterion to calculate the strength parameters of heterogeneous rock masses, it achieves refined characterization of the mechanical parameters of the surrounding rock, forming a data foundation for stability analysis.

(2) Using the proposed method, rock quality grade zoning was performed for different levels of the Jiangfeng Iron Mine. For levels 115, 135, and 220, the rock mass quality grade between exploration lines 31 and 36 was evaluated as Grade IV, indicating that the mining structure parameters in these areas should be appropriately reduced.

(3) Simulation results indicate that as the thickness of the isolation pillars increases, the stability of the Phase I and Phase II isolation pillars gradually improves. When a 20 m isolation pillar thickness is maintained, overall stability is optimized. However, stress concentration zones are likely to form at the junctions of the isolation pillars and fault zones, leading to potential rock instability. Mining structure adjustments in these areas should be carried out.