Drill-and-blast tunneling remains one of the most prevalent excavation methods, yet it inevitably produces vast quantities of rock waste. In long tunnel projects, the volume of this waste can easily exceed millions of cubic meters over a construction cycle spanning one to two years. The conventional treatment of this material, often involving simple stockpiling and landfilling, not only represents a significant waste of potential resources but also carries substantial risks of geological hazards and environmental pollution. The inherent value of tunnel muck lies in its potential for resource utilization, provided its fragmentation characteristics are thoroughly understood. The geometric properties of rock blocks, particularly their diameter and morphology, are the decisive parameters dictating their suitability for various applications, from direct use as roadbed fill to processing into aggregates for concrete. Traditional manual measurement techniques, such as sieving and direct caliper measurement, are notoriously inefficient, offer limited spatial coverage, and are heavily influenced by subjective judgment, making them inadequate for large-scale, high-frequency assessments.
The advent of high-resolution drone technology has fundamentally transformed the landscape of geotechnical surveying and analysis. Modern unmanned aerial vehicles, integrated with Real-Time Kinematic (RTK) positioning modules, offer a compelling combination of affordability, maneuverability, and centimeter-level spatial accuracy. This study leverages drone technology to bridge the gap between the massive scale of tunnel waste dumps and the need for precise, granular characterization of the muck pile. By deploying a DJI Phantom 4 RTK platform, equipped with a 20-megapixel camera, we systematically captured the waste site of a major tunnel project located in a mountainous region under challenging geological conditions, predominantly composed of sandstone, mudstone, and conglomerate. The drone technology enabled rapid, non-contact data acquisition over a large area, laying the foundation for a fully automated analysis pipeline.
The core of our methodology hinges on the fusion of drone-based photogrammetry and deep learning. The image acquisition parameters were carefully calibrated to meet the precision requirements of fine aggregate analysis. The flight altitude was set between 20 and 50 meters. The ground sampling distance (GSD), a critical metric representing the physical size of a single pixel on the ground, can be calculated using the following formula:
$$ G = \frac{H \cdot p}{f} $$
In this equation, $H$ represents the vertical distance from the camera to the ground, $p$ is the physical pixel size of the sensor, and $f$ is the focal length of the camera lens. For our setup, this calculation yielded a GSD of approximately 0.9 mm. Given that planar measurement accuracy is typically estimated as 2 to 3 times the GSD, our system achieves a theoretical accuracy of roughly 2.7 mm, which is fully sufficient for measuring the block-size distribution of typical tunnel muck. This high level of precision demonstrates the suitability of advanced drone technology for detailed geotechnical characterization.
Following data acquisition, the sequential images are processed using a photogrammetric pipeline based on Structure from Motion (SfM). The SfM algorithm assumes a pinhole camera model, where the projection of a 3D world point onto the 2D image plane is given by:
$$ \lambda P = K[R|t]P_w $$
Here, $\lambda$ is an arbitrary scale factor, $K$ is the camera intrinsic matrix containing focal lengths and principal point coordinates, $[R|t]$ represents the extrinsic parameters (rotation matrix $R$ and translation vector $t$) describing the camera’s position and orientation in world space, and $P_w$ is the homogeneous coordinate of the point in the 3D world. This process begins with feature extraction using the Scale-Invariant Feature Transform (SIFT) algorithm to establish robust correspondences between overlapping images. The sparse point cloud generated by SfM is then densified using Semi-Global Matching (SGM), resulting in a highly detailed 3D point cloud model of the entire waste dump. This model, measuring 51 meters in length, 40 meters in width, and 15 meters in height, provided the foundation for all subsequent analyses.
The segmentation of individual rock blocks from the generated orthophoto was achieved using the Segment Anything Model (SAM). This advanced deep learning model, pre-trained on a vast corpus of visual data, demonstrates exceptional adaptability to the irregular shapes and textures of rock fragments. The output consists of precise polygonal contours for each visible block on the surface. However, we observed that the model’s performance degrades under conditions of severe occlusion or overlap, and its accuracy for fragments smaller than 50 mm is currently limited, often requiring manual verification. To systematically analyze the spatial variability of the muck pile, the orthophoto was divided into four distinct zones based on topographical features: the slope crest, the slope surface, the slope toe, and the latest dumping area. This zonal approach is critical for understanding the transport and deposition mechanics acting on the muck.
| Zone | Description | Dominant Block Size |
|---|---|---|
| Slope Crest | Upper plateau area | Fine to medium |
| Slope Surface | Intermediate inclined zone | Medium to coarse |
| Slope Toe | Base of the slope | Coarse to very coarse |
| Latest Dumping | Recently deposited material | Highly variable, reflects blast |
The geometric characteristics of the blocks revealed a clear pattern of gravity-driven spatial sorting. Analysis of the aspect ratio (long axis/short axis) showed that the majority of blocks across the entire dump are sub-rounded to rounded, with aspect ratios concentrated between 1.2 and 1.8. However, significant variations exist between zones. Blocks at the slope toe are predominantly large, angular, and elongated, suggesting they rolled to their final position with minimal attrition. In contrast, blocks at the crest and surface are smaller, more equidimensional, and sub-rounded, likely due to increased inter-particle collisions and weathering during the initial deposition. The frequency distribution of these morphological traits is a direct consequence of gravitational sorting, where larger, denser fragments accumulate at the toe, while finer material remains near the crest.
| Morphology Parameter | Crest Zone | Surface Zone | Toe Zone | Latest Dumping |
|---|---|---|---|---|
| Modal Aspect Ratio | 1.3 | 1.5 | 1.8 | 1.6 |
| Dominant Shape Class | Sub-rounded | Sub-angular | Angular | Angular |
| Shape Uniformity | High | Moderate | Low | Moderate |
The most impactful result from our analysis using drone technology was the quantitative assessment of block size distribution. The overall median block diameter ($D_{50}$) across the entire waste dump was found to be approximately 300 mm, with a maximum observed diameter ($D_{max}$) exceeding 1,500 mm. The size distribution curves for each zone were distinctly different. The crest zone was dominated by fine material, with more than 80% of the blocks passing a 100 mm sieve. The surface zone showed a wider distribution, while the toe zone was characterized by a significantly coarser gradation, with 90% of the material passing a 1,000 mm sieve only. The latest dumping area, which best represents the “as-blasted” condition prior to gravitational sorting, had a block size distribution that offered direct feedback on the efficacy of the drilling and blasting operations. The oversized material ratio (blocks exceeding 1,000 mm in diameter) was found to be 16.4% for the entire dump but only 2.5% in the latest dumping area, indicating a recent improvement in blasting practice, though large fragments of up to 2,000 mm were still present.
| Gradation Parameter | Crest | Surface | Toe | Overall Dump |
|---|---|---|---|---|
| $D_{10}$ (mm) | 15 | 25 | 100 | 40 |
| $D_{50}$ (mm) | 80 | 180 | 500 | 300 |
| $D_{90}$ (mm) | 250 | 450 | 1200 | 800 |
| Oversize Ratio (>1m) | 0% | 5% | 25% | 16.4% |
The spatial distribution of block sizes directly informs a zonal resource utilization strategy. The fine to medium-grained material abundant at the slope crest and on the slope surface is ideally suited for direct use as general fill or as road base material, requiring minimal processing. Conversely, the coarse, angular rocks that accumulate at the slope toe are less suitable for direct application but serve as excellent feedstock for mechanical crushing and screening to produce high-quality coarse and fine aggregates for concrete and other value-added products. The latest dumping area, specifically, acts as an in-situ laboratory. By correlating its block size distribution with the known blast design parameters, we can optimize drilling and blasting patterns to minimize oversize generation. The analysis suggests that a significant cause of oversized blocks is the current blasthole spacing, which approaches 1.3 meters, particularly for cut and bottom holes. To reduce the generation of blocks larger than 1,000 mm, it is recommended to decrease the blasthole spacing and optimize the inter-hole delay timing, as quantified by the relationship between burden and blasthole diameter. The charge concentration per hole, currently at 1.3 to 2.0 kg/m, must be carefully adjusted in conjunction with the adjusted spacing to maintain proper fragmentation energy.
In conclusion, this study demonstrates the immense potential of integrating advanced drone technology with state-of-the-art deep learning for the efficient, accurate, and scalable characterization of tunnel waste rock. The application of drone technology allows for rapid, non-invasive surveying of entire waste dumps, overcoming the limitations of traditional manual methods. The derived geometric and gradation data provide a rational basis for developing a zonal resource utilization plan, directing fine material to low-specification uses and coarse material to high-value processing. Furthermore, this approach provides a powerful feedback mechanism for optimizing drill-and-blast operations, targeting specific fragment sizes to reduce secondary blasting and increase overall operational efficiency. The future of sustainable tunnel construction lies in the digitalization of its waste streams, with drone technology serving as the primary tool for data acquisition and intelligent analysis.