Introduction
In the mining industry, the size distribution of rocks following a blast—known as fragmentation—is a critical Key Performance Indicator (KPI). Proper fragmentation is essential for operational efficiency; oversized rocks can damage hauling equipment and cause bottlenecks at primary crushers, while excessive "fines" (very small particles) indicate wasted explosive energy and increased processing costs. Traditionally, measuring this distribution has been a choice between the highly accurate but labor-intensive sieve analysis and manual 2D image analysis, which is faster but often lacks spatial resolution and exposes technicians to hazardous environments.
This paper presents a proof-of-concept system designed to automate rock fragmentation analysis using quadrotor Unmanned Aerial Vehicles (UAVs). By integrating an autonomous aerial pipeline with industry-standard analysis software, the researchers achieved a significant reduction in data collection time while maintaining accuracy levels suitable for industrial application.
System Architecture and Methodology
The proposed system leverages a sophisticated software and hardware stack to move from image capture to data output in near real-time. The core of the flight control and data management is built on the Robot Operating System (ROS), which coordinates the UAV's movements and image streaming.
The experimental setup utilized a Parrot Bebop 2 quadrotor. To simulate a reliable positioning environment indoors, a Vicon motion capture system consisting of ten cameras was employed to track reflective markers on the UAV, effectively serving as an "indoor GPS." The workflow follows a structured path:
- Autonomous Flight: The UAV follows a predefined path, capturing high-resolution images at varying altitudes with a 50% overlap to ensure comprehensive coverage.
- Data Streaming: Images are streamed to a ground workstation.
- Automated Processing: Using MATLAB for data management and Split-Desktop for image processing, the system automatically delineates particle boundaries.
- Statistical Analysis: The resulting data is fitted to a Swebrec function to generate particle size distribution curves.
Experimental Results
The system was validated in a laboratory setting using a custom-built rock pile of known composition. A ground truth was established through traditional sieve analysis, against which the UAV-based results (16 automated photos) and a manual "fixed-camera" baseline (11 manual photos) were compared.
Efficiency and Speed
The most striking result was the gain in operational efficiency. The manual image collection and analysis process took approximately 55 minutes and 52 seconds. In contrast, the automated UAV pipeline completed the entire cycle—from setup to the final export of distribution data—in just 10 minutes and 2 seconds. This represents an ~82% reduction in time effort, or a 5x speedup in data throughput.
Accuracy and Consistency
The UAV method demonstrated high accuracy, predicting the rock size distribution within 17% of the sieve analysis ground truth. Furthermore, it remained within 6% of the results obtained via manual image analysis. While individual frames from the UAV showed higher variance than curated manual shots—likely due to the lack of human "curation" of the angles—the system's ability to aggregate a larger volume of data quickly allowed the cumulative results to stabilize and provide a reliable statistical representation of the muck pile.
Discussion and Key Insights
The research highlights several advantages of an autonomous aerial approach:
- Data Aggregation: The "more is better" effect suggests that the speed of UAV collection allows for a higher sampling frequency, which can compensate for the lack of human curation in individual images.
- Safety: By utilizing "expendable" UAV hardware, mining operations can remove human technicians from the "muck pile," a high-risk zone characterized by unstable ground and proximity to heavy machinery.
- Real-Time Feedback: The rapid turnaround of fragmentation data enables "on-the-fly" adjustments to downstream milling operations or future blast designs, a capability currently hindered by the lag in manual reporting.
Conclusion and Future Work
While successful as a proof-of-concept, the system faces challenges in transitioning to full-scale mining environments. The researchers noted "edge effects" where the boundaries of the laboratory bin were occasionally misidentified as large fragments. Future iterations will focus on replacing physical scale objects with onboard sensors such as LiDAR or stereo cameras for calibration. Additionally, implementing Simultaneous Localization and Mapping (SLAM) will be necessary for operation in GPS-denied or high-dust environments typical of active mine sites.
Summary updated