Abstract
Rock fragmentation is a critical initial phase of the mining extraction process and has a substantial impact on downstream operational productivity. It is crucial to continuously measure rock fragmentation to optimize blast design, potentially leading to significant enhancements in mill throughput and slope stability. Presently, analyzing rock fragmentation involves physically screening the blasted rock material or utilizing image analysis on the resulting muck piles. Physical screening is costly, time-intensive, and impractical on a large scale. On the other hand, image analysis techniques are cost-effective, but their accuracy and outcomes are influenced by factors such as lighting conditions, variations in rock colour, shadowing, areas of fines, and the presence of overlapped and non-overlapped particles.
This study pioneers a novel approach by harnessing laser scanner data for point cloud segmentation, offering a promising solution to overcome the limitations of lighting conditions, colour variation, and shadowing in traditional image analysis methods. To support this novel technique, creating a 3D labeled dataset becomes crucial for training a deep learning model capable of automating the segmentation of rock piles in point cloud form. Recognizing the time-intensive and expensive nature of creating such labeled datasets, this study introduces an innovative approach by generating an automatically labeled synthetic dataset of blasted muck pile point clouds using Blender, a versatile platform widely employed in autonomous robotics and the automotive industry.
Trained on the synthetic 3D dataset, the deep learning model enables precise segmentation of rock instances in three-dimensional coordinates. This stands in contrast to current methods that yield 2D segmentation outputs, providing an accurate representation of the rock object in 3D. The results offer a foundation for detailed analysis of size distribution and particle shape/angularity, advancing the accuracy and efficiency of rock fragmentation assessments.
Importantly, this method depends on coordinate data instead of RGB information, rendering it particularly applicable in challenging conditions such as underground mining, night shifts, or situations where maintaining optimal lighting conditions is difficult or costly. The findings present a significant leap forward in rock fragmentation analysis, opening avenues for enhanced practices in diverse mining environments.