Drone-Assisted Stockpile Volume Estimation in Open and Confined Spaces

Abstract

This thesis aims to develop unmanned aerial system solutions for stockpile volume estimation in both open and confined spaces. It starts with a comprehensive literature review that examines both traditional and recent stockpile volume estimation techniques employed in various environments. It was found that recently emerging aerial methods, such as drone-borne LiDAR sensors, can enable notable advantages including speed, safety, occlusion elimination, and enhanced accuracy compared to current typical industrial solutions. However, there is still a notable gap in research represented in the underdevelopment of cost-effective aerial solutions for safe and precise volume estimation within confined spaces. The research in this thesis starts with a detailed investigation of the state-of-the-art in utilising drones for operations within treacherous conditions, particularly within industrial confined spaces. It was found that existing studies have not thoroughly examined drone missions under operational constraints such as absence of GPS signals, dust-laden air, and poor/lack of visibility. These limitations defined the way for a relatively novel application where drones could be deployed to enhance inspection while augmenting safety measures. Following the establishment of this fresh perspective, mission planning, instrument development, and implementation of control and navigation strategies were assessed across diverse confined spaces and for various stockpile volume estimation missions in both simulated and real-world scenarios. A thorough cost-benefit analysis elucidated that drone-based solutions for stockpile volume estimation within confined spaces achieve a high Cost-Benefit Priority Factor (CPF) of 133-200. Moreover, this approach surpasses traditional industrial fixed sensor systems in flexibility, initial cost savings, and ability to serve multiple sites. Advancing further, a low-cost, yet effective approach was proposed that relies on actuating a single-point light detecting and ranging (1D LiDAR) sensor using a micro servo motor onboard of a drone. The collected LiDAR ranges were converted to a point cloud that allows the reconstruction of 3D stockpiles. The proposed approach was assessed via simulations of a wide range of mission operating conditions. The influences from modulating the drone flight trajectory, servo motion waveform, flight speed, and yawing speed on the mapping performance were all investigated. Comparing the volumetric error values, the average error from the proposed actuated 1D LiDAR system was 0.9% as opposed to 1% and 0.8% from the 2D and 3D LiDAR options, respectively. That said, compared to 2D and 3D LiDARs, the proposed system requires less scanning speed for data acquisition, is much lighter, and allows a substantial reduction in cost. Experimental tests on drone-based solutions for scanning a reference stockpile were conducted with either single or multiple drones equipped with 1D LiDAR sensors, achieving an average volumetric error of 2%. In contrast, the actuated single-point LiDAR system exhibited a higher volumetric error of 5% due to the significant number of outlier points involved. Finally, as the previously presented solutions required an external localisation system for their operation within confined spaces, this thesis paved the way to get rid of such requirement via applying an ICP (Iterative Closest Point) algorithm that can operate independently of such systems. The proposed algorithm uniquely employed a low-rate, low-dense LiDAR scan, specifically focusing on the horizontal layer of a 3D LiDAR for localisation and scan matching. Furthermore, a wall-following navigation strategy was employed for indoor navigation and path-planning to further streamline the mapping process. It was shown that the estimated volume of the reconstructed stockpiles has an average volumetric error of 3.7%, but this figure was enhanced to 0.4% when applying loop closure. Moreover, mapping using an actuated single-point LiDAR approach was processed using the ICP localisation method, resulting in a 1.4% volumetric error.