- 183 views
- 193 downloads
Data-Driven and Artificial Intelligence Approach to Dynamic Truck Fleet Dispatching and Shovel Allocation Planning in Open-Pit Mines
-
- Author / Creator
- Noriega, Roberto
-
An open-pit mine is a highly dynamic environment where different equipment resources are allocated to mining areas to extract metal-bearing rock and waste, for pit development, following a set flow of activities. The material mined is then transported through the mine road network to different destinations like processing plants, waste dumps, or stockpiles. Open-pit short-term planning defines the mining sequence, destinations, and equipment allocation decisions over short periods of time to meet the production and pit development targets established by the long-term strategic plan. It is a process that relies on estimating equipment productivity and mining areas' rock properties for the decision-making process. Therefore, it incurs high uncertainties from the operating cycles of mining equipment and their breakdowns, the interaction of loading and haulage equipment, and imperfect geological information from the mining areas.
This research proposes a practical framework for robust open-pit short-term planning based on Deep Reinforcement Learning (DRL). DRL is the Machine Learning (ML) branch that deals with computational approaches to learning an optimal sequential decision-making policy in an uncertain environment. Recent advances in ML have allowed the development of DRL frameworks for sophisticated production control systems in different industrial settings that outperform current practices. However, applications of DRL within the mining industry are still scarce. This research aims to fill that gap by developing a novel DRL framework for open-pit short-term planning, including shovel allocation and truck dispatching decisions. The proposed framework starts with developing a discrete event simulation (DES) that serves as the open-pit mine production environment based on the equipment dispatch and mine planning database. The DES then serves as a platform that provides feedback on decisions made by the DRL algorithm. Afterwards, the DRL algorithm is designed to achieve targets aligned with the two problems considered in this research, shovel allocation planning and truck dispatching control. The details of the DRL design process include the choice of algorithm, the state and action representations, the design of the function approximator, i.e., the architecture of the deep neural network learning the decisions, and the design of the reward structure to achieve the desired goals.
The DRL framework proposed for the truck dispatch control in open-pit mines provides real-time truck assignments to their next loading and dumping locations to achieve mining, processing, and blending targets. A case study is presented in an iron ore deposit where the trained agent learns a robust dispatching policy to achieve the ore and waste mining targets and maintain the metal concentration of the ore feed to the processing plants within a desired range. The DRL framework proposed for the shovel allocation planning goal is to learn a robust shovel allocation strategy for the next production quarter, 3-month, to meet the tonnes per hour (TPH) production target to be delivered to the crusher feeds, by interacting with the production simulator. Also, the framework is tested in a case study of an iron ore open-pit mine where the shovel allocation agent successfully learns a strategy that consistently delivers the desired production target. This research is expected to contribute to transferring and adopting Artificial Intelligence and ML technologies within the mining industry. -
- Graduation date
- Fall 2023
-
- Type of Item
- Thesis
-
- Degree
- Doctor of Philosophy
-
- License
- This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.