Optimizing ore transportation at open-pit mining sites requires an intelligent solution able to coordinate multiple agents while dealing with possible equipment failures. When ore is extracted from a site, it is first transported to ore preparation plants (OPPs) before moving on to processing plants. The haul trucks responsible for transporting the ore to the OPPs consume a large amount of diesel, making them a significant expense for the mining operation.
We partnered with Simwell to demonstrate how reinforcement learning and simulation can improve haul truck routing to minimize distance travelled while also optimizing around equipment failures. The Pathmind policy outperformed a shortest queue heuristic to increase the total amount of ore sent to the OPPs.
An open-pit mine contains routes connecting five shovels, or ore extraction sites, and five OPPs. A fleet of ten haul trucks picks up ore at the shovel sites and moves it to the OPPs. The simulation monitors total distance driven, mean cycle time, percentage of equipment utilization, the amount of ore transported, and the transportation rate by hour.
During operations, equipment failures at the shovel sites and the OPPs can occur. These malfunctions need to be factored into the optimization strategy since they can greatly impact the best routing choices. A simple strategy of always sending a truck to the shortest line might produce good results if all equipment worked without failure, but a real-world optimization strategy needs to take those malfunctions into account.
Why Pathmind Was Needed
The optimization method to solve this routing problem needed to be smart enough to work around equipment failures to increase overall ore processing. Many optimization tools cannot meet the challenge because they are not able to learn from new data and respond quickly. Reinforcement learning is adaptable, helping to both reroute trucks as equipment failures happen, and predict if something is about to break down and reroute the haul trucks before it happens.
Pathmind’s reinforcement learning was able to optimize the movements of the entire haul truck fleet while working around equipment failures to keep the operation running efficiently. The final solution increased ore processing by 19%.
The simulation used in this project is also a good example of how reinforcement learning can be used to continue optimizing operations after reaching your initial goals. If the AI had determined that one OPP is never used because it lowers the overall performance of the operation, it could be a good business decision to remove that plant to further reduce costs. Adding reinforcement learning to your optimization strategy can often reveal surprising new insights.
SimWell is an Industrial Engineering firm creating a simulation model for every need. It specializes in simulation and optimization in the sectors of mining, manufacturing, supply chain and government.