Haul-Truck Fleet Digital Twin and Dispatching Using Discrete-Event Simulation and Reinforcement Learning

A digital twin of an open-pit haulage system (4 loading units, a crusher pocket, a waste dump, 6-32 CAT 793F-class trucks, stochastic loading, travel and dumping, shovel and truck breakdowns) is built as a discrete-event simulation, calibrated from fleet-management telemetry and used to compare truck dispatching rules with a PPO reinforcement-learning dispatcher, on a standard laptop CPU, 16 GB RAM. All data are SIMULATED: no public fleet- management data exist without an operator agreement. The 'real mine' is the same simulator with hidden parameters that the calibration never reads; every result is a result about this simulator, not about a real mine. Twin: C2 predicts the real mine's production within 0.40 % (handbook twin +10.98 %). Test (200 shifts, common random numbers): Greedy 128.26 kt per shift, +5.3 % over fixed allocation; PPO 128.42 kt, +0.16 kt over Greedy (CI +0.07 to +0.26), so the specification's RL target (+2-8 % production or -10-30 % queue time) is NOT MET. Research and teaching prototype on simulated data: not a dispatching system for a real mine.

Category: Mining Engineering

Mining Engineering Project MN08 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.