This thesis investigates whether refuelling, cleaning and resting decisions on fixed multi-day routes can be improved by an integrated, data-driven decision aid for Nijhof–Wassink. The problem is modelled as a Markov Decision Process and addressed with a Deep Q-Network trained on historical three-day routes enriched with station and cost information. The resulting policy is evaluated against a heuristic, the current decision method and a Mixed-Integer Linear Programming benchmark. On the constructed test routes, the DQN reduces expected operational costs by 19.33% compared to the current decision approach and by 3.73% compared to the heuristic, with an average optimality gap of 13.40% relative to the MILP solutions. Scenario and sensitivity analyses indicate that the policy remains effective under changes in station availability and cost parameters, suggesting that the approach is a promising basis for an integrated decision-support tool in bulk logistics.