Integrated source procurement, tanker configuration and 24/7 job-chaining for bulk spring water logistics: owned, leased and spot sources, water-profile compatibility, PBS road-access limits, licence caps, driver shifts and full-load pickup-and-delivery on the real Victorian road network. Supply Chain Lab, RMIT University.
Spot water is the expensive recourse tier: the optimiser holds it back until owned and leased licence caps bind (summer peak or outages), mirroring the company's three-tier procurement structure.
Customers accept water only from nominated sources (taste/composition). Sites marked ▲ cannot receive A-double combinations due to road-access restrictions.
Baseline: dedicated shuttle lanes, the common manual-planning pattern: each tanker is locked to one customer lane with its traditional sources and shuttles back and forth all shift. Both modes queue for limited loading/unloading bays (Millbrook, Kinglake and metro plants have 2 bays; other sites 1), so congestion and waiting emerge naturally as demand and service times rise. Optimised: a multi-start randomised construction (GRASP-style, 12 starts) followed by inter-route local search - swap and relocation of planned jobs between vehicles on the dispatch board (no water is ever transferred between tankers) - jointly choosing procurement tier, source, vehicle configuration and job sequence. Distances and travel times come from the OSRM road network matrix (heavy-vehicle time approximated as car +8%; straight-line fallback if offline). This in-browser heuristic demonstrates the concept; the production model is an exact time-space MILP solved in Gurobi with a tactical allocation layer.
A bulk spring-water producer buys water at several springs, hauls it in tankers to bottling plants, and profits on the margin between what the water costs and what the delivered load earns. Sourcing and routing cannot be decided separately: the cheapest spring is worth nothing if the road to it bans a heavy combination, or if every tanker is queuing at its loading bay.
The page solves the whole problem in your browser. It builds a road distance and time matrix from OSRM on the real Victorian network, then runs a multi-start construction with inter-route local search over source choice, tanker configuration and 24/7 job chaining, respecting PBS road-access limits, licence caps, water-profile compatibility, bay queues and driver shifts. Drag the season slider to re-solve; click a spring to knock it out and watch the plan recover.
Watch what binds. In a wet season the licence caps rarely bite and the answer is driven by haul distance; in a dry season the cheap springs run short and the plan pays for longer hauls to keep the plants running. The Monte-Carlo panel is the honest test: a plan that wins on the mean day and collapses across 200 sampled days is not a plan worth signing.
Basemap © OpenStreetMap contributors; routing via public OSRM / FOSSGIS services. Interactive prototype on synthetic data.
