EN
Back to the archive

The encyclopedia · Engineering & Operations · Operational decision · 1981–1983

Air Products planned gas deliveries with a solver that handled a huge model

Air Products linked customer tanks to delivery trucks in one optimizer, saving 6-10% of operating costs.

Air Products and Chemicals

the move

Air Products delivers industrial gases into tanks at customer sites, so serving a customer depends on how much gas is left in a tank and whether a truck can reach it in time.

The old way asked dispatchers to judge both by feel. The firm built a system that forecasts each customer's usage, builds a time-and-distance network, and solves the day's routes with mixed-integer programming using Lagrangian relaxation, which could push models of hundreds of thousands of variables near to optimality.

Since October 1981 the optimizer has been saving 6 to 10% of operating costs, and the model also lets a dispatcher change a schedule and immediately see the cost, so the plan stays both optimal and practical.

why it works

  • Forecasting tank use means the truck arrives before the customer runs out, avoiding emergency runs.
  • Solving inventory and routing together replaces two guesses with one plan.
  • Near-optimal mixed-integer solutions on a very large model capture savings that manual routing leaves on the road.
  • A fast interactive interface means a dispatcher can keep the solution and still handle exceptions.
the payoffRoute the tanker from what each tank needsinspired

what transfers

When tanks and trucks bind the business, solving delivery and tank levels together beats dispatching each truck on experience.

what came after

The system stayed in production and became a reference for solving large vehicle-routing and inventory problems; the Interfaces paper is among the most-cited from the early Edelman years. It helped establish that a well-built optimizer could be made to run in the field, not just in a study.

references

spotted an error? The archive wants to know.

same kind of clever