Locomotive fuel management with inline refueling

•We study the inline refueling as augmenting the fuel in locomotive tanks by hauling wagons with tanks.•We describe the theoretical complexity of the problem.•We develop a mixed-integer program based on a time-space network.•A case study in Australia is presented to evaluate the adoption of inline refueling.•We investigate the impacts of inline refueling on costs incurred by unpredictable fuel consumption increases. Fuel and fuel-related expenses constitute a major part of the operating costs of railway companies. Hence, improvements in fuel management often lead to significant annual operational cost savings. The traditional approach to reduce the fueling costs is to fill the locomotives at inexpensive stations to bypass the more expensive stations. However, this approach is not applicable to rail networks with long-haul operations. Railway companies have started adopting inline refueling tanks, a supplementary reservoir which can refuel locomotives during a trip, and can be refueled, attached, detached, and swapped at any station, independently of locomotives. This new technology offers many opportunities for railway companies, in particular to facilitate long-hauls and to leverage fuel cost differences. As always, new opportunities mean new challenges, as inline tanks engender combinatorially many new possible refueling plans. Moreover, since the inline tanks are a substantial investment, the number of available inline tanks is limited. To tackle this, we design a first optimization model that determines the assignment to trains and refueling plans for a fleet of inline tanks. Our analysis shows on an Australian case study and on the INFORMS data set for the USA that inline tanks can lead to a substantial return on investment, thereby informing strategic-level decisions such as the purchase of inline tanks or the closing down of fuel stations. Furthermore, we provide analyses that reveal further managerial insights on the benefits of inline refueling. Finally, we empirically demonstrate the difficulty of this problem and develop a heuristic to tackle large instances.