A Heuristic to Reduce the Maximum Work-in-Process for Aircraft Maintenance

Abstract

Air Logistics Complexes (ALCs) provide depot maintenance for aircraft fleets for the United States Air Force (USAF). The rate at which aircraft are inducted into maintenance is in constant flux. For example, initiation of new modification programs, aging aircraft fleets, and demand for new work all impact future workload requirements and planning. The induction schedule identifies the dates that aircraft arrive at a maintenance depot. A good induction schedule can help minimize flow days (time an aircraft is in maintenance) while a poor induction schedule can result in increased work-in-process (WIP) and long aircraft maintenance queues. Depot maintenance can be modeled as a job shop scheduling problem. Many heuristics and metaheuristics have been proposed to solve variations of job shop problems.

This paper utilizes a simulation tool, RAMP, which was developed to simulate the maintenance performed on aircraft and produces results about the utilization, queue lengths, and other performance metrics of the system. This thesis describes a developed heuristic, which is used to alter induction schedules in order to smooth WIP curves and reduce queue times. The algorithm keeps most of the induction schedule un-altered by manipulating induction dates for a relative few aircraft around important times in the simulation, namely around large extrema.

The algorithm uses a numerical derivative and an application of the second derivative test to detect important extrema. Once these extrema are identified, it attempts to reduce aircraft WIP around peaks and fill aircraft WIP into valleys. This is performed by shifting the induction dates of aircraft around those extrema.

The algorithm was successful in reducing the maximum WIP for 12 out of 16 schedules tested. Of the successful runs, there was an average improvement of 8.7% on maximum WIP. Performance of the heuristic is dependent on the parameters used; however, modifications were made to mitigate this dependence. The algorithm can function well in its current version; however, there is still additional work that can be done to further increase performance.