Public dependence on air transportation has grown to its largest point in history. Along with this increased dependence is a heightened awareness of safety concerns and the need for pilots to cover all scheduled flights. All commercial pilots are certified for the particular aircraft they are flying by satisfactorily completing a Federal Aviation Administration (FAA) approved course. During the training course, a number of training devices are used including a Full Flight Simulator (FFS) and a Flight Training Device (FTD). The procurement, installation, operation, and maintenance costs of these devices are expensive. In addition, the amount of time pilots spend in training is costly because they continue to be paid their salary rate although they are unable to fly revenue-generating flights for the airline. Due to these high costs, training is scheduled very tightly, with the goals of maximizing training device utilization and minimizing the pilot's training footprint (time spent in training). Any disruption to the tight schedule, such as a simulator breakdown, or pilot illness, renders the original schedule obsolete and demands rescheduling of activities.

In this research, the rescheduling problem is investigated through the development and application of several different rescheduling approaches. The problem is decomposed by first investigating a single resource model and insights gained from this experimentation are transferred to the multiple resource model. The solution approaches developed for experimentation include: right-shift rescheduling (RSR), rescheduling of affected activities (RAFF), and rescheduling of all activities (RALL). Performance measures used to compare the various approaches include the minimization of the pilot footprint, the minimization of pilot tardiness, and the minimization of the deviation that a revised schedule has from the original schedule. A case study and a series of experiments involving random disruptions to original schedules were used to analyze the solution approaches.

For the data sets analyzed, the RAFF algorithm outperformed other methods with respect to the majority of the measure collected. Analysis performed on the amount of slack in the original schedule revealed that diminishing returns were observed beyond a certain level of slack. Further analysis on the impact of the location of this slack showed that the majority of the slack should be placed at the end of the schedule, or the slack should be dispersed almost evenly over the entire schedule.