In operational airline crew scheduling, one of the major tasks is to allocate flight activities to individual crew members. This task is referred to as the crew rostering problem; an optimization problem that aims to assign flights to crew members in a cost-effective way, while also considering preferences of crewmembers. Flights are combined into multiple-day activities called pairings that consist of flight duty days and rest period days. For intercontinental long-haul destinations specifically, pairings may take more than a week. During flight duty days of these long-haul pairings, crew members are not situated at the airline base location. During rest period days, however, they have the opportunity to be at their homes or with their families. The pairings that crew members on long-haul crew divisions operate, thus, directly impact their life
apart from work as well as the locations at which they are situated during flight duty days. Therefore, crew preferences are highly valued by crew members, as these provide their only means to influence duties. This is a critical reason for crew preferences to be a recurrent topic in negotiations on collective labour
agreements between crew unions and airlines that influence crew resources. As crew costs are the biggest expense for airlines next to fuel (Belobaba et al., 2015), crew preferences are considered in an environment that is predominantly designed to manage crew resources as efficiently as possible. For long-haul cockpit crew in European airlines, crew preferences can be expressed via requests for the operation of specific pairings. However, the problem with these pairing requests is that the requests concern pairings that commence months later than the pairings considered by the crew rostering problem. This
problem is typically solved only one month before operation of the pairings. It is impractical to consider the pairing requests as part of the crew rostering optimization problem, since the problem size of the optimization problem becomes too large to be solved within practical computation time. In other words, a decision on whether or not to grant a certain pairing request has to be made with limited information available on the consequences of that decision since it concerns a part of the schedule that is not optimized for. When a pairing request is granted for a crew member, this invokes a pre-assigned pairing for this crew member. Such pre-assigned pairings constrain the construction of this crew member’s roster in the crew rostering problem, as only a limited number of options exist to efficiently construct this crew member’s roster. When many of these pre-assigned pairings exist in the schedule, this leads to inefficiencies in the schedule. Such inefficiencies cause loss days without productive activity in the schedule, which are an indicator of inefficient use of crew resources leading to higher crew costs.

Pilots are an important asset in the operations of airlines, as with insufficient pilots, an airline cannot operate the planned schedule. The long-term crew planning problem aims to optimally plan the supply and demand for pilots as well as methods and strategies for closing the gap between these. The demand is determined by the number of pilots required at each crew position based on the demand for the flight schedule, training requirements, contractual agreements for holidays and off-days. The supply, on the other hand, is determined by the current pilot workforce and changes over time by factors such as retirements and illness. Finally, the inevitable gap between supply and demand can be closed by planning transitions for pilots to different crew positions and by recruiting new pilots. This project addresses the cockpit crew transition planning problem. This is an optimisation problem that aims to optimally assign transitions for pilots with the goal to minimise the shortages and surpluses in the supply of pilots, constraint to for example contractual and governmental agreements and capacity limitations. A heuristic local search method has been developed to plan and evaluate transitions for pilots between different crew positions. A tree search method is then used to select, at each iteration, the transition with the highest potential. In an experiment on various configurations of the model, the results are compared and recommendations are made on the best method to model the cockpit crew transition planning problem.

Airline schedules are constantly being updated to deal with disruptions originating from crew absence, upstream delays, mechanical failures and other sources. A robust schedule is capable of dealing with- or absorbing negative effects of such unexpected events. Schedule robustness, accounted for proactively in the scheduling g phase, is extensively researched for the crew scheduling problem, but the subproblem of simulator-based crew training is neglected. Airlines periodically train cockpit crew members for legal licensing purposes (recurrent training) and crew conversion between fleets and ranks (conversion training). This operation involves up to thousands of working days for large airlines. Any disrupted training activity potentially impacts crew availability by means of missed due dates in case of recurrents or postponed employment in case of conversion. The vast cost and risk related to this problem has led to the development of a robust cockpit crew training schedule. First, an integrated Training Scheduling & Assignment Model (TS&AM) is developed to construct a training schedule. A Monte Carlo Simulation is then applied to generate disruption scenario’s, which are then solved using a rule-based recovery heuristic. The output of the recovery model is captured in features that translate into expected recovery cost, which are then used in a feedback loop. Two feedback variants are tested: (1) Proportional Feedback (PF) and (2) Neural Network (NN) feedback. The feature dependent expected recovery cost is then used in another cycle of the TS&AM to output a robust training schedule. As a final step, the robust training schedule is tested using the same simulation evaluation framework. Experiments have shown that a 16.75 percent improvement on recovery cost or cost of crew unavailability due to missed due dates can be achieved by learning ways of introducing schedule robustness. The lowered recovery cost also translates to a 1.11 percent overall cheaper operation of the training schedule.

In an operational airline environment, disturbances to the planned flight schedule cannot be avoided. In the case of crew absences, airlines can employ reserve crews to cover open positions in crew schedules. The airline reserve crew pairing problem is concerned with the determination of optimal airline reserve crew patterns, which are defined by the number, lengths, and start times of reserve crew pairings. Currently, there are no existing solutions that solve the airline reserve crew pairing problem for long-haul cockpit crew.

This project addresses the airline reserve crew pairing problem for long-haul cockpit crew. A reserve pattern evaluation model has been developed that uses airline simulations to determine the quality of existing reserve patterns. This evaluation model is used iteratively by four optimisation algorithms that aim to generate optimal reserve patterns. The solutions generated by the algorithms are compared to each other in a number of experiments. Following these experiments, recommendations are made on how the gap between expected reserve demand and scheduled reserve capacity can be minimised.