Personnel scheduling for aircraft line maintenance-solution methods for the tactical and operational decision level.

Labor is one of the most significant costs for companies in highly developed countries. In the last decades, many companies have relocated their production facilities to low wage countries. Since service needs to beprovided on site, service companies, however, cannot escape from the costly wage expenses. Therefore, it is very important that overstaffing and idleness are avoided by creating good personnel schedules. The personnel scheduling problem we will study in this dissertation originates froma line maintenance department of an aircraft maintenance company. A line maintenance program represents the regular short inspections of aircraft between their arrival and their departure at some specific airport. Whereas the workload that stems from the tasks that need to be done during these inspections is rather stable, the main source of uncertainty forthese inspections is the timing of the demand. Delays can originate forinstance from earlier flights, bad weather conditions, or air traffic management. In this dissertation, we propose methods which should facilitate the scheduling in practice, and offer thereby an alternative for thecurrent practice of scheduling by paper and pen. Chapter 2provides a classification of the personnel scheduling literature from 2004 onwards. The classification fields, such as applied methods, decision delineation, and uncertainty incorporation, guide the reader in his/her search for specific problem settings and the corresponding approaches of the research community. The current literature mainly addresses the staffing and/or scheduling of workers considering fixed inputs. We adviseresearchers to integrate multiple decisions into the personnel scheduling problem such as demand forecasting, machine scheduling, or considering multiple locations. The last decades, companies more and more consideremployee preferences (such as requests for specific working days or shifts, assignments to a specific location or working partner, preferred durations or start times) in order to satisfy the workforce and to allow them to flexibly manage their personal lives. There are still great opportunities in finding algorithms that efficiently cope with those preferences. Lastly, most papers appear to feature a deterministic approach, while real-world personnel scheduling problems have to deal with a variety of uncertainty sources. In situations where uncertainty has a strong effect on the personnel schedule, such as volatile demand or last-minute changes, it could prove very beneficial to incorporate this uncertainty inthe decision-making process. Instead of integrating this uncertainty, researchers could also test the robustness of their solutions, for instance by simulating the stochastic behavior of demand, or worker availabilities. In the literature, however, we hardly ever encountered this type of analysis. In Chapter 3 the different types of aircraft maintenance that appear in literature are discussed. We found the terminology consisting of many overlapping definitions very confusing. To obtain a better insight into the different types of aircraft maintenance, we provide a taxonomy that classifies all types and we discuss the literature that addresses the combination of aircraft maintenance and personnel scheduling. In the second section of this chapter we provide the details of a given personnel scheduling problem of an aircraft maintenancecompany. In this problem, personnel rosters need to be created while minimizing the labor costs that stem from the provision of a line maintenance program. This problem setting is the basis for the problems studied in Chapters 4, 5 and 6. Chapter 4 is dedicated to a three-step evaluation procedure for personnel rosters. Due to the usage of heuristics and multi-objective performance criteria, solution procedures typically generate multiple solutions, or in this case, personnel rosters. For the management, it is difficult to create a ranking among these solutions. Therefore, we present an approach in which the rosters, which arecreated based on a deterministic problem setting, are used in a stochastic environment. This results in performance criteria, such as service rate and tardiness values, that are evaluated by means of a Data Envelopment Analysis (DEA). This approach should enable that management decisions are no longer based on instinct but on objective criteria. In Chapter 5, we study two variants of the resource loading problem (RLP), in which a number of jobs with time windows need to be scheduled with monotone execution patterns. This latter characteristic ensures that a given job can only be assigned a non-decreasing (resp. non-increasing)number of workers over the periods in which the job is processed. We distinguish between RLP-selection in which no non-regular capacity isavailable and selected jobs yield a reward, and RLP-capacity in which all jobs need to be executed and non-regular capacity is minimized. The resulting integer programming (IP) formulation for real-life problem instances tends to be too hard in order to be solved in limited time with a commercial solver. Therefore, we carry out a Dantzig-Wolfe decompositionand try to solve the resulting problem with column generation. Unfortunately, these two CG methods do not give good results. The CG method thatuses execution schemes (i.e., columns which state in which period and with which intensity a job should be executed) has the advantage that it is the only solution method which is able to find results for the large instances (with 300 flights). Finally, Chapter 6 follows with the presentation of a proactive/reactive policy to deal with the maintenance job scheduling problems on the operational level which areinfluenced by uncertainty in the arrival times of the jobs. The proactive policy consists of creating a personnel roster and a baseline schedule which take into account the delay distributions of the jobs and schedules each job as late as possible. In the reactive policy, we use a rolling horizon algorithm which makes use of the most recent information withrespect to the arrival times of the jobs. The results indicate that thequality of information is highly important. Better estimations on the real release times of the jobs (i.e., the arrival delays of flight) and the earlier these are retrieved, the cheaper the reactive scheduling policy. Also the job characteristics play an important role with respect to the costs which have to be made to create feasible schedules.status: publishe