The aerospace industry annually provides transport for billions of passengers along trillions of kilometers. The industry is continuously aiming to provide these services in a more efficient and sustainable way. One possibility is to consider improving airside airport operations, both current types and those expected in the near future. Scheduling airport operations requires taking into account flight planning, airport layout, routing requirements and personnel planning. Current operational planning is characterised by application of linear programming tools for strategic planning, and manual adjustment for adaptive planning.

This dissertation aims to develop data-driven optimisation models, to increase the efficiency and sustainability of various airside airport operations, and to apply these models to airport case studies. The focus is first put on external electric taxiing, a new taxiing technique using electric towing vehicles (ETVs) to tow aircraft from gates to runways and vice versa. Many airports are considering to implement this technique, as it offers a large improvement in reducing their greenhouse gas emissions, noise levels and air pollution, which is an improvement for passengers, airport personnel, and local residents.

The first goal is to create a comprehensive overview of the operational aspects of external electric taxiing, by reviewing existing research work and industry sources. This overview includes the expected specifications of ETVs and the future procedures for electric taxiing movement. Electric taxiing introduces a new airside operation to the airport: ETV-to-aircraft scheduling. Studies on this new operation, as well as on vehicle routing, vehicle fleet sizing and battery charging optimisation models, which are needed for electric taxiing, are reviewed. The overview also includes the remaining research challenges to achieve large-scale ETV implementation in the next few decades.

The second goal is to develop an optimisationmodel to performETV-to-aircraft scheduling that takes into account realistic airport circumstances. A more efficient ETV-toaircraft schedule, which allows more aircraft to be towed by an ETV fleet, will reduce airport emissions more. Some studies have already proposed ETV-to-aircraft scheduling models. However, they do not include all elements needed to make the model realistic and comprehensive, such as routing with conflict and collision avoidance, ETV charging and discharging, and airport surface movement specifications. Two more elements are added to this list in this work: airport electricity capacity and achieving a time-efficient model. Two models are developed for full-day ETV-to-aircraft scheduling, a Mixed-Integer Linear Programming (MILP) model and an Adaptive Large Neighbourhood Search (ALNS) model. Both models limit ETV charging to the electricity capacity of the airport. The ALNS model is able to create near-optimal full-day schedules for large fleet sizes within a few hours, for a large airport case study. The ALNS model is tested with various daily electricity capacity profiles, which shows the necessity of night charging and the effects of increasing amounts of charging during the day.

The third goal is to develop an optimisation approach to retain efficiency for electric taxiing in a real-time situation. The models developed for the second goal are applicable for strategic scheduling. During operation, disruptions to the strategic schedule will occur, and adaptive scheduling is required to continue operation. In this dissertation both a strategic and disrupted scheduling model are developed. The disrupted model reassigns delayed aircraft to ETV, aiming to minimize the changes to the original schedule. The model is used to create an adaptive schedule in a large airport case study using historical flight data. At the start of every half hour period, the disruptions due to flight delays of the next period are incorporated in a new schedule. The results show the efficacy of the disrupted model in minimizing schedule changes, which does not come at the expense of emission savings. In addition to electric taxiing, this dissertation focuses on improving the efficiency and robustness of airside operations by predicting airport disruptions, to avoid additional use of resources and to provide a better service. Where the previous part consists of using models to react to flight delays, operations can also be improved by predicting them in advance. In existing works, delays are predicted by classification or as point prediction. In this dissertation, probabilistic prediction is applied to flight delay, using two machine learning algorithms: Mixture Density Networks and Random Forests Regression. In addition, metrics suited to probabilistic prediction are developed and used to evaluate the algorithm performance. In a small airport case study, the algorithms are shown to be able to predict delays within a Continuous Ranked Probability Score (CRPS) of eleven minutes. The probabilistic prediction algorithms generate estimated delay distributions, which include extended uncertainty information. To illustrate the utility of the predictions for airport operations, they are applied in a probabilistic model aimed to increase the robustness of the flight-to-gate assignment problem. The proposed model is shown to reduce the number of gate-conflicted aircraft by up to 74% when compared to a deterministic flight-to-gate assignment model. The robustness of the assignment can be controlled with a model parameter.

Another method for predicting flight delays is binary classification, which is popular in literature. However, when posed as a binary problem, flight delay and also flight cancellation prediction suffer from a large data imbalance. This causes a distorted view when using metrics such as accuracy. This dissertation develops a systematic approach to binary prediction with imbalanced data, by considering a range of sampling ratios and various sampling techniques. Two machine learning algorithms are applied to a small airport historical flight dataset. The results underline the need to investigate the influence of varying data imbalance ratios on the performance of classification algorithms in various metrics.

Throughout this dissertation, the focus has been on improving the sustainability and efficiency of airport operations through data-driven approaches. These approaches include MILP models, heuristics and machine learning models. The developed models provide support for airport planners to improve current and future scheduling tasks. However, it remains future work to apply similar techniques to other airside operations and to further improve the realism and real-time usability of the current models. In addition, airports’ spatial planners, air traffic controllers and ETV developers will play a critical role in the further development and implementation of electric taxiing. Overall, this dissertation forms a starting point for airport planners aiming to use data-driven methods to improve the sustainability and efficiency of airports, to ensure more durable and reliable air transportation services.@en

Predictive aircraft maintenance is a maintenance strategy that aims to reduce the number of failures, the number of inspections, the number of maintenance tasks and the aircraft maintenance costs. Aircraft are equipped with health monitoring systems, where sensors continuously measure the condition of the aircraft components. In predictive maintenance, these sensor measurements are used to estimate the time left until the failure of these components, called the Remaining Useful Life (RUL). These RUL prognostics are subsequently used to optimize the aircraft maintenance schedule. There are several challenges that complicate the implementation of predictive aircraft maintenance in practice. In this thesis, the threemain challenges are addressed.@en

Aircraft maintenance methods are shifting from conservative maintenance approaches such as a periodic maintenance approach towards predictive maintenance approaches, leading to a reduction of costs, less unexpected aircraft-on-ground events and less wasted useful life of components. In this paper, we propose a new remaining useful life prognostics approach for aircraft engines and integrate this into a maintenance planning framework. First, an explicit health indicator is constructed from an implicit multi-sensor aircraft engine degradation data set using principal component analysis. Then, the remaining useful life prognostic model is developed using a polynomial chaos expansion approach, which allows for uncertainty quantification in the form of a probability density function which is faster than Monte Carlo simulation. A Markov decision process is used to determine the optimal time of maintenance for aircraft engines. During a case study, these optimal times for individual engines being part of a pool of operational engines are integrated using a linear programming model and a rolling horizon approach to obtain an optimised maintenance schedule. The prognostic model performs in the mid-range compared to other papers using the same data sets. Furthermore, with the current cost parameters the integrated remaining useful life maintenance planning strategy reduces costs by a factor of 3 and 2.5 compared to a periodic maintenance strategy or a run-to-failure maintenance strategy, respectively. The waste is reduced by a factor of 2.5 compared to periodic maintenance, while no failures occur. This research has demonstrated that the polynomial chaos expansion remaining useful life prognostic approach can be used for optimal maintenance planning for aircraft engines using a Markov decision process, showing benefits in terms of costs, waste and unexpected failures.

Aircraft maintenance is critical to an airline's operations to ensure the reliability, availability, and safety of their assets. Recently, the approach of using component prognostics in aircraft maintenance has received increasing attention in academic- and industrial research. Predictive maintenance has demonstrated promising results in using sensor-based prognostics for maintenance decisions. In this paper, we propose a novel predictive maintenance framework that is capable of mapping the individual component degradation levels to an optimal maintenance decision. The independent component degradation levels are computed by a supervised learning model, called "Long Short-Term Memory Networks". Subsequently, the computed degradation levels are utilized in a multi-component maintenance decision framework, by using a model-free reinforcement learning technique named "Deep Q-Learning". The predictive maintenance framework aims to minimize a cost objective based on the type and frequency of a maintenance action. In addition, we analyzed several key performance indicators, such as the number of components used, the component utilization level, as well as the wasted component lifetime. The predictive maintenance framework was evaluated using NASA's turbofan degradation dataset. Ultimately, the results of the numerical experiments showed that the proposed predictive maintenance framework resulted in lower costs than when using a time-based and corrective maintenance policy and competitive costs compared to an ideal maintenance policy. The proposed predictive maintenance framework opens new directions for multi-component sensor-based maintenance decisions. The results found form the basis for application suggestions and future research directions in practice.

This research is relevant to any airline that wants to innovate its maintenance scheduling process bymaking use of data-driven techniques. Shifting towards a data-driven organization holds the potential benefits of increased scheduling efficiency and profits, while decreasing the time required to schedule maintenance on a daily basis. Optimizing the airline maintenance process will lead to lower aircraft downtime and less wasted life of aircraft components. This will result in a higher passenger satisfactory, a more sustainable way of operation and a higher profit for the airline. All of these aspects stress the importance of optimizing the airline maintenance scheduling process, both from a economic and societal point of view.

This Thesis proposes an optimisation model to minimize operational cost of a fleet of electric thin-haul aircraft under a minimum RPK (Revenue-Passenger Kilometer) constraint. A solution that minimises cost per RPK can be found by varying the minimum RPK level. The solutions describes how many aircraft are needed, as well as a schedule for each aircraft for a single day. We consider a set of airports where charging infrastructure is assumed to be present. All aircraft must start and end the day from the assigned hub airport. Next, we consider a discretised time space between a start time and end time with constant time steps. The problem is represented on a time-space network where each node uniquely defines a location (airport) and point in time, and arcs connect the nodes. An arc connecting two consecutive nodes at the same airport are ground arcs and represent waiting on the ground. An arc connecting two nodes at different airports are flight arcs. The cost, duration and energy consumption on flight arcs is determined in advance. A schedule is represented as a sequence of arcs. The number of passengers on a flight is limited by the demand. We developed a method to find a schedule that minimises costs while meeting a minimum RPK constraint. This method was then illustrated on a network with 5, 10, 15, 20 and 30 airports. The results show that our method is much faster than a traditional linear programming model. The obtained solution is a local minimum and is very close to the global optimum. We found that cost per RPK shows a sawtooth pattern, gradually decreasing as aircraft utilisation increases but spiking up when an additional aircraft is added to the fleet. Furthermore, the sawtooth pattern shows an increasing trend. The schedule shows a strong preference for connections with a longer distance and with high demand. The algorithm first fills up these connections, often using back-and-forth flights, until they are largely saturated. When there is little demand left on these connections, the algorithm starts adding flights on connections with a shorter distance or less demand.

To ensure continuous operations at airports, operational schedules need to be able to cope with early and late arriving/departing flights. To optimize such schedules for these events, flight delay predictions are necessary. Till now, flight delay has been studied mainly from a binary and regression standpoint. These type of predictions lack a certainty indication, which is needed by decision-makers to optimize a schedule. This research focuses on probabilistic predictions of flight delay on an individual flight level. Flight operation data from Amsterdam Schiphol airport and weather data from METAR is used to train four machine learning models that predict probabilistic flight delay distributions on a prediction horizon of one day. The first two developed models, a Dropout neural network and a Random Forest, predict empirical distributions. The other two models, a Mean Variance Estimator (MVE) and a Mixture Density Network (MDN) model, predict continuous distributions by predicting the parameters of an assumed distribution. The MVE predicts a single normal distribution, while the MDN model predicts a mixture of multiple normal distributions. Two interval-based performance metrics are suggested which assess the probabilistic predictions from different perspectives. It is concluded that arrival delay is best modelled with a MDN model with ten components, while departure delay is best modelled with only three components. Both models outperform a baseline statistical method on all considered interval widths. The MDN models outperform the other machine learning models by predicting more early arrivals and early departures correctly. The proposed models show that they can provide certainty indications of flight delay on an individual flight level, as opposed to binary and regression models. In the future, these certainty indications can assist airport operators with optimizing operational schedules, ensuring continuous operations and making ultimately air travel more pleasant.

In an effort to improve an airport operation optimization model, this research investigates the possibility of predicting probability distributions of flight delays with machine learning algorithms. The research is centered around Rotterdam The Hague Airport, a regional airport in the Netherlands. The first objective is to test how well machine learning classifiers can predict whether a flight will be delayed for a regional airport. This results in accuracies of around 70%, while taking precision and recall into account. The second objective is to predict the probability distributions of flight delays, for which three models are selected: a modified Random Forest Regressor, a Mixture Density Network and a Dropout Network. The main finding is that accurately predicting distinctive delay probability distributions for individual flights is possible. As a final objective, the predicted flight delay distributions are incorporated into an existing Flight-to-Gate Assignment Problem. It is found that this improves the robustness of the resulting schedules, although associated with a small reduction in their efficiency. The overall conclusion of this research is that machine learning-based prediction of flight delay distributions is possible, sufficiently accurate, and can improve at least one airport operation optimization problem. Further research will have to show whether this approach can be extended to other airports, other aviation optimization problems, or even optimization problems in other research areas.

During operation aircraft brakes degrade due to wear. This degradation can be continuously monitored using brake degradation sensors. Using this monitored degradation data the remaining useful life of the brakes can be estimated by means of a prognostic model based on a Gamma probability distribution. Using the remaining useful life estimations a maintenance schedule can be optimized for a fleet of aircraft each fitted with multiple brakes accounting for maintenance constraints. Considered maintenance constraints include the availability of the maintenance hangar and the aircraft flight schedules. Because the problem contains multi-unit systems opportunistic maintenance can be applied, specifically economic dependance between components is considered in the optimization. Using a rolling horizon approach a long-term maintenance schedule can be created and evaluated. This long-term schedule is simulated using Monte Carlo simulation to achieve robust results on which meaningful conclusions can be drawn. To compare the results of the condition-based maintenance strategy a time-based maintenance strategy with fixed replacement intervals is used. It has been shown that the application of a condition-based maintenance strategy which makes use of prognostics can significantly improve the scheduling performance.

In order to reduce aircraft emissions during on-ground operations, electric taxiing systems (ETS) have been intensively researched to take over or assist in part of the taxiing phase of a flight. One of these ETS is the TaxiBot, deployed by Smart Airport Systems (SAS). While a number of papers have been researching this vehicle routing problem (VRP) in order to minimise costs, fuel consumption or another metric, most research uses deterministic input data. However, sudden changes are inevitable, disrupting the resulting schedule. In this paper, we propose both a strategic model and a tactical model for airport surface movement, taking into account stochastic delays to the flight schedule. The subsequent tactical schedule can be produced with only little deviations from the strategic schedule. The tactical model can be generated in the order of seconds, making it useful for real-life traffic management decision support systems. We found that the tactical schedule does not worsen remarkably. Still 47.9% of the flights can be towed by a TaxiBot, which was 48.5% in the strategic schedule. Different case studies have been performed in order to determine the effect of e.g. the size of the flight schedule and the number of TaxiBots used, on the aircraft taxiing coverage and TaxiBot efficiency. With this, both busy and calm days, now and in the near future are assessed and the optimum number of TaxiBots necessary can be determined. An upper limit is reached at an asymptote starting from 34 TaxiBots, while a lower limit is dependent on a trade-off between flight coverage and spare TaxiBot capacity.

Electric taxiing (ET) is a novel concept that focuses on replacing engine-powered aircraft taxiing by taxiing using electrically powered towing vehicles, called ET vehicles. The main purpose of ET is reducing the impact of aviation on climate change while at the same time saving fuel costs. In this paper, we propose two models that can be used consecutively to analyse the operational implications of ET. Our goal is to determine the minimum number of ET vehicles required to perform all taxi procedures on a single day at a hub airport. First, we determine the optimal taxi routes for a set of aircraft towed by ET vehicles using a vehicle routing model. Then we find an optimal assignment of ET vehicles to these towing operations, taking into account time and energy constraints and scheduling the moments ET vehicles charge their batteries. We illustrate our models for a quiet and a busy day at Amsterdam Airport Schiphol. The models successfully give concrete guidelines on the required ET vehicle fleet size and infrastructure needed for the implementation of ET. The number of required ET vehicles can be decreased by tactically distributing battery charging over the entire day. Improved battery capacity and power can also effectively decrease ET vehicle fleet size.

Over the recent years a significant amount of research has been conducted to develop models which are able to estimate a components Remaining Useful Life (RUL) based on available sensor readings. In this research a deep learning (DL) model in combination with a similarity-based curve matching technique is used to estimate the RUL of a component. The data-driven RUL estimation scheme consist of an online and offline step. In the online step, a 1-dimensional Convolutional Auto-Encoder (1DConvNet-AE) is trained in an unsupervised way to convert multi-sensor readings, collected from historical run-to-failure instances, into a 1-dimensional reconstruction error vector. The set of generated 1-dimensional reconstruction error vectors is used to generate 1-dimensional Health Index (HI) curves which represent the various degradation paths of the run-to-failure instances. The HI curves based on historical run-to-failure instances are stored in the HI library. In the online step, multi-sensor readings of a test instance are converted into an online HI curve, representing the degradation path of an operational component. By using the similarity-based curve matching technique, the online generated HI curve is matched with the HI curves situated in the HI library. The RUL is estimated based on a weighted average of a set of matches which pass a set similarity threshold value. The proposed procedure is tested on an operational dataset. This research shows the existence of a relation between the increase of reconstruction error and health deterioration over time. The RUL estimation performance is considered in an operational evaluation procedure in which a similarity threshold value is included. This research shows a usable RUL can be estimated directly from raw multi-sensor input data by the proposed model.

As on-time performance is one of the main contributors to success in the world of commercial aviation, predictions on flight delays and cancellations can significantly improve operational efficiency and thus quality of service. Since flight delays and cancellations are occasional and infrequent events, operational on-time performance data is inherently imbalanced. This is especially the case for cancellations, as on average 1.6% of flights are cancelled, while about 33% of the flights is delayed. For this research, flight operational data is combined with weather data to predict flight delays and cancellations on prediction horizons of hours to months before the flight, by means of Neural Network and Random Forest machine learning algorithms. Since these algorithms naturally tend towards the usage of balanced data, the need exists to find a systematic approach to deal with the imbalance issues, in order to make accurate predictions. Hence, an imbalanced data approach is proposed, which analyses model performance with indicators such as precision and F1-score on varying data imbalance ratios. The imbalance ratios are obtained through the use of sampling techniques such as Synthetic Minority Oversampling and Random Undersampling. It is concluded that the highest precision is found without any sampling while for the highest F1-score sampling is essential. Additionally, the research confirms that severely imbalanced data, like the cancellation data, yields the worst performance when compared to medium imbalanced data, like the delay data.

This paper addresses the airline centred Arrival Sequencing and Scheduling problem aimed at the smart distribution of arrival delays, considering the explicit preferences from users. We consider the scenario in which actions are executed solely in the en-route phase with the available leeway present in the
current ATM system. The arrival process at the destination centre alongside equity rules such as ”First-Come, First-Served” remain untouched. A Mixed-Integer Linear Programming approach is presented in order to evaluate the fleet-wide impact of speed changes by individual aircraft in order to come to a global
(airline specific) optimum. The approach presented is evaluated using operational data in the form of a case study of a large European hub-style carrier. Case study results indicate the ability to decrease delay related cost by over 15% through the more efficient distribution of delay times between aircraft. Overall aircraft timeliness in the case study for both the controlled airline as well as competing airlines shows a slight improvement of several seconds of average delay per aircraft. In addition, a number of variations to the base model are presented, investigating a possible trade-off between model priorities.

The competition in the airline industry has rapidly increased during the last decades, especially with the entrance in the market of low-cost carriers. The high costs incurred in Maintenance, Repair and Overhaul (MRO) activities are generating a great interest in the improvement of maintenance operations as a way to stay competitive in the market. At the same time, the new generations of aircraft are increasingly being equipped with sensors that monitor the component's health condition. This stimulates the shift towards data-driven predictive aircraft maintenance, which is enabled by prognostics. This study proposes a model for maintenance scheduling of a fleet of aircraft based on component Remaining-Useful-Life prognostics and a limited stock of available spares. A discrete-time, rolling horizon approach is proposed, resulting in a sequence of scheduling time windows. For each time window, the goal is to find an optimal maintenance schedule. Moreover, the scheduling model considers three stages in decreasing order of maintenance priority, from critical aircraft leading to grounded condition, to predictive alerts, to non-critical failures. The results show that a cost-efficient maintenance schedule for a large fleet of aircraft is generated with an outstanding computational performance. Moreover, the aircraft operating costs are significantly reduced in the long-run, even when considering limited spares.

Airlines plan the trajectory of their flights in advance. However, this plan is not always followed since, during the actual flight, aircraft deviate either horizontally by rerouting, or vertically by choosing a different Flight Level. The issue arises when some airlines frequently deviate from their flight plans frequently, to the disadvantage of other airspace users. With this research we analyse aircraft en-route trajectory deviations for clusters of European airlines. We focus on two metrics: vertical flight level deviation, and horizontal deviation. We analyze these deviations for clusters of airlines operating in the European Civil Aviation Conference (ECAC) airspace in 2017. The airline clusters are obtained using unsupervised machine learning algorithms with operational and Complex Network features. The results show that major Low-Cost Carriers deviate on average, per flight, 34% more vertically than major Network Carriers, while Network Carriers have on average, more efficient horizontal trajectories. However, NCs show a larger fraction of flights with equally planned and realized efficiency, which points to less horizontal deviations overall. The findings reflect a general, network-wide horizontal inefficiency in trajectory planning, and that LCCs deviate horizontally a larger percentage of their flights. Moreover, they manage to use the available Flight Levels consistently to their advantage. This research could be used for future policy-making on a fair and equitable routes and slot allocation.

The flight planning process is an extensive and long process to direct and maintain a high level of operations within the airspace. As air traffic demand grows year after year, it's worthwhile to optimise the European air traffic system further. One way of optimising the system, is by creating optimal flight schedules that solve for demand-capacity imbalances. These schedules require an evaluation with respect to punctuality and fuel consumption in the tactical phase. For this article, an air traffic model has been developed in BlueSky to simulate air traffic while taking the variance of wind and delay into account. Subsequently, the performance of several optimised schedules will be assessed and compared with respect to wind, speed changes, punctuality and fuel consumption.

To safely and efficiently accommodate the air traffic growth in the coming years, a new concept of a time window (TW) flight has been recently proposed. A TW is a period of time during which the aircraft is required to arrive at a waypoint in the flight trajectory. It is characterised by an opening time, which represents the earliest possible time to cross the waypoint, and by a closing time, which represents the latest possible time to cross the same waypoint. Any delay from the TW in crossing the waypoint is as much a distortion of the trajectory as a change in horizontal position. The width of the TW indicates the amount of flexibility that airlines have during flight execution, as aircraft have to remain within the TW to avoid potential demand and capacity imbalances in sectors. The larger the TW width, the more flexibility a flight has.

TW flights allow for flexibility in operations, through the TW width, while simultaneously maintaining predictability. The flight schedules increase traffic predictability as they are developed in the strategic (months before the day of operation) and pre-tactical (seven to one day before operation) flight planning levels. Therefore, they facilitate the development of flight schedules that satisfy demand and capacity imbalances in the air traffic network early in the planning process and provide a common view of the air traffic environment on the day of operation.

This research tackled the challenge of evaluating the impact of the TW concept of operations on individual flight performance during the tactical phase (day of operation) of air traffic flow management. The flight time and fuel consumption needed to adhere to the TWs are under consideration. For the tactical assessment, a control model based on speed adjustments that allow aircraft to track contiguous TWs placed at critical waypoints along the trajectory was proposed.