Reserve cockpit crew allocation at major airlines is typically determined by scaling historical averages to planned network demand, without uncertainty quantification or systematic incorporation of operational drivers such as crew illness rates, seasonal absence patterns, and calendar effects. Within the broader workforce forecasting literature, uncertainty quantification is largely absent, and moreover, only two prior studies have addressed reserve cockpit crew demand forecasting directly, neither of which covers the European short-haul fleet or produces probabilistic forecasts across multiple planning horizons. This paper presents a probabilistic forecasting framework for cockpit crew reserve demand, evaluated on a cockpit division at KLM Royal Dutch Airlines across forecast horizons of 30, 90, 180, and 360 days. A direct multi-step forecasting strategy is applied, with dedicated models trained per horizon. A diverse set of classical, machine learning (Random Forest, LightGBM), and deep learning models (DeepAR, NHiTS, Temporal Fusion Transformers) is benchmarked on point accuracy and probabilistic calibration, with quantile regression and level-set forecasting (LSF) as uncertainty quantification methods. The experimental results show that the LightGBM model, optimized via quantile regression, achieves the lowest mean pinball loss at every horizon, leading the next-best model by up to 38.0%, but exhibits under-coverage relative to the empirical 90% target at forecasting horizons of 30, 90, and 180 days. Despite its strong probabilistic performance, quantile LightGBM’s point accuracy is modest: its mean absolute error (MAE) is only 4.5% lower than a naive mean baseline at its best horizon, underscoring how little exploitable signal a point forecast alone can extract from this series. The best-performing point model overall, Random Forest, improves on the mean baseline by 9.1% MAE, still a modest margin that highlights the intrinsic difficulty of the forecasting problem and motivates a probabilistic approach. Feature importance and ablation studies show that autoregressive lags and incidental illness rates dominate short-horizon predictions, while planned network demand becomes the primary driver at H = 360; weather alarm indicators are uninformative across all horizons. The framework is extended to planned reserves, voluntary reserves, and incidental illness as separate targets: voluntary reserve usage is up to roughly four times harder to forecast than planned usage on a normalized scale, while incidental illness is the most predictable sub-target. The results establish a proof of concept for data-driven probabilistic reserve planning on the European short-haul fleet and provide a basis for an explainable decision support tool for operational crew planning. The probabilistic forecasting framework reveals that reserve demand predictability differs across forecast horizons, quantile levels, and reserve types, information that a single point-forecast approach cannot provide. ... Read more

External (weather or ground operations) and internal (passenger connections and aircraft rotations) uncertainties affecting airline operators drive flight delays and generate disrupted operations as a consequence. These disruptions can be mitigated proactively if accurate predictions of events such as flight arrivals are available. This paper investigates the flight arrival delay prediction problem from an airline-centered perspective, considering both external factors for flight delays and internal factors for delay propagation. This paper proposes a network-aware Graph Neural Network trained on discrete-time snapshots of a hub airline network. The model considers aircraft rotations and passenger connections as independent delay propagation sources between flights, incorporating them into flight arrival delay prediction. The data set used for the case study concerns SWISS International Airlines and Edelweiss flights in 2025, with a Zurich-hub emphasis. Two Gradient Boosted Decision Trees (GBDT) baselines (with and without explicit graph features) are designed to assess the performance of the proposed Discrete-Time Heterogeneous Graph Neural Network. The main findings of the work demonstrate the benefits of dynamic network-aware models, offering more than 1 minute of lower mean absolute error than proposed baselines. Furthermore, operationally relevant features absent from previous work, including weather conditions and airport congestion, are shown to improve predictions across both model families. Finally, it is shown that the model reports decreasing performance with respect to the time horizon, but implicitly learns operational constraints such as the night curfew, where prediction accuracy recovers relative to other periods in the second half of the day. The presented model can be used as an input for more accurate early operational decisions, such as early re-booking or identification of critical connecting passengers. ... Read more

Continuous Descent Operations (CDOs) can reduce fuel consumption and CO2 emissions. However, their implementation in constrained airspace is often limited by operational procedures and altitude restrictions. Previous studies have evaluated CDO performance under idealised conditions, resulting in insufficient quantification of the effects of real operational constraints. This study investigates how operational constraints on arrival routes influence aircraft vertical descent profiles and the resulting fuel consumption. A framework is introduced that is capable of quantifying fuel consumption across different descent trajectories. The simulation-based framework enables the analysis of incremental modifications to operational restrictions, including changes to level-off altitude and duration. The methodology is applied to arrival traffic at Schiphol Airport, using a dataset of over 8,000 recorded arrivals from the Aircraft Condition Monitoring System (ACMS) and one month of Eurocontrol Demand Data Repository (DDR) traffic data. Level-off segments between Top of Descent (TOD) and the Initial Approach Fix (IAF) are identified and linked to waypoint-based restrictions specified in Letters of Agreement (LoAs) and the Route Availability Document (RAD). The BlueSky air traffic simulator and the Base of Aircraft Data (BADA) 3.16 performance model are used to quantify the fuel impact of modified descent scenarios. The results show a clear relationship between level-off duration, altitude constraints, and fuel consumption. Higher level-off altitudes and shorter durations consistently reduce fuel burn. Regression analysis of recorded flights confirms these trends. A Key Performance Indicator (KPI) based route assessment identifies arrival routes with the greatest potential to reduce fuel consumption, while taking into account operational complexity. The findings show that measurable fuel savings can be achieved through targeted adjustments in airspace restrictions without requiring a complete redesign of the airspace.
... Read more

The anticipated growth in air traffic is expected to increase demand on air traffic control capacity beyond its current limits, motivating the development of autonomous conflict detection and resolution (CD\&R) methods for self-separating airspace. While both state-based and intent-based conflict resolution methods have been studied extensively, a direct comparison between the two approaches under identical conditions is lacking. This paper addresses that gap by comparing the intent-based A3 CD\&R model to the state-based Modified Voltage Potential (MVP) resolution method. Performance is evaluated through verification scenarios and random traffic simulations at three density levels, using simulations in the BlueSky ATM simulator. Results show that the A3 method achieves substantially fewer conflicts and losses of separation than MVP across all density levels, with a markedly lower domino effect parameter indicating better airspace stability. However, when losses of separation do occur under the A3 model, the intrusions are larger than those observed with MVP. Regarding efficiency, MVP incurs large flight time increase due to speed-based resolutions, whereas the A3 model resolutions result in minimal flight time increases at the cost of larger route deviations. These findings demonstrate that intent information provides meaningful advantages for conflict resolution performance. ... Read more

In the approach phase of a flight, aircraft transitions from the en-route phase to the final approach. The transit time between these phases is highly uncertain and frequently subject to delays. Efficient and reliable prediction of this time is essential for airline fuel planning and flight scheduling, yet current practice still relies largely on fixed deterministic buffers. Most existing work on arrival delay prediction focuses on deterministic models and aggregate indicators (e.g the difference between planned and actual arrival times), often at forecast horizons during the airborne phase. This paper develops and validates an explainable probabilistic forecasting model for flight duration within the Amsterdam Schiphol (AMS) Flight Information Region (FIR), with a forecast moment in the pre-departure phase. The primary objective is to forecast the duration within the AMS FIR using information available at planning while providing interpretable contributors of delay that can be clearly communicated to flight dispatchers and pilots. The study uses a historical operational dataset of roughly 280,000 inbound flights, combining airline planning data, AMS traffic data, and METAR/TAF weather reports. On the held-out test set, the model achieves a reduced MAE of 33% and a reduced RMSE of 26% relative to the current operational baseline, while capturing about one third of the variance in FIR duration (R2= 0.33). Error analysis shows that typical delay contributors are captured logically and have intuitive effects on the predictions. It is found that the largest under-predictions are mainly driven by weather forecast errors and unexplained tactical ATC interventions. The results indicate that quantile-based transit time forecasts can provide airlines with a more risk-aware basis for fuel and schedule planning than fixed deterministic buffers. However, the relatively low R2 shows that a substantial share of the variation in FIR duration remains unexplained, largely associated with tactical ATC interventions that occur under otherwise acceptable weather and capacity conditions. ... Read more

With a rising demand for air travel, Airline Disruption Management (ADM) ensures successful schedule recovery in the event of disruptions. The Aircraft Recovery Problem or ARP, part of ADM, solely focusses on aircraft. Previous research has concentrated on exact optimisation, as well as simple-, meta-, or hybrid-heuristic solution methods. However, in order to prevent significant delays, resolution decisions must be made fast. This need combined with the rise of deep learning, has led to the emergence of deep reinforcement learning (DRL) as a viable solution strategy. Nevertheless, the performance of DRL remained often limited to specific state space formulations and reward designs.
In order to close this gap, the primary objective of this work is to further optimise a reinforcement learning (RL) formulation for the aircraft recovery problem (ARP) while minimising disruption effects. It investigates and compares two models with alternate state space formulations. First, we test a single, aircraft-centric and continuous design. Second, we presents a dual, sparse, flight-centric, and primarily binary formulation. Each model compares computational efficiency, action distribution, and conflict resolution effectiveness across three DRL environments; proactive, reactive, and myopic, subject to different levels of stochastic state information. It was found that the state space formulation significantly influences computation time, which is a prominent issue faced by big action- and state space sizes. Furthermore, it is shown that proactive environments result in better conflict resolution.
However, significant challenges of the model were revealed by the unexpected negative learning trend. This counterintuitive result was further underlined by the notably higher performance during exploration than during exploitation, indicating the DRL agent’s inability to learn an optimal policy. Finally, sensitivity analyses of the reward and a hyperparameter underlined the high susceptibility of RL to minor parameter tweaks, stressing the challenging implementation of DRL models for real-life applications. ... Read more

More robust aircraft maintenance planning can be achieved by accounting for workforce availability uncertainty, thereby reducing the risk of understaffing and costly Aircraft on Ground (AOG) situations. However, traditional forecasting methods in this domain typically rely on point estimates and fail to account for the long-term unpredictability of manpower due to sickness, vacation, or educational leave. This study addresses this gap by developing a probabilistic time series forecasting framework to model daily uncertainty in workforce availability by skill group for horizons up to one year. Drawing on a novel dataset of workforce absence and schedules provided by KLM Engineering \& Maintenance, the study benchmarks a diverse set of architectures, including classical statistical baselines, tree-based methods (LightGBM), and Deep Learning models (DeepAR, NLinear, TSMixer), on their ability to generate well-calibrated predictive distributions. The experimental results demonstrate that the LightGBM architecture, optimised via quantile regression, consistently outperforms complex deep learning and statistical alternatives. Furthermore, it improves point accuracy by approximately 10\% over the current internal forecasting standard at short horizons and 5\% at long horizons, while maintaining experimental calibration above 91\% for its forecasted 95\% confidence intervals. Feature ablation studies reveal distinct temporal dynamics: while long-term, one-year-ahead forecasts are predominantly driven by deterministic calendar and holiday features, short-term accuracy benefits from recent historical absence patterns. A critical operational validation further highlights that while incorporating planned absences improves uncertainty estimates during training, inclusion leads to overconfident predictions due to the addition of significant feature uncertainty. Instead, this study recommends a robust, feature-based probabilistic approach that leverages SHAP values to provide planners with actionable, transparent insights into availability risk.
... Read more

Accurate prediction of aircraft turnaround time (TAT) is essential for mitigating reactionary delay, yet present methods remain constrained. Existent work uses discrete event simulations to predict individual ground activities but accumulate error and uncertainty, and in turn, other data-driven studies still provide a single static point estimate that is not updated with the latest operational information. It is therefore difficult to validate the use of these models in real operational environments. Although rolling, continuously updated forecasts have recently been explored for departure delay prediction, no study has yet extended such dynamic modelling to the turnaround itself, leaving a critical gap in operational decision support, as departure delay may include additional delay sources external to the turnaround process. This study develops and operationally tests a probabilistic machine learning framework that continuously updates full predictive distributions of TAT for a European hub carrier. Real time airline, meteorological and air‑traffic feeds are merged into gradient boosting tree ensembles trained with quantile regression. Evaluation on past turnarounds yields a median absolute error of 10 minutes immediately after the preceding take‑off, falling to 8 minutes at on‑block. Results show that the uncertainty of the prediction reduces by a quarter as updated operational data like delays or air-traffic control slots come in. These findings show that uncertainty in TAT can be quantified accurately and in near real-time using data streams already present in airline operations, enabling controllers and optimisation engines to target mitigation measures proportionately and thereby reducing cascading delay, cost, and emissions. ... Read more

Aircraft maintenance planning plays a large role in ensuring operational efficiency and safety while minimising costs. Hangar maintenance scheduling can be trivial due to various uncertainties, such as non-routine tasks, resource availability, and unforeseen delays. Deterministic methods might struggle to account for these complexities and do not scale well with large, heterogeneous fleets, causing frequent and costly adjustments to the schedule. Previous research has focused on incorporating different uncertainties using robust scheduling methods. This research aims to develop and assess a stochastic and scalable aircraft hangar maintenance planning model that can provide insight in the robustness of the planning, next to incorporated uncertainties, to reduce the need for frequent planning revisions. The proposed method creates a schedule using a Genetic Algorithm (GA) that minimises maintenance costs and interval losses while adhering to operational constraints. After that, a Monte Carlo simulation is applied to assess the feasibility of the schedule under randomly generated check duration scenarios. Critical checks that can cause grounding of aircraft due to exceeded due dates are modified in a feedback loop to improve the robustness of the schedule. The maintenance optimisation is tested in a case study, provided by a European airline and discusses the trade-off between maintenance cost, interval loss, run time, and feasibility in hangar maintenance planning under uncertainty. The schedule is compared to a Mixed-Integer Linear Programming (MILP) benchmark model. Results show that the MILP outperforms the MILP in terms of cost and run time, but the GA might be useful in more complex scenarios. The simulations give insight in the robustness of the planning and show that delay propagation and grounding probabilities can be decreased by adjusting critical checks during re-optimisation. The overall grounding probability can go down by 9 to 40%, with 5 to 10% of checks fixed in time, respectively. This can lead to a more robust schedule, minimising revisions. An airline can use this framework as a decision-support tool to create variations on the planning and assess the impact of its decisions on the robustness of the schedule. ... Read more

Effective scheduling is challenging in Component Maintenance, Repair, and Overhaul (CMRO) operations due to the complexity of dynamically allocating resources across multiple jobs with varying priorities and technical constraints. Current industry practices typically rely on static, manual scheduling, resulting in suboptimal resource allocation and insufficient adaptability to operational disruptions. Most existing studies approach specific job shop problems by incorporating individual features, such as job prioritization or resource constraints, without considering the combined operational complexities of CMRO shops. Therefore, this research presents a scheduling model using the Flexible Job Shop Scheduling Problem (FJSSP), tailored to dynamic CMRO environments. The model uses Mixed Integer Linear Programming (MILP) to simultaneously schedule technicians and machines while accounting for skill requirements, resource constraints, and job prioritization. The approach balances multiple objectives, including tardiness and earliness, to enhance shop performance metrics such as Turnaround Time (TAT) and On Time Delivery (OTD) rates. Outcomes from a case study applied to real-world data from CMRO shops demonstrate significant operational improvements, achieving a reduction in TAT of up to 34% and an improvement in OTD by approximately 23% relative to historical shop performance. Furthermore, the model incorporates schedule robustness measures, minimizing deviations from planned schedules, despite operational uncertainties. Additionally, comparative analysis with a traditional heuristic dispatching rule model confirms the superior performance of the proposed optimization framework. This framework can be broadly applied to improve scheduling efficiency and stability in CMRO shops and similar workshop environments.
... Read more

This thesis presents a novel framework for solving the Multi-Skill Resource-Constrained Multi-Modal Project Scheduling Problem with maximum time lags, addressing the challenges of scalability, deadline adherence, and uncertainty in job durations. The research is conducted through a case study with the maintenance department of a large European airline, using real-life maintenance scheduling data. To improve scalability, the framework integrates batching techniques that segment the scheduling horizon and a priority-rule-based heuristic to hot start the solver, significantly reducing computational runtimes for large problem instances. A range of objective functions, including single and multi-objective formulations, are explored to evaluate their impact on scheduling performance. The results demonstrate
that multi-objective formulations provide the best balance between throughput and deadline adherence and consistently outperform a priority-based heuristic. A clear trade-off is observed between optimizing for maximum tardiness and average tardiness, where minimizing maximum tardiness improves deadline adherence at the cost of lower throughput, while minimizing average tardiness has a more consistent throughput but allows slightly more deadline misses. To address job duration uncertainty, adaptive buffering strategies based on historical job performance are introduced and shown to outperform static buffers by tailoring slack times to individual job characteristics. In the examined case study, the combination of an adaptive buffering strategy with a multi-objective function combining makespan and
average weighted tardiness offers the most effective trade-off between robustness and efficiency. Overall, the framework proves to be scalable, adaptable, and well-suited to real-world scheduling environments with high variability and complex constraints.
... Read more

The goal of this project is to develop a firefighting aircraft capable of meeting the demands of a market that lacks purpose-built aircraft and faces an ever-increasing threat. The W-132 is capable of making precise, targeted drops thanks to its high manoeuvrability, whilst maintaining a very high cruise speed and payload relative to its competitors. This report includes logistics, operations, sustainability, design process, feasibility, risk and cost analyses, providing a thorough overview of the team’s achievements. ... Read more

The Europe-Asia-Oceania air route is experiencing rapid growth, traditionally served by direct legacy flights but increasingly dominated by hub-based carriers. These airlines leverage large, single-hub models to capture transfer traffic. However, limited research exists on how to efficiently design and operate multi-hub networks for international connectivity. Existing models either oversimplify hub capacities or focus solely on fleet planning and re-timing, lacking an integrated, schedule-based approach. This research develops an integrated decision support model that combines a capacitated multi-allocation p-hub location problem with airline schedule design under operational constraints and competitive dynamics. A multi-step iterative method connects hub assignment and scheduling using a genetic algorithm to ensure feasibility, connectivity, and profitability. The findings reveal that adding hubs initially boosts network efficiency and profitability, though marginal benefits diminish after a certain number of hubs. The integrated model significantly narrows the gap between theoretical and actualized profitability, showing a 7.6\% increase in daily profit for the scheduling model over iterations. This research offers a crucial decision-support tool for long-term airline network planning, particularly in rapidly expanding aviation markets. Its ability to jointly optimize hub locations and flight schedules under operational constraints provides substantial utility for diverse airline network planning scenarios, leading to more viable and profitable network configurations. ... Read more

Several studies have demonstrated that an integrated approach to the airline schedule recovery problem, optimising multiple facets simultaneously rather than using traditional sequential methods, yields improved solution quality; however, at the cost of model simplification, with most studies focusing solely on cost minimisation. This paper introduces a multi-objective Benders decomposition approach to a detailed integrated aircraft and crew recovery model that considers a heterogeneous fleet, individual crew members, and basic passenger considerations, thereby addressing both model simplifications and lack of multi-objective considerations. A lexicographic multi-objective optimisation scheme is integrated into Benders decomposition, allowing for the optimisation of multiple distinct objectives: minimising cancellations, minimising recovery costs, maximising on-time performance, and minimising changes made. Computational tests were conducted using real schedule data provided by Transavia, with results showing high-quality solutions within the model’s defined assumptions. However, Benders decomposition alone was insufficient to be applicable within live disruption management, with performance times ranging from 5 to 30 minutes depending on the scenario’s complexity. The model’s design and ability to consider crew on an individual level show strong promise in solution quality and applicability to the industry, as crew are often considered at both the pairing and individual levels. This provides a strong foundation for practical disruption management support tools if deficiencies in time performance can be addressed. ... Read more

The deployment and subsequent development of an Advanced Air Mobility (AAM) transportation system is expected to take an incredible amount of resources in terms of planning, time and capital. Due to the system not yet being operational anywhere, and consequently, the lack of clear operational boundaries set, researchers and developers are left with an enormous design space. Currently, parallel independent developments are taking place in all aspects of the system, and different perspectives have enabled a wide range of concepts to be formulated in each. Whereas independent assessments provide crucial insights into the individual components of the system themselves, they fail to capture the inherent interdependencies. They also do not capture the growth of the aircraft fleet in correlation with the growth of the vertiport network. This gives rise to the need for a framework which is capable of establishing a preliminary vertiport network to allow for the study of the aircraft, fleet and total system performance in a coherent manner, and the scalable correlated evolution thereof. Consequently, this study aims to develop a unified framework for the formation of a scalable vertiport allocation plan in conjunction with system-performance-based heterogeneous fleet sizing. The scalable vertiport allocator employs a distance-based agglomerative clustering algorithm to determine the clusters in the ultimate vertiport network and the k-means clustering algorithm to determine the preliminary location of the vertiport per cluster. This is followed by a commute-distance based vertiport elimination procedure to establish each vertiport network to be assessed. The optimal fleet at each stage of network growth is established through a parameter sweep conducted across the fleet size and composition. The system-based performance metrics of the combinations of vertiport network and fleet are then assessed using an on-demand agent-based simulation. The framework is applied to New York (NY) state and models of existing multi-rotor (MR) and tilt-rotor (TR) aircraft are utilized as test case. The test case results show the applicability of the framework in the establishment of a scalable preliminary vertiport allocation plan to maximize system commute distance, and the correlated growth of the optimal fleet based on the maximum system performance. ... Read more

Disruptive events pose a significant challenge to airlines’ everyday operations due to the highly optimized nature of their schedules. Unforeseen events force airlines to rapidly reschedule and adjust their operations. Current disruption management methods rely mostly on reactive and static models that fail to capture the dynamic and probabilistic nature of airline recovery. This study presents a model-based reinforcement Reinforcement Learning (RL) method for aircraft recovery under disruption uncertainty that anticipates future potential disruptions. The Aircraft Recovery Problem (ARP) is formulated as a Markov Decision Process (MDP) and a framework is proposed in which an Approximate Dynamic Programming (ADP) algorithm that relies on Value Function Approximation (VFA) determines optimal recovery actions considering the immediate and future impact of each action. The uncertain disruptions are modelled as aircraft unavailabilities with a fixed probability of realizing. The aim of the model is to keep flight delays and cancellations at a minimum while exploiting stochastic information on potential aircraft unavailabilities. The model is tested on multiple scenarios with different objectives and levels of disruptions and is benchmarked against an exact optimization algorithm. Results indicate that a proactive approach outperforms reactive models, particularly in high-disruption scenarios with high aircraft utilization. The comparison with the exact benchmark shows that the RL method can achieve sub-optimal solutions with considerably less corrective actions. This framework offers a decision support tool that allows airline operators to find more resilient solutions in uncertain environments by incorporating probabilistic predictions on disruptions in the decision-making process ... Read more

Engine shop visit (ESV) scheduling is a critical component of airline maintenance planning, directly impacting operational continuity, cost management, and long-term fleet value. Despite its importance, existing approaches often overlook fleet-level considerations, such as additional lease engines and spare engine management. Whilst maintenance planning has been widely studied, the specific dynamics associated with engine shop visit planning remain relatively unexplored. This paper presents a Mixed-Integer Linear Programming (MILP) framework to solve the engine maintenance problem as an adaptation of the Resource Constrained Project Scheduling Problem (RCPSP). The classical formulation has been adapted significantly, as time precedence constraints have been omitted, and extensions have been introduced to incorporate engine health metrics and component-level scheduling. Furthermore, the model has been extended to allow for additional lease engine activation and to manage the number of available spare engines. The framework is applied to the operational contexts of a major European airline operating wide-body aircraft in a mixed global network, integrating airline-specific constraints and assumptions. Through sensitivity analyses on key parameters and several use-case scenarios, including an Unexpected Engine Removal (UER), the model successfully generated feasible shop visit plans under varying conditions whilst providing valuable insights to decision-makers. The results highlight the benefits of integrated planning and support operational and strategic engine fleet management. ... Read more

Managing spare parts for Aircraft Maintenance, Repair, and Overhaul (MRO) is challenging because there is a significant gap between long-term maintenance schedules and daily procurement decisions. While existing research often addresses demand forecasting and inventory control in isolation using abstract assumptions, a new framework is presented to bridge this gap by directly connecting day-today procurement decisions with the fixed, fleet-wide maintenance schedule. A task-based approach enhances traditional planning, which often relies on aggregate forecasts and can miss the specific needs of individual checks. The result is a transparent, cost-based model built for operational utility, where every decision accounts for probabilistic predictions and remains auditable. This traceability is crucial in an environment where complete historical data is often unavailable. The framework consists of two modular stages. First, a demand forecasting methodology converts raw maintenance tasks into a usable, time-phased, and probabilistic demand signal. To accomplish this, maintenance tasks are systematically grouped based on their technical attributes. The outcome is a repeatable method to describe the demand potential of each scheduled task. Second, a daily procurement optimization model was created to act on this detailed forecast. The algorithm replicates a planner’s decision-making process by explicitly comparing the expected future costs of buying, waiting, or selling surplus stock. To mirror operational reality, the model utilizes regular orders with uncertain lead times, reactive express orders, and pre-procurement. Every decision becomes a justifiable trade-off regarding the cost of purchasing and holding inventory, and the high financial penalty of a stockout. Finally, the model was validated against two benchmarks: a fully conservative (100% service) strategy and a standard periodic-review policy. The proposed model reduced total net costs for both benchmarks, achieving savings of 17.5% and 9.2% respectively. The analysis shows this advantage stems from the strategic acceptance of controlled risks when doing so leads to a lower expected total cost. Further sensitivity analysis revealed that the cost-driven logic is robust even under poor forecasts, as it automatically compensates to maintain a safe inventory level. The analysis also identifies key non-linear trade-offs, finding that total net cost is minimized at a moderate level of caution regarding stockout penalties, rather than at the extremes of underor over-estimation. The framework ultimately provides a practical and transparent decision support tool, demonstrating that a task-specific, dynamic, and cost-based approach is more effective and resilient than traditional, static planning rules. ... Read more

Reliable autonomous recovery of Unmanned Aerial Vehicles (UAVs) on moving maritime platforms remains a critical challenge, primarily due to complex, stochastic deck motion, particularly vertical heave, and unpredictable environmental disturbances. Existing Reinforcement Learning (RL) approaches often simplify this environment, limiting their real-world applicability. This thesis investigates the robustness trade-offs of RL-based guidance controllers under realistic, high-dynamicity maritime conditions. We benchmarked a classical Proportional-IntegralDerivative (PID) controller against two RL architectures trained using Soft Actor-Critic (SAC) in a high-fidelity PyBullet simulation: a Full RL 3D controller and a novel Hybrid RL 1D controller, which strategically applies RL only to the critical, stochastic vertical (heave) axis. The results demonstrate that the Hybrid RL 1D architecture (86.6% success rate) achieved superior overall robustness and efficiency. Notably, the RL controllers dramatically reduced average landing time (RL_1D: 3.31 s vs. Baseline: 11.51 s), though the classical PID baseline maintained higher horizontal precision (Err𝑋𝑌 of 0.17 ± 0.17 m ). The Hybrid RL 1D maintained a superior success rate up to 89% in high sea states (SS7) and exhibited greater resilience to sensor noise. However, a critical limitation was identified: both RL-based policies experienced a pronounced performance collapse under strong, untrained wind disturbances, a regime where the non-adaptive classical PID baseline proved unexpectedly stable. These findings confirm the benefits of hybrid control for maximizing robustness and highlight that the system’s ability to handle wind disturbance rejection remains a significant, unresolved shortcoming for current RL guidance systems. ... Read more

Reducing uncertainty in air traffic flow management is crucial for maintaining safety and efficiency in modern aviation. Additionally, forecasting Actual Take-Off Times (ATOT) for flights across Europe is particularly challenging due to the diverse flight-specific variables and operational conditions. This study focuses on enhancing ATOT prediction for flights arriving at Amsterdam Schiphol Airport from European out-stations by leveraging machine learning techniques, specifically a Long Short-Term Memory (LSTM) neural network, augmented with a Multihead Attention mechanism. A model capable of capturing complex temporal dependencies and operational factors influencing the ATOT is developed utilizing data from Electronic Flight Data (EFD) messages, weather reports and a EUROCONTROL dataset. The model’s performance is evaluated against traditional ensemble methods and the current Decision Support Tool (DST) system used by Luchtverkeersleiding Nederland (LVNL). Results indicate that the LSTM model outperforms existing models including a reproduction of the DST, achieving a Mean Absolute Error (MAE) of 12.05 minutes at a forecast horizon of 4 hours, demonstrating significant improvements. This assessment underscores the importance of factors such as the knock-on effect in delay prediction and suggests that integrating advanced machine learning models can significantly enhance demand forecasting, leading to more efficient air traffic management and reduced delays at Schiphol Airport. ... Read more