Background

Cargo handling is a 24/7 industry, with ground services, air transport, oceanic shipping, and more all happening around the clock. While it is common for cargo facilities and ports to run 24/7 or close at night, it is less common for part of a cargo facility to temporarily close at night while operations in other areas remain operational. This setup has not been significantly researched and therefore remains unoptimised. Such a facility setup exists at KLM Cargo
facilities at Amsterdam Airport Schiphol, where one of its three cargo terminals temporarily closes while the other two remain open throughout the night, causing operational inefficiencies.

System analysis

Terminal 1 (T1) at KLM Cargo’s primary facilities handles the airline’s express shipments (XPS), but is subject to nightly closures, requiring express products delivered at night to temporarily be stored at Terminal 3 (T3), which primarily handles general cargo exports, until T1 reopens in the morning. This leads to several inefficiencies and problems, including XPS packages frequently being delayed and missing early morning flights due to not being transferred and sorted in time after T1 reopens, as well as XPS packages impacting the general cargo operations at T3.

In this project, presented by the Business Process and Improvement (BPI) department at KLM Cargo, the XPS package processes were mapped, analysed, and redesigned in order to optimise the cargo delivery process during the night and eliminate operational inefficiencies and hindrances within the current setup. This was achieved through a variety of methods, including literature research, data analysis, in-person observations, interviews, and discrete-event simulation modelling.

Future requirements analysis

Many stakeholders are interested in seeing the current process revitalised and replaced with a new process that ensures XPS packages are sorted sooner, do not get lost, are prioritised better, and reduce the impact on T3 operations. It is also important that the project satisfies all safety and security requirements, fits within the physical spaces of the terminal floor plan, and falls within the monetary limits of the company.

Design solution

The newly proposed design involves installing automated drop-off machines in the walls of T1 that connect to scanning technology and subsequently feed directly into the sorting machine within T1. This allows XPS packages that are cleared and ready to be shipped to directly enter the sorting machine of T1 instead of waiting at T3 until the morning. Additionally, holding bins allow some XPS packages to be set aside should the system detect potential safety and security risks that require additional human checks. The primary goal of the new solution is to reduce the average amount of time XPS packages take from the moment they arrive on the premises to the moment they are inserted into the sorting machine, while also removing as many XPS packages from T3 at night as possible.

Results

Discrete-event simulation (DES) models were used to replicate both the current and proposed systems. The model for the new process tested various percentage combinations of XPS packages that were either accepted directly or stored in a temporary holding bin awaiting additional checks in the morning. Data provided by KLM Cargo included values for the interarrival times of trucks carrying XPS shipments at night, as well as the number of packages arriving, allowing for accurate data-based representations of the processes. The simulations produced several outputs for both the current and proposed processes, including average XPS flow times, average times at which the first XPS package entered the sorter, and the average number of packages entering the sorter before T1 reopened. These results were compared between the two systems to support conclusions and recommendations.

Discussion and recommendations

The results obtained from the DES simulations indicate that an automated drop-off system can significantly improve the operations and execution of the night-time cargo delivery process for express shipments at T1. The results show that having at least 15% of night-time XPS arrivals directly accepted into the sorter during the night is sufficient to surpass the current system in performance. Additionally, the results support the notion that the more XPS packages are directly inserted into the sorter, the less impact and fewer problems arise in T3 operations during the night. It is therefore strongly recommended that KLM Cargo continues development and research aimed at turning this drop-off proposal into reality.

Global supply chains increasingly experience disruptions due to climate events, geopolitical changes, and cyber threats. Artificial intelligence (AI) presents opportunities to enhance supply chain resilience; however, these systems may introduce additional risks if not implemented with adequate safeguards. This study examines strategies for designing and governing AI systems to strengthen supply chain resilience while minimizing unnecessary complexity. A mixed-methods approach is employed, combining interviews with AI and supply chain experts and two focus groups. The research identifies three agent design archetypes that clarify the practical contributions of AI to supply chain resilience: Diagnostic and Monitoring Agents, which provide early warnings for slow-moving risks; Response and Coordination Agents, which accelerate disruption recovery by synthesizing fragmented data and take limited actions; and Interface and Learning Agents, which disseminate tacit knowledge and reinforce feedback loops. Furthermore, a two-dimensional evaluation framework is introduced. The first dimension (R-Dimension) as- sesses whether a proposed AI concept credibly enhances supply chain resilience, while the second dimension (D-Dimension) determines the appropriate level of autonomy and implementation feasibility. The Dimensions Interaction Matrix maps resilience value against governability across six decision zones, guiding practitioners from concept rejection to bounded autonomy. The framework also maps Generative AI (GenAI) and Agentic AI to specific Levels of Automation (LoA), enabling practitioners to justify whether a deterministic approach, GenAI, or Agentic AI is suitable for each supply chain decision point. The findings indicate that effective deployment of Agentic AI requires investment in disruption readiness through synthetic scenario testing and organizational preparedness with defined escalation protocols and human override authority. This research provides practical frameworks for practitioners and identifies areas for future validation with real-world data.

The Dutch healthcare system currently operates under intensifying fiscal and workforce pressures, transforming nurse scheduling from a simple operational task into a complex strategic conflict involving divergent stakeholder priorities. Addressing this "wicked problem," this study addresses nurse scheduling as a strategic decision challenge, introducing a novel methodological framework that bridges Operations Research and Game Theory. Unlike traditional optimization models, this research introduces the "Nurse Scheduling Game" (NSG), formulating the problem as an Integer Programming Game (IPG) to capture the behaviors and interactions of stakeholders explicitly.

The interaction is analyzed through two distinct equilibrium concepts representing different governance structures. First, the uncoordinated state is modeled as a simultaneous Nash Equilibrium, formulated as a Generalized Nash Equilibrium Problem (GNEP). Second, the potential for strategic improvement is explored through the hierarchical Stackelberg Equilibrium, formulated as a Bilevel Integer Problem (BIP). In the latter, the Hospital Manager (Leader) explicitly anticipates the Nurses' (Followers) reactions. To solve this computationally intractable bilevel problem, the study implements a novel Monte Carlo Multilevel Optimization (MCMO) framework.

Applied to a representative case study of a mid-sized Dutch hospital, the computational results quantify the significant costs associated with uncoordinated planning. Under Nash dynamics, the system converges to a state of "defensive buffering," resulting in outcomes approximately twice as expensive as the coordinated alternative. Conversely, the Stackelberg Equilibrium demonstrates the value of strategic anticipation. By transitioning from volume-based to precision-based allocation, the hierarchical model achieved a 51.0% reduction in total system costs and an 11.8% reduction in patient waiting times compared to the Nash baseline.

These findings translate into actionable policy implications, suggesting that the solution to budget overruns lies in shifting from reactive to anticipatory governance. The study supports the implementation of Algorithmic Workforce Management systems that couple budget setting with schedule design. Ultimately, this research offers a unified game-theoretic optimization framework that reconciles financial constraints with workforce autonomy, providing a viable pathway toward sustainability for the Dutch healthcare system.

Maritime ports are critical nodes in global logistics network, but are vulnerable to disruptions, especially from extreme weather events. Efficient recovery by optimized repair sequences can minimize vessel waiting times and operational losses. This study proposes a Deep
Q-Network framework integrated with Agent-Based modelling to optimize repair strategies
for disrupted waterway networks in ports. A case study of the Port of Rotterdam demonstrates
the model structure, implementation and comparison with a greedy heuristic benchmark. The
DQN was trained on simulated disruption scenarios and evaluated on unseen test cases, showing superior performance in high priority disruptions. In low priority scenarios with minimal
network impact, the greedy approach performed comparably or better. Results highlight the
potential of reinforcement learning to adapt repair strategies dynamically based on real-time
conditions, offering a decision support tool for port authorities. The findings suggest that, with
further integration of operational constraints and real world data, such methods could enhance
port resilience and reduce recovery times following major disruptions.

Urban congestion is an increasing challenge for accessibility, travel times and livability. Although on-demand water cabs offer a flexible and sustainable alternative, the design of optimal mooring locations has been underexplored. To address this problem, this article presents a new method for the strategic placement of mooring locations. The proposed method involves identifying potential mooring locations and applying a customised Adaptive Large Neighbourhood Search (ALNS) to manage computational complexity. The model minimises weighted travel time and cost indicators for all origin destination pairs by selecting between public transport and water cab for each trip. The method was evaluated in a case study of Rotterdam. The findings indicate that the proposed ALNS algorithm outperforms both Large Neighbourhood Search (LNS) and Mixed-Integer Linear Programming (MILP) in computation time while simultaneously generating high quality solutions. Moreover, the model is shown to be capable of identifying an efficient network, resulting in greater coverage, shorter travel times and improved accessibility. These results demonstrate that the method constitutes a powerful tool for policymakers in designing robust, efficient and future proof water cab networks.

This thesis examines the integration of carbon-reducing technologies and distribution decisions within the cement industry's supply chain, with a focus on the impact of these decisions on emerging carbon pricing mechanisms, particularly the European Union's Carbon Border Adjustment Mechanism (CBAM). As the cement industry is a major contributor to global carbon emissions, there is significant pressure to reduce its environmental footprint. Despite the potential of carbon capture technology, its adoption within the cement industry has been slow and remains limited. Moreover, decision‐support tools to guide these decarbonisation decisions remain scarce. This thesis develops and validates an integrated framework that combines long-term demand forecasting, strategic investment in carbon capture technologies, and distribution planning under evolving carbon-pricing regimes.


First, an extensive data collection process assembles detailed information on global cement production facilities, bilateral trade flows, and regional emission factors, providing a robust empirical foundation for the analysis. A country-level consumption series for 1995--2023 is then reconstructed by combining historical trade flows with production capacity data, applying outlier detection and interpolation techniques to ensure trend consistency. Building on this foundation, a systematic forecasting exercise produces reliable country-level demand -- which is thereafter distributed over the three biggest populated cities in the country -- projections for 2025--2050, which are spatially disaggregated to define demand nodes for the optimisation model.


The core of the methodology is a Mixed-Integer Linear Programming (MILP) model that simultaneously optimises binary investment decisions in carbon capture retrofits and continuous cement flows across a 25-year planning horizon. The model’s objective function maximises profit by balancing expected revenues against production, transportation, emissions-related costs, storage and transportation costs for captured carbon, and capital expenditures for retrofit investments.


To evaluate model robustness and the influence of future policy environments, the study conducts comprehensive parameter sensitivity and scenario analyses. Sensitivity analysis systematically perturbs key input parameters -- such as production and transportation costs -- to identify which uncertainties most affect optimal investment and distribution strategies. Scenario analysis contrasts three carbon pricing pathways (STEPS, APS, and NZE), each evaluated with and without the CBAM, to investigate how different policy trajectories shape investment decisions and trade flows.


Results show that the CBAM consistently accelerates retrofitting investments and reshapes international cement trade flows across the STEPS and NZE carbon pricing pathways. Under more aggressive carbon pricing scenarios (APS and NZE), domestic European retrofit projects become viable even without the CBAM, although the mechanism continues to redirect marginal investments toward lower-cost regions. Sensitivity tests reveal that assumptions about regional production costs exert the strongest influence on investment outcomes, while variations in transport-related costs have comparatively limited effects.


This thesis makes three key contributions. Methodologically, it provides a validated, end-to-end decision-support framework that integrates comprehensive data preparation, long-term demand forecasting, mathematical optimisation, and detailed post-analysis. Empirically, it offers novel insights into how the CBAM implementation and alternative carbon pricing trajectories jointly determine the geography of carbon capture investments and global cement flows. Practically, it delivers a strategic tool enabling industry stakeholders and policymakers to plan effective long-term decarbonisation strategies under uncertainty, highlighting the importance of targeted support measures in higher-cost regions and the need to align regulatory designs with the economic realities of global supply chains.

Rapid expansion of the e-grocery sector has intensified the need for integrated scheduling of production and distribution under operational uncertainty. This study introduces the first exact solution to the e-grocery integrated production and distribution scheduling problem (IPDSP), specifically designed for fulfillment-center operations and decentralized-hub distribution under demand uncertainty. The IPDSP model simultaneously optimizes single-machine order batching, homogeneous single-trip vehicle routing with delivery time windows, demand splitting, and multi-hub routing. Separate models—a mixed-integer linear programming (MILP) formulation and a genetic algorithm (GA)—are each formulated, presented, and evaluated for solving the IPDSP within realistic runtime constraints. The deterministic MILP is further extended into a stochastic formulation to capture demand variability by generating multiple demand scenarios, incorporating the maximum demand across those scenarios, and allowing the model to optionally exclude certain scenarios. Computational experiments on real-world fulfillment center (FC) data demonstrate that the MILP finds feasible solutions to the deterministic case in 44% of instances (100% for the two smallest FCs) and proves optimality in 7.6% of runs within a one-hour limit. The GA reduces average runtime by 99.6% while incurring only a 3.4% increase in objective value and yields feasible schedules in all cases. The stochastic MILP shows that increasing driving time by 21.3% achieves full scenario coverage, with 80%–90% coverage offering the best cost–robustness trade-off. For large or time-sensitive problems, the GA is recommended; the MILP remains suitable for small instances requiring strict optimality. Overall, these findings advance resilient and efficient e grocery scheduling under uncertainty.

This study analyses the effectiveness of different recovery strategies for transport systems after disruptions, with special attention to the influence of network topology and recovery objectives. By means of simulations on four networks with different structural characteristics (central, radial, decentralized and ring-radial) it is shown that no single strategy is universally applicable. The performance of recovery strategies such as proximity, hierarchy, recovery time and dynamic recovery appears to be strongly dependent on both the degree of fragmentation and the specific structure of the network. Furthermore, the optimal approach differs per functional recovery objective, such as accessibility, efficiency, resilience or robustness. The study therefore not only provides scientific insights into network recovery, but also practical tools for policymakers and planners who have to make decisions under time pressure about recovery after infrastructure disruptions.

Hyperconnectivity (HC) is an emerging concept in logistics that allows for large-scale collaboration of logistics services through shared assets, information exchange, standardised protocols, and flow alignment. A hyperconnected last mile delivery (LMD) system could lead to more optimised routing decisions through processes such as consolidation with fewer vehicle kilometres and more sustainable operations as a result. However, little is known about the combination of connected services that are deemed promising among stakeholders. The Q-methodology was applied to reveal the dominant perspectives among stakeholders on the implementation of HC characteristics in LMD services. Subsequently, the results were fed back to participants to validate whether the analysis was performed accurately. This paper concludes that there is a consensus among stakeholders that the current LMD system is too segmented and that services such as shared parcel lockers and white label concepts could help consolidate delivery operations. Important barriers identified were the fear of losing a competitive market position by sharing information, as well as economic concerns.

The European Union’s (EU’s) Critical Raw Materials Act (CRMA) aims to strengthen the EU’s resource resilience by increasing the autonomy of Critical Raw Material (CRM) supply through EU-based extraction, processing, and recycling, thereby reducing external dependencies and promoting a circular economy. The CRMA mandates, among others, that at least 25% of CRMs come from EU-based recycling. Copper, a key CRM, faces growing demand, and with mining alone, future supply will not fulfill demand. This research analyzed the copper supply chain before and after the CRMA. It was found that recycling is the most promising solution for adherence to the CRMA and a sustainable and resilient future. The study develops a Mixed Integer Linear Programming (MILP) model to optimize the European copper recycling network under the CRMA recycling requirement. The model minimizes total costs while determining the optimal number, locations, and capacities of recycling facilities. The model is tested through a case study of a natural resource company and is usable to similar firms. Results identify four optimal facility locations: Stuttgart, Geithain, Bologna, and Barcelona, which were selected based on, among others, geographic centrality and supply and demand quantities. Sensitivity and scenario analyses reveal the network is vulnerable to facility outages, supply shortages, and significant demand increases. Decentralizing the network by adding a fifth facility improves resilience with limited additional cost.

This thesis investigates how control systems based on Artificial Intelligence (AI) can be safely implemented to optimise the water-related processes in critical infrastructures across Europe, focusing on wastewater treatment plants (WWTPs) in Italy, France, and the Netherlands. The study addresses the overarching question: “How can AI-based solutions be technically effective, economically viable, safe, and acceptable for sustainable water management in Europe?”
The research follows a dual-track approach combining engineering development and policy analysis. On the engineering side, the work develops and evaluates deep reinforcement learning (DRL) controllers for aeration optimisation within activated sludge systems using real operational data provided by SUEZ Digital Solutions. Two state-of-the-art agents, Soft Actor–Critic (SAC) and Twin-Delayed DDPG (TD3), are trained interactively on a linear model for aeration, to respect operational constraints and improve the process. The agents were trained with two different configurations: with and without a buffer of historical transitions that is used as previous knowledge. After training, the agents were benchmarked across multiple disturbance scenarios, generated from real data of energy price and inflow load. Results demonstrate significant improvements compared to baseline control, achieving lower energy consumption, stable dissolved oxygen levels, and better values of redox potential in the tank. These findings confirm the technical feasibility and scalability of DRL-based aeration control for real-world deployment.
On the policy side, the research explores the institutional and governance readiness for adopting AI-based control in critical water infrastructures across Italy, France, and the Netherlands through fifteen semi-structured interviews with regulators, utility managers, and researchers. Using the Transition Model Canvas (TMC) and Multi-Level Perspective (MLP) frameworks, the analysis identifies key barriers and leverage points. In the first group, fragmented governance, infrastructural limits, and lack of AI literacy among stakeholders at all levels were identified, while the second one included regulatory sandboxes, digital-skills training, and pilot projects. Comparative insights show that France benefits from strong national coordination and incumbents (SUEZ, Veolia), Italy faces heterogeneous regional governance and uneven digitalisation, while the Netherlands provides a model of integrated and innovation-oriented regulation.
By integrating both perspectives, the thesis proposes a Transition Model Canvas for AI-based wastewater infrastructure at a European level. It maps how landscape pressures (EU AI Act, Green Deal, and Urban Wastewater Treatment Directive recast) interact with regime actors and niche innovations to shape transition pathways. The work concludes with a set of policy and design recommendations for safe and responsible AI adoption, structured into short-, medium-, and long-term phases.
Overall, this thesis demonstrates that AI-based control systems can substantially improve energy efficiency, regulatory compliance, and sustainability in wastewater management. Their successful adoption, however, requires coordinated regulatory frameworks, skills, and investment in digital infrastructures. The study illustrates how combining systems engineering with policy analysis can support the responsible digital transformation of Europe’s critical water infrastructures.

This study investigates the key factors influencing labor hour demand during the early-life Infant Mortality Phase of Complex Capital Goods (CCGs), using ASML lithography systems as a case. The objective of this research is not to construct a high-performing predictive model, but rather to uncover and understand the key factors underlying labor hours in CCGs when these are not yet known and only historical labor hour data from predecessor machines are available. By elucidating these factors, and understanding the causal mechanisms behind them, a stronger analytical foundation can be established, enabling a more accurate interpretation of the available data and serving as a bridge for future predictive modeling of the labor required to navigate the infant mortality phase.
A mixed-methods approach was adopted, combining expert interviews, focus groups, Multi-Criteria Decision Analysis (MCDA), and Fixed and Random Effects panel regression, combining qualitative expert insights with quantitative statistical modeling. The qualitative phase identified both pre-deployment factors and live operational factors influencing labor demand. These were subsequently quantified and tested using a balanced panel dataset of historical labor hour and performance records from predecessor machines.
The methodological findings demonstrate that even in data-scarce environments, structured inference models can uncover consistent and interesting patterns in early-life labor demand. However, several methodological limitations were identified. The reliability of time-writing and performance data posed challenges for model validation and may have led to the exclusion of potentially relevant factors from the analysis. Additionally, the limited sample size constrained the generalizability of certain results, while inconsistencies in labor hour recording practices introduced potential measurement bias that may further affect the robustness of the inference outcomes.
The contributions of this study are twofold: (i) the development of a methodological framework for inferring labor hour influencing factors of CCGs in the absence of direct input parameters, and (ii) the adaptation of classical statistical inference techniques for novel applications in CCG maintenance planning. While demonstrated using ASML lithography systems, the approach is transferable to other CCG contexts, such as wind turbines, jet engines, and railway vehicles.

The rapidly growing e-grocery sector demands fulfillment centers (FCs) that operate with high efficiency and robustness, even under strong variability in task durations. Traditional scheduling approaches, such as the Earliest Due Date–First Served (EDD-FS) heuristic, lack real-time adaptability and neglect uncertainty. This study develops a real-time adaptive scheduling framework that combines predictive analytics with optimization to improve the timeliness and reliability of picking operations in a large-scale e-grocery FC. Gradient Boosted Decision Tree (GBDT) models are trained to predict active and inactive batch picking durations, significantly outperforming baseline methods. These predictions feed into two Mixed-Integer Linear Programming (MILP) schedulers: a deterministic variant using point estimates, and a scenario-based variant that incorporates uncertainty via sampled task durations. Evaluated in Picnic’s proprietary warehouse simulation system, the deterministic scheduler reduces total delay by 16% and delayed batches by 8.5%, though performance depends on the EDD-FS weighting. The scenario-based scheduler achieves further improvements, decreasing total delay by 6.3% to 25.6% and maximum delay by 16.6% to 37.9%, with higher scenario counts improving outcomes at the cost of computation. These results show that integrating predictive models with deterministic and scenario-based optimization enhances schedule robustness compared to heuristic baselines. Future work should validate the framework in live FC operations, explore rolling-horizon and metaheuristic extensions, and assess reinforcement learning as an alternative for adaptive decision-making. This study thus provides a novel methodological framework with both academic significance and direct operational relevance for the rapidly expanding e-grocery sector.