Deliverable D3.3 presents the safety, security, and resilience assessment of the AUTOFLEX inland waterway transport system. Led by the National Technical University of Athens (NTUA), the work develops a structured risk-based methodology that combines hazard identification, network science, and risk mitigation to assess how disruptions may affect the continuity of cargo flows across the AUTOFLEX network.
Background and Scope
As the AUTOFLEX transport system integrates autonomous vessels, Temporary Port Terminals, Mobile Distribution Centres, and Stow & Charge hubs into an interconnected logistics network, understanding how local failures can propagate and affect the wider system becomes essential. Task 3.6 addresses this by developing a methodology specifically designed to capture systemic interdependencies and quantify how disruptions may affect network connectivity, operational continuity, and supply chain performance.
The work builds on the transport system design from D3.2 and focuses exclusively on risks that may cause supply chain delays, covering operational, infrastructural, environmental, technical, organisational, and cybersecurity-related disruption mechanisms.
The Three-Pillar Methodology
The assessment is structured around three closely connected pillars:
- Hazard Identification: A HAZID-based process was used to systematically identify and classify hazardous events relevant to the AUTOFLEX transport system and its innovations, covering ports, TPTs, Stow & Charge hubs, MDCs, vessel routes, and Oskar 2.0. Three dedicated online workshops were held with project partners: Workshop 1 with ISE (11/03/2026), Workshop 2 with DFDS (11/03/2026), and Workshop 3 with DST (16/03/2026). Each hazardous event was assessed using a 5×5 risk matrix and ranked according to probability and impact. Risk levels were further refined using node-specific TPT building block factors and route-specific factors including traffic intensity, CEMT class, and route length.
- Risk Assessment: The transport system was modelled as a weighted graph, with nodes representing ports and AUTOFLEX infrastructure locations and edges representing vessel routes. Node and edge weights were derived directly from the HAZID risk levels, making this a risk-informed network rather than a purely structural one. Graph theory metrics were applied to characterise the network’s structural properties, and percolation theory was used to simulate how the network degrades under five removal scenarios: random removal, node weight, edge weight, weighted degree, and weighted betweenness centrality.
- Risk Mitigation: Three mitigation strategies were formulated and evaluated. An exploratory revised network configuration with 30 nodes and 68 directed edges was assessed and improved percolation thresholds across all five disruption scenarios while also reducing the network diameter, demonstrating that increased connectivity and route alternatives can improve the resilience of the AUTOFLEX transport system.
Key Findings
The original AUTOFLEX network has a sparse, corridor-based structure with Rotterdam playing a dominant intermediary role. The low network density and absence of local clustering indicate limited redundancy and strong dependence on a small number of central nodes.
The percolation analysis showed that the network is more tolerant to random failures than to targeted disruptions. Under random removal the critical percolation threshold is pc = 0.250, meaning the network begins to fragment when approximately 25% of nodes are removed at random. Under targeted removal based on edge weight the threshold is pc = 0.176. Under targeted removal based on weighted degree, weighted betweenness centrality, and node weight, the threshold drops to pc = 0.125, confirming that the removal of just 2 nodes is sufficient to fragment the network.
Utrecht consistently ranks as the most critical node across all metrics. By node weight, the top four critical nodes are Utrecht, Moerdijk, Rotterdam, and Leiden. The risk-weighted analysis reordered criticality rankings compared to a purely topological assessment, revealing vulnerabilities that an unweighted analysis would have missed. The impact analysis of Utrecht’s removal showed that criticality redistributes onto a concentrated set of secondary nodes: Moerdijk, Rotterdam, Den Haag, Leiden, Delft, and Eindhoven 2, which should therefore be included in resilience planning.
Outlook
The methodology and results from D3.3 provide a quantitative basis for identifying critical nodes and edges, assessing disruption thresholds, and prioritising resilience interventions. Future work is recommended to integrate performance-based indicators such as expected delay, rerouting cost, and berth and charging capacity constraints, as well as a formal sensitivity analysis and the integration of penetration testing results as the AUTOFLEX digital systems mature.
Partner Contributions
This deliverable was produced by contributions across the AUTOFLEX consortium:
NTUA – Lead authors: Lianna Serafeim, Marios Koimtzoglou, Konstantinos Louzis
SO – Reviewer: Renan Guedes Maidana
