A Multi-Criteria Decision-Support Framework for Evaluating Alternative Fuels and Technologies Toward Zero Emission Shipping

Abstract

This paper presents an MAUT-based decision-support framework, developed within the NAVGREEN project, to enable the evaluation of alternative fuels and technologies in shipping decarbonization pathways toward zero-emission targets. The framework integrates stakeholder-derived weights elicited through the Analytic Hierarchy Process (AHP) and systematically evaluates alternatives across five criteria: cost, technological maturity, safety and regulatory compatibility, carbon footprint, and social acceptability. Alternatives are mapped into a common utility space through criterion-specific utility functions and aggregated into a composite utility score, enabling transparent and reproducible comparison of single and combined solutions. To strengthen applicability beyond a single illustrative application, the study incorporates a structured scenario and sensitivity analyses (policy stringency, infrastructure constraints, conservative regulatory environments, and weight and parameter perturbations) to assess ranking stability under plausible future conditions. A case study on an Ultramax bulk carrier is used solely to demonstrate the operability and workflow of the method, rather than to empirically validate technology choices across all ship types. Optional AI-assisted elicitation may be used as a supporting aid to harmonize indicative inputs when data are incomplete; however, validation of AI-generated estimates is outside the scope of the present study and is identified as future work.

1. Introduction

The decarbonization of the maritime sector is a critical component of global efforts to mitigate climate change. International shipping contributes approximately 2–3% of global greenhouse gas (GHG) emissions, and this share is expected to rise if no effective mitigation actions are taken [1]. The International Maritime Organization has adopted a revised GHG strategy aiming to reach net-zero emissions by around 2050, aligning with the goals of the Paris Agreement [2].

Achieving zero-emission shipping, however, involves complex trade-offs among alternative fuels and propulsion technologies [3]. Options such as liquefied natural gas (LNG), biofuels, green methanol, ammonia, hydrogen, and battery electric systems differ substantially in terms of carbon footprint, cost, maturity, safety, regulatory readiness, and social acceptance [4,5]. Selecting an appropriate solution for a given vessel or route requires structured decision-making under multiple, often conflicting, criteria and uncertainty. Beyond technology selection, implementation pathways and readiness constraints strongly shape the feasibility of zero-emission solutions [6].

In this study, the term “toward zero-emission shipping” is used in a pathway sense, i.e., to denote decarbonization trajectories aligned with net-zero targets. Accordingly, the alternatives considered may include low/zero-carbon candidates and baseline or transitional solutions that can offer partial GHG reductions and near-term implementability but do not constitute zero-emission solutions on their own. Baseline (transitional) options are retained as benchmarks to make short- to medium-term decision trade-offs explicit and to indicate the conditions (e.g., carbon pricing, infrastructure availability, standards, and regulatory maturity) under which lower-carbon or zero-carbon options become preferred.

Within this context, Multi-Criteria Decision-Making (MCDM) methods provide a rigorous and transparent framework for the systematic evaluation of alternative energy sources in maritime applications [7]. Among them, Multi-Attribute Utility Theory (MAUT) is particularly suited for situations involving quantifiable trade-offs, stakeholder preferences, and heterogeneous data sources [8]. Incorporating stakeholder input through participatory approaches such as the Analytic Hierarchy Process (AHP) enhances the legitimacy and robustness of the decision process [9]. Since MAUT is a mature and widely used MCDA method, this paper does not claim to introduce a new MAUT. Instead, the contribution is a transparent, stakeholder-grounded decision-support workflow for shipping decarbonization.

Furthermore, recent advances in artificial intelligence (AI) can support the estimation of performance indicators, enabling more consistent and data-driven evaluation of technologies. Techniques such as machine learning and expert systems can be used to process large and uncertain datasets, extract patterns, and generate normalized scores that feed into the utility model [10].

MCDM methods can be broadly categorized into two families: Multi-Attribute Utility Theory (MAUT)-based and outranking methods. MAUT approaches rely on the explicit construction of a utility function for each alternative, aggregating performance scores across all criteria, weighted according to their relative importance [11]. A key advantage of MAUT is that it provides a cardinal ranking of alternatives, allowing for quantitative interpretation of trade-offs. This makes it particularly suitable for applications where preferences are known or can be elicited, and criteria can be normalized or transformed into comparable utility scores.

Outranking methods such as ELECTRE and PROMETHEE, by contrast, construct preference relations through pairwise comparisons, without assuming full compensability between criteria [12]. These methods are useful when the decision context involves strong incommensurability or when stakeholders are unwilling to assign precise weights or utility functions. However, outranking methods may be more difficult to communicate to non-expert users and can be sensitive to threshold values and preference structures.

Among other approaches, the Analytic Hierarchy Process (AHP) and Analytic Network Process (ANP) offer flexible frameworks for decomposing complex decision problems and eliciting preferences through pairwise comparisons [9,13]. AHP has been widely applied in transport, energy, and environmental planning due to its simplicity and ability to involve stakeholders. However, its use of ratio scales and eigenvalue calculations has been criticized for potential inconsistency and rank reversal issues [14].

Hybrid methods that combine MCDM with fuzzy logic, AI techniques, or stakeholder consultations are gaining traction in the assessment of sustainability and the decarbonization of the maritime sector [15]. These methods seek to overcome the limitations of single MCDM techniques by integrating expert judgment, uncertainty quantification, and data-driven modeling.

Overall, the literature confirms that MCDM provides a robust methodological foundation for supporting maritime energy transition decisions, provided that the selection of the method matches the nature of the criteria, data availability, stakeholder engagement, and decision objectives.

Despite the growing body of maritime decarbonization research using multi-criteria decision-making to compare alternative fuels and technologies, a clear gap remains in transferable, transparent, and stakeholder-grounded decision-support frameworks. Many published applications are tailored to a single vessel type, route, or regional context, which limits their usefulness as general decision-support tools. In addition, performance indicators are often combined without an explicit, reproducible mapping into a common utility space, making it difficult to compare heterogeneous technical, economic, regulatory, and social inputs on a consistent basis. Stakeholders’ preferences are frequently elicited from limited actor groups or reported at a high level, even though criteria such as safety/regulatory compatibility and social acceptability inherently require structured participatory input. This study addresses these limitations by proposing an MAUT-based workflow that maps heterogeneous indicators into criterion-specific utility functions, integrates AHP-derived weights within a quadruple-helix stakeholder structure, and evaluates ranking robustness under plausible uncertainty and future conditions through scenario and Monte Carlo sensitivity analysis.

The framework can be populated using evidence from literature, validated models, and structured expert elicitation. Optional AI-assisted input elicitation may be used only to structure indicative inputs when data are incomplete. The work is developed in the context of the NAVGREEN project (Green Shipping of Zero Carbon Footprint), which is funded through the Recovery and Resilience Fund under Greece’s National Recovery and Resilience Plan (“Greece 2.0”). NAVGREEN adopts a holistic approach to “green shipping” by developing and evaluating alternative fuels and energy solutions (including production, storage, conversion, and onboard energy management), investigating energy upgrading and footprint reduction measures for ships and ports, advancing digitalization solutions, and promoting circular-economy pathways such as green recycling and energy recovery, supported by a multi criteria decision support tool to map and compare optimal solutions and their combinations.

The aim of this paper is to yield a transparent and reproducible methodology for ranking candidate energy alternatives by maximizing an overall utility function, which aggregates normalized performance across five selected criteria and explicitly reflects stakeholder priorities through weights. In this study, five alternative energy options were evaluated for a bulk carrier case study to demonstrate the model’s applicability. By integrating technical, environmental, economic, and social criteria, the framework provides a holistic tool to support policymakers, shipowners, and technology providers in navigating the transition to sustainable maritime transport.

The framework is vessel-agnostic by design and can be applied to any ship type and size, as well as to any available energy solution or combination thereof, if criterion performance inputs are available from literature, measured data, validated models, or structured expert elicitation. The case study included in this paper is used solely as an illustrative implementation to demonstrate operability and workflow, rather than to empirically validate technology choices for all ship categories. To support generalizability, the paper additionally reports a structured scenario and sensitivity analysis examining ranking stability under plausible future conditions (policy stringency, infrastructure readiness, and regulatory conservatism).

Figure 1 summarizes the overall workflow of the proposed framework and the sequence of methodological steps applied in this study.

2. Materials and Methods

2.1. Justification for Using MAUT

Among the available Multi-Criteria Decision-Making (MCDM) techniques, the Multi-Attribute Utility Theory (MAUT) is particularly suitable for the structured evaluation of alternative marine fuels and propulsion technologies, especially when the objective is to combine multiple quantifiable criteria into a single utility score representing overall preference.

While utility-based approaches such as MAUT aggregate cardinal performance information through explicit utility functions and weights, outranking families such as ELECTRE and PROMETHEE rely on pairwise dominance logic using preference/indifference thresholds and concordance–discordance (ELECTRE) or preference functions and net flows (PROMETHEE). In practice, outranking methods can better represent non-compensatory decision philosophies (i.e., avoiding “good performance” offsetting unacceptable drawbacks), but they typically require additional parameterization and may yield partial orderings or be sensitive to preference-function settings and rank-reversal phenomena. Recent methodological work highlights that PROMETHEE outcomes can be materially affected by preference-function parameters and compensation behavior, motivating careful sensitivity checks when outranking is used. In contrast, the Analytic Network Process (ANP) can be preferable when explicit interdependencies and feedback among criteria must be modeled. However, the present study adopts AHP-derived weights within an additive MAUT structure because it provides a transparent and fully comparable utility scale across heterogeneous criteria and supports direct integration with the robustness analysis (Monte Carlo), which requires repeated evaluation of a single scalar utility score under uncertainty [15].

MAUT is a compensatory method, meaning that poor performance in one criterion may be offset by strong performance in another, provided that appropriate weights reflect the importance of each criterion. This property is desirable in energy and transport decision-making contexts, where trade-offs between cost, emissions, maturity, and safety are inevitable [16]. The method assumes that decision makers or stakeholders are able to express preferences over individual criteria and that these preferences can be translated into utility functions, often normalized on a [0, 1] scale.

The selection of MAUT for the present framework is justified by several factors:

• Quantitative comparability: MAUT allows performance indicators measured on different scales (e.g., €/kWh, g CO₂/MJ, TRL score) to be transformed into a common utility scale, facilitating aggregation and comparison.
• Transparency and interpretability: The additive utility model is intuitive and easy to communicate, especially when supported by visualizations of utility scores and weight distributions.
• Stakeholder integration: MAUT can be effectively combined with participatory methods such as AHP or Delphi for the elicitation of criteria weights, allowing incorporation of expert or stakeholder opinions into the final assessment [17].
• Adaptability to AI-enhanced modeling: The structure of MAUT readily accommodates the integration of AI-generated estimates for performance indicators, especially in cases where empirical data are sparse or uncertain.

Given the availability of stakeholder-derived criteria weights (from structured questionnaires developed in the NAVGREEN project) and the possibility of normalizing technical indicators, the MAUT framework provides a robust and flexible decision-support tool tailored to the objectives of zero-emission shipping.

2.2. Model Description

The proposed decision-support framework is structured around a classic additive Multi-Attribute Utility Theory (MAUT) model, which integrates multiple criteria into a composite utility score for each technological alternative under evaluation. The approach is designed to facilitate transparent, traceable, and stakeholder-informed comparisons of decarbonization options toward zero-emission targets.

Let A = {a₁, a₂, …, aₘ} m ∈ ℕ, denote the set of m technological alternatives (e.g., fuels or propulsion systems), and let G = {g₁, g₂, …, gₙ} n ∈ ℕ, denote the set of n evaluation criteria (e.g., cost, carbon footprint, safety and regulatory compatibility). For each alternative a_j, (j = 1, 2…, m) and each criterion g_i, (i = 1, 2…, n), the alternative’s performance is first represented by a criterion-specific indicator uᵢⱼ (raw score in the original measurement units). The raw indicator is then mapped to a normalized utility value ${\mathrm{e}}_{\mathrm{i}\mathrm{j}}={\mathrm{e}}_{\mathrm{i}}\left({\mathrm{u}}_{\mathrm{i}\mathrm{j}}\right)\in [0,1]$, where higher values indicate greater desirability of a_j with respect to g_i.

Criterion weights ${\mathrm{w}}_{\mathrm{i}}$ reflect the relative importance of each criterion and are elicited through stakeholder pairwise comparisons within the AHP. The weights satisfy the normalization constraint:

$${\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}}=1,\hspace{1em}\hspace{1em}0 \le {\mathrm{w}}_{\mathrm{i}}\le 1.$$

(1)

The overall utility score of alternatives ${\mathrm{a}}_{\mathrm{j}}$ is computed using the additive MAUT aggregation [16]:

$$\mathrm{E}({\mathrm{a}}_{\mathrm{j}})={\mathrm{E}}_{\mathrm{j}}={\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}}{\mathrm{e}}_{\mathrm{i}\mathrm{j}}.$$

(2)

The model yields a cardinal ranking of alternatives and supports structured trade-off and robustness analysis by examining how results respond to plausible variations in weights and/or utility mappings.

The utility function eᵢⱼ (uᵢⱼ) is typically linear, assuming either benefit-type (“the higher the better”) or cost-type (“the lower the better”) behavior:

Benefit-type criterion (higher is better):

$$\begin{array}{cccc}& e_i(u_{ij})={\displaystyle \frac{u_{ij}-u_i^{\mathrm{m}\mathrm{i}\mathrm{n}}}{u_i^{\mathrm{m}\mathrm{a}\mathrm{x}}-u_i^{\mathrm{m}\mathrm{i}\mathrm{n}}}},{ u}_i^{\mathrm{m}\mathrm{i}\mathrm{n}}\le u_{ij}\le u_i^{\mathrm{m}\mathrm{a}\mathrm{x}}& & \end{array}$$

(3)

Cost-type criterion (lower is better):

$$\begin{array}{cccc}& e_i(u_{ij})={\displaystyle \frac{u_i^{\mathrm{m}\mathrm{a}\mathrm{x}}-u_{ij}}{u_i^{\mathrm{m}\mathrm{a}\mathrm{x}}-u_i^{\mathrm{m}\mathrm{i}\mathrm{n}}}}, u_i^{\mathrm{m}\mathrm{i}\mathrm{n}}\le u_{ij}\le u_i^{\mathrm{m}\mathrm{a}\mathrm{x}}& & \end{array}$$

(4)

where $u_i^{\mathrm{m}\mathrm{i}\mathrm{n}}$ and $u_i^{\mathrm{m}\mathrm{a}\mathrm{x}}$ denote the minimum and maximum values used to normalize criterion $g_i$ across the assessed alternatives (or across a predefined feasible range). This linear rescaling maps all criterion performances onto [0, 1], ensuring comparability before aggregation in the MAUT model [18].

Each of the criteria is quantified through its corresponding performance indicator and mapped to a normalized utility score via the criterion-specific utility function defined in the MAUT model. The criterion weights are obtained from the stakeholder-based weighting process and, as reported in Section 3.1, while they are application-dependent, they may vary across vessel categories and operational contexts.

2.3. Selection and Categorization of Alternative Energy Solutions

A comprehensive set of candidate energy and propulsion alternatives was established by mapping currently available and emerging technological pathways that can contribute to reducing ships’ carbon footprint and supporting the energy transition in shipping. In defining the scope of the present framework, the focus was placed explicitly on energy-related (technological) solutions, because other decarbonization measures, such as design improvements (e.g., hydrodynamic optimization) and operational measures (e.g., speed management, routing, and operational planning), can, in principle, be implemented in parallel with any selected energy solution, largely independently of the selected fuel or propulsion pathway. Therefore, to address the central research objective—namely, the comparative investigation of the energy transition toward green shipping (low carbon footprint energy carriers and propulsion technologies), the alternative set was deliberately anchored on energy pathways and their enabling onboard systems, while recognizing that complementary technical and operational measures may be layered onto any option without altering the definition of the underlying energy pathway.

Following this scoping rationale, a baseline set of 23 candidate alternatives, A = {a₁, a₂, …, a₂₃}, was retained and organized into a structured taxonomy to ensure completeness, transparency, and methodological consistency for the subsequent MAUT-based assessment. As shown in

Table 1. Set of baseline selected alternative energy solutions.

j	Alternative Energy Solutions aj	Groups of Solutions
1	Internal Combustion Engines (I.C.E.)-heavy fuel oil	Conventional Internal Combustion Engines
2	I.C.E.-marine diesel oil
3	I.C.E.-natural gas (LNG)	Internal Combustion Engines/LNG-LPG
4	I.C.E.-petroleum gas (LPG)
5	Dual fuel I.C.E.–biodiesel (Β30)	Dual fuel Internal Combustion Engines
6	Dual fuel I.C.E.–methanol/diesel
7	Dual fuel I.C.E.–hydrogen/diesel
8	Dual fuel I.C.E.–ammonia/diesel
9	Dual fuel I.C.E.-LNG/ammonia
10	Low-temperature fuel cells (PEMFC)-hydrogen	Hydrogen fuel cells
11	High-temperature fuel cells (SOFC)-hydrogen
12	Low-temperature fuel cells (PEMFC)-ammonia
13	High-temperature fuel cells (SOFC)-ammonia
14	Batteries	Storage of Energy
15	Sails	Renewable Energy Sources
16	Fletner rotors
17	Kites
18	Wind turbines
19	Photovoltaic panels
20	Waste heat recovery	Waste heat recovery
21	Shore power supply (cold ironing)	Shore power supply (cold ironing)
22	Carbon capture and storage (CCS)	Carbon capture and storage (CCS)
23	Nuclear Power	Nuclear Power

, the alternatives are grouped into coherent families according to their primary energy carrier and conversion pathway, namely conventional internal combustion engine (ICE) baselines, ICE options using gaseous fuels, dual fuel ICE configurations, fuel-cell-based hydrogen/ammonia pathways, electrification and energy storage, wind-assist and other renewable onboard generation, energy-efficiency enabling technologies (e.g., waste heat recovery), shore-side electrification (cold ironing), carbon capture solutions, and nuclear power. This classification enables systematic screening and selection of case-study alternatives by supporting comparisons within and across technology families, while remaining extensible as new solutions and combinations mature.

Importantly, the framework is designed to support the evaluation not only of single technological alternatives but also of combined solutions, where two or more technologies are installed and operate as an integrated configuration. Therefore, the selection may be performed either among the baseline solutions listed in Table 1 or among solutions formed as combinations of those technologies.

2.4. Selection of Criteria

The evaluation framework developed under the NAVGREEN project specifies five primary criteria for assessing alternative zero-emission marine fuels and propulsion technologies. These criteria were defined through an iterative process combining targeted literature review, expert consultation, and stakeholder engagement via structured questionnaires, with the aim of capturing the most decision-relevant dimensions of feasibility and sustainability in maritime decarbonization. This approach is consistent with participatory applications in maritime decision-making, where stakeholder input is used to structure preferences and explore trade-offs in complex policy and operational contexts [19]. Each of the selected criteria is linked to a corresponding performance indicator suitable for normalization and aggregation within the MAUT utility model:

• Economic Cost. Economic performance is captured through cost indicators accounting for both capital expenditure (CAPEX) and operational expenditure (OPEX), expressed in normalized terms (e.g., €/MJ or €/kWh delivered). The indicator reflects not only fuel costs but also ship conversion or new building requirements, onboard storage and fuel-handling systems, and infrastructure implications where relevant. Integration of cost elements follows a total cost of ownership (TCO) logic, which is widely adopted in shipping decarbonization assessments [20,21].
• Technological Maturity. The technological feasibility of each pathway is represented by its Technology Readiness Level (TRL) on a standardized 1–9 scale, complemented by evidence of deployment in maritime applications and supply-chain availability (e.g., certified components and vendors). Alternatives with lower TRL are associated with higher implementation and scale-up risk, which is particularly important when comparing established fuels with emerging solutions [5].
• Safety and Regulatory Compatibility. This criterion captures both operational safety (e.g., toxicity, flammability, storage pressure/temperature, and handling complexity) and regulatory feasibility under existing or evolving frameworks at IMO, EU, flag-state, and class rule levels. The inclusion of regulatory compatibility is essential for assessing novel fuels, such as hydrogen or ammonia, where safety case maturity and prescriptive requirements may still be developing [4].
• Carbon Footprint. This criterion represents the life-cycle greenhouse gas (GHG) emissions of each alternative, expressed in grams of CO₂ equivalent per unit of energy (g CO₂eq/MJ). It reflects consistency with the regulatory trajectory of the EU [22] and the IMO decarbonization pathway [1]. The underlying indicator follows a well-to-wake perspective, incorporating emissions from fuel production, transport, onboard storage, and use, consistent with established LCA practice [23,24].
• Social Acceptability. The fifth criterion reflects perceptions of and acceptance across relevant stakeholder groups, including seafarers, port and local authorities, coastal communities, and society more broadly. It captures issues such as perceived risk, nuisance factors (e.g., noise or odor), trust in safety arrangements, and willingness to support public or private investment in enabling infrastructure. While inherently more difficult to quantify than technical indicators, social acceptability can be approximated using structured stakeholder consultations and evidence from transition experiences in shipping energy pathways [25,26].

The selection of these five criteria was guided by the principles of completeness, non-redundancy, stakeholder relevance, and practical measurability, consistent with best practice for MAUT model construction. The additive MAUT formulation assumes preference independence across criteria, enabling a separable utility representation in which each criterion is mapped to a utility score and aggregated through weights. While real-world criteria may be correlated (e.g., maturity affecting cost), explicit interaction terms are not modeled in the baseline formulation to preserve transparency and avoid double-counting [18].

During the NAVGREEN project, broader sets of candidate criteria were initially considered, including additional environmental indicators (e.g., air pollutants), operational flexibility, and supply/geopolitical dimensions. However, based on expert workshops and a Delphi-like stakeholder consultation process, the final set was prioritized because it aligns directly with the dominant policy drivers of maritime decarbonization, remains applicable across vessel types and operational profiles, can be robustly weighted through AHP without imposing excessive cognitive burden on participants, and can be operationalized through measurable indicators or defensible expert/AI-supported estimates [27]. Overall, the final selection represents a methodologically sound compromise between scientific rigor, policy relevance, and participatory legitimacy, providing a stable basis for subsequent weighting, normalization, and ranking of alternative solutions.

2.5. Criteria Performance Estimation and Data Sources

The credibility of the proposed MAUT-based decision-support framework relies heavily on the robustness and consistency of the performance indicators used to assess alternative fuels and propulsion technologies. Each of the five selected evaluation criteria (carbon footprint, cost, technological maturity, safety and regulatory compatibility, and social acceptability) is linked to a specific performance indicator that has been carefully selected based on scientific validity, policy relevance, and data availability.

The identification of appropriate performance indicators and their measurement units (Table 2) was carried out through a multi-stage process involving literature review, expert elicitation, and alignment with EU regulatory requirements. The objective was to ensure that each indicator:

• accurately captures the core dimension of the corresponding evaluation criterion;
• can be expressed quantitatively or semi-quantitatively on a continuous scale;
• is measurable through available datasets, models, or expert knowledge;
• allows for normalization and integration into the MAUT utility framework.

Table 2. Criteria performance indicators.

i	Criterion	Performance Indicator	Unit
1	Cost	Total Cost of Ownership (TCO) per energy delivered	€/kWh or €/MJ
2	Technological Maturity	Technology Readiness Level (TRL)	1–9 scale
3	Safety and Regulatory Compatibility	Composite Safety/Compliance Index (expert-derived, normalized)	0–1 (utility)
4	Carbon Footprint	Well to Wake GHG emissions (CO2 eq) per energy unit	g CO2-eq/MJ
5	Social Acceptability	Stakeholder Perception Score from Delphi or expert panel analysis	0–1 (utility)

Table 2. Criteria performance indicators.

i	Criterion	Performance Indicator	Unit
1	Cost	Total Cost of Ownership (TCO) per energy delivered	€/kWh or €/MJ
2	Technological Maturity	Technology Readiness Level (TRL)	1–9 scale
3	Safety and Regulatory Compatibility	Composite Safety/Compliance Index (expert-derived, normalized)	0–1 (utility)
4	Carbon Footprint	Well to Wake GHG emissions (CO2 eq) per energy unit	g CO2-eq/MJ
5	Social Acceptability	Stakeholder Perception Score from Delphi or expert panel analysis	0–1 (utility)

The estimation of the criteria performance indicators’ values for each technology or fuel alternative a_i was conducted using a hybrid approach.

The criteria performance indicator values (uᵢ) for each alternative a_i were compiled using a hybrid evidence base that combines:

• peer-reviewed literature and LCA studies for environmental metrics (e.g., well-to-wake GHG emissions for hydrogen, ammonia, and e-methanol);
• industrial databases and technical reports for technology status and cost ranges;
• supplier specifications and publicly available techno-economic information for CAPEX/OPEX and TRL-related inputs;
• structured expert elicitation and stakeholder questionnaires for criteria that are partly qualitative, particularly safety/regulatory compatibility and social acceptability.

When empirical values were unavailable, outdated, or highly variable, transparent proxy assumptions and harmonization procedures were applied to ensure comparability across heterogeneous sources. Where needed, optional AI-assisted elicitation was used only to help structure indicative inputs. However, the present study does not validate AI-generated estimates, and benchmarking is identified as future work. In line with recent maritime AI literature, data-driven approaches (e.g., supervised learning and regression) have been explored to approximate technical and environmental variables from vessel characteristics and operational descriptors, while knowledge-based or hybrid approaches can help translate qualitative assessments into reproducible scoring scales for multi-criteria aggregation [28,29]. In this study, any AI-related use is limited to an optional, supporting role for structuring indicative inputs when evidence is incomplete or heterogeneous, while the MAUT framework itself remains independent of the estimation mechanism. Inputs and assumptions were documented and then normalized through the utility functions to ensure comparability across alternatives within the MAUT framework.

Regarding the qualitative-to-quantitative transformation used to estimate criterion performance for social acceptability, the following concept is applied. Social acceptability is operationalized as a stakeholder perception indicator and quantified using a 5-point Likert scale, where 1 = very low acceptability, 2 = low acceptability, 3 = moderate/neutral acceptability, 4 = high acceptability, and 5 = very high acceptability. Stakeholders express their perception of each alternative on a bounded ordinal scale, where the lowest response option represents minimal acceptability and the highest response option represents maximal acceptability. The resulting social acceptability performance score is then normalized to the $\left[0,1\right]$ interval by anchoring 0 to the minimum Likert value and 1 to the maximum Likert value, using a linear rescaling:

$${\mathrm{e}}_{5j}={\displaystyle \frac{{\mathrm{L}}_j-{\mathrm{L}}_{jmin}}{{\mathrm{L}}_{jmax}-{\mathrm{L}}_{jmin}}}$$

(5)

where ${\mathrm{L}}_j$ is the reported Likert rating (or its aggregated central value across respondents), and ${\mathrm{L}}_{jmin}$ and ${\mathrm{L}}_{jmax}$ are the scale endpoints (1 and 5) for any alternative solution a_j. This transformation yields an interpretable, unitless score that is directly comparable across alternatives and can be used in the MAUT aggregation alongside other normalized criteria.

2.6. Evaluation and Ranking of Alternative and Combined Solutions

Following the definition of the evaluation criteria, the estimation of their weights, and the derivation of each alternative’s normalized performance values, the proposed framework computes, for each candidate solution of the selected set of m (m ∈ N) energy solutions A = {a₁, a₂, …, aₘ} and for a specified ship type and voyage profile, the corresponding value of the normalized utility function using the MAUT aggregation described in Section 2.2. The technological (energy-related) alternatives considered for evaluation are those listed in Table 1. Evaluation may be performed either (i) among individual alternatives in Table 1 or (ii) among combined solutions, where two or more technological alternatives are installed and jointly contribute to the ship’s power demand.

For combined (hybrid) solutions, the framework applies conservative aggregation rules that reflect how system-level outcomes are typically constrained by the most limiting component.

Cost and carbon footprint are approximated as power/energy-share weighted mixtures of the constituent solutions, because of the following:

(i)

direct CO₂ emissions are proportional to fuel use via emission factors (fuel consumption × emission factor) [30,31];

(ii)

propulsion-assisted technologies reduce the share of delivered propulsive energy and therefore reduce fuel/emissions approximately in proportion to that share under a first-order accounting view [32].

For safety and regulatory compatibility, the combined option is treated as a bottleneck attribute and is therefore aggregated using a max operator (i.e., the most adverse safety/regulatory burden dominates), consistent with conservative risk aggregation logic commonly represented through max–min operators in fuzzy/rule-based inference [33].

For technological maturity (TRL) and social acceptability, the combined option is constrained by the least mature/least accepted component, and therefore, a min operator is used (weakest-link principle). This reflects established TRL practice [34], where system readiness cannot exceed the readiness of its least ready critical element, and it provides a conservative feasibility screen for multi-component solutions.

The above scheme is intentionally conservative and intended as a transparent baseline for illustrating how hybrid/combined solutions can be represented within the MAUT structure. In practice, interactions and synergies may exist (e.g., mitigation measures that reduce safety burden, learning effects that improve maturity, or market dynamics that improve acceptance), in which case alternative aggregation operators or interaction-aware models could be used. In this study, the influence of plausible variability in weights and input scores, including qualitative criteria, is examined through the Monte Carlo robustness analysis and ranking-stability metrics reported in Section 3.3.

For a combined solution a_c, consisting of k (k ≤ m; k, m ∈ N) component alternatives, the normalized performance for the cost and carbon footprint criteria is calculated as a power-share weighted average:

$${\mathrm{e}}_{\mathrm{i}\mathrm{c}}={\sum }_{\mathrm{j}=1}^{\mathrm{k}}{\mathrm{p}}_{\mathrm{j}}{\mathrm{e}}_{\mathrm{i}\mathrm{j}}\hspace{1em}\hspace{1em}\mathrm{i}=1,4$$

(6)

where ${\mathrm{e}}_{\mathrm{i}\mathrm{j}}$ is the normalized performance of alternative a_j under criteria ${\mathrm{g}}_1$ (cost) and ${\mathrm{g}}_4$ (carbon footprint), and ${\mathrm{p}}_{\mathrm{j}}$ is the power participation index of alternative a_j in the combined solution a_c:

$${\mathrm{p}}_{\mathrm{j}}={\displaystyle \frac{{\mathrm{M}}_{\mathrm{j}}}{{\sum }_{\mathrm{j}=1}^{\mathrm{k}}{\mathrm{M}}_{\mathrm{j}}}}, \hspace{1em}\hspace{1em}\mathrm{j}=1,2,\dots ,\mathrm{k};\hspace{1em}\mathrm{k}\le \mathrm{m}\mathrm{a}\mathrm{n}\mathrm{d}\mathrm{k},\mathrm{m}\in \mathrm{N}$$

(7)

with ${\mathrm{M}}_{\mathrm{j}}$ denoting the power attributed to the alternative ${\mathrm{a}}_{\mathrm{j}}$ within ${\mathrm{a}}_{\mathrm{c}}$, satisfying:

$${\sum }_{\mathrm{j}=1}^{\mathrm{k}}{\mathrm{p}}_{\mathrm{j}}=1\mathrm{and}0\le {\mathrm{p}}_{\mathrm{j}}\le 1.$$

(8)

This formulation assumes that the normalized performance of each alternative with respect to cost and carbon footprint varies linearly, monotonically, and non-decreasingly with the associated power share.

For the remaining criteria, conservative composition rules are applied: for technological maturity $\left({\mathrm{g}}_2\right)$ and social acceptability $\left({\mathrm{g}}_5\right)$, the combined solution takes the minimum normalized performance among its components, while for safety $\left({\mathrm{g}}_3\right)$ it takes the maximum:

$${\mathrm{e}}_{2\mathrm{c}}=\min \left({\mathrm{e}}_{2\mathrm{j}}\right),{\mathrm{e}}_{5\mathrm{c}}=\min \left({\mathrm{e}}_{5\mathrm{j}}\right),{\mathrm{e}}_{3\mathrm{c}}=\max \left({\mathrm{e}}_{3\mathrm{j}}\right),\hspace{1em}\mathrm{j}=1,2,\dots ,\mathrm{m};\hspace{1em}\mathrm{m}\in \mathrm{N}$$

(9)

All evaluated solutions (combined or not) are then ranked in descending order of their overall normalized utility score:

$${\mathrm{E}}_{\mathrm{c}}={\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}}{\mathrm{e}}_{\mathrm{i}\mathrm{c}}$$

(10)

where ${\mathrm{w}}_{\mathrm{i}}$ denotes the relative weight of criterion ${\mathrm{g}}_{\mathrm{i}}$, with

$$0\le {\mathrm{w}}_{\mathrm{i}}\le 1\mathrm{and}{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}}=1$$

(11)

and ${\mathrm{e}}_{\mathrm{i}\mathrm{c}}$ is the normalized performance of solution ${\mathrm{a}}_{\mathrm{c}}$ under criterion ${\mathrm{g}}_{\mathrm{i}}$.

In the special case of a non-combined solution, ${\mathrm{a}}_{\mathrm{c}}={\mathrm{a}}_{\mathrm{j}}$, then ${\mathrm{e}}_{\mathrm{i}\mathrm{c}}={\mathrm{e}}_{\mathrm{i}\mathrm{j}}$ and ${\mathrm{E}}_{\mathrm{c}}={\mathrm{E}}_{\mathrm{j}}$.

Consequently, solution selection is recommended to proceed from the highest-ranked alternative to the lowest-ranked one, based on the corresponding utility value ${\mathrm{E}}_{\mathrm{c}}$.

2.7. Sensitivity and Robustness Analysis (Monte Carlo)

To assess the stability of the MAUT-based ranking under uncertainty and preference variability, a Monte Carlo sensitivity and robustness analysis is performed [35]. In each simulation run, both the criterion weights and the criteria performance inputs are perturbed within predefined plausible ranges, and the full MAUT workflow (utility mapping and aggregation) is re-executed to generate an updated overall utility score and ranking.

Starting from the baseline ship weight vector, each criterion weight is independently perturbed within ±20% of its baseline value, and the resulting weight vector is re-normalized to ensure ${\sum }_{\mathrm{i}}{\mathrm{w}}_{\mathrm{i}}=1$. This procedure captures plausible variability in stakeholder preferences while preserving the relative structure of the weighting scheme.

For each alternative and criterion, performance indicator values are sampled within uncertainty ranges reflecting variability in available evidence and assumptions. For quantitative inputs (e.g., cost-related indicators and carbon footprint factors), values are sampled using symmetric ranges around the baseline (default: ±20% unless otherwise stated). For criteria relying partly on qualitative scoring (safety and regulatory compatibility and social acceptability), the corresponding numerical scores are perturbed within a bounded interval (default: ±0.10 on the normalized 0–1 scale) to represent plausible scoring variation while preserving scale consistency. All sampled values are constrained to remain within their valid domains and are then transformed into utilities using the same criterion-specific utility functions applied in the baseline evaluation.

The simulation is executed for $\mathrm{N}$ runs (default $\mathrm{N}=\mathrm{10,000}$). For each run $\mathrm{k}$, utilities ${\mathrm{u}}_{\mathrm{i}\mathrm{j}}^{\left(\mathrm{k}\right)}$ are computed and are aggregated into an overall utility score ${\mathrm{U}}_{\mathrm{j}}^{\left(\mathrm{k}\right)}$ for each alternative $\mathrm{j}$ using the additive MAUT model. Robustness is quantified through:

(a) the probability of each alternative being ranked first $\mathrm{P}\mathrm{r}({\mathrm{rank}}_{\mathrm{j}}=1)$,

(b) the frequency of rank reversals relative to the baseline ordering, and

(c) the distribution of overall utility scores (e.g., median and percentile bands) for each alternative.

These indicators provide an explicit measure of ranking stability and help identify which criteria and inputs most strongly drive ranking changes.

3. Results

3.1. Criteria Weights’ Estimation and Stakeholder Involvement

In the proposed decision-support framework, the estimation of weights for the five selected evaluation criteria is a crucial step in ensuring that the preferences of relevant stakeholders are systematically integrated into the assessment of alternative zero-emission solutions. This process was implemented using the Analytic Hierarchy Process (AHP) [13], in line with the guidelines of the NAVGREEN research project.

The selection of participating experts followed established multi-criteria guidelines for Delphi and AHP procedures [17], based on the following conditions:

• Possession of specialized knowledge in the relevant domain (alternative energy solutions in maritime transport).
• Absence of conflict of interest and ability to maintain impartiality.
• Willingness to participate in collaborative consultation processes.
• Balanced representation of the scientific fields required to address the problem.
• Diversity and inclusion of potentially opposing views or stakeholder interests.
• All selected participants belonged to one of the four categories of the quadruple helix: academia, industry, government, and civil society.

This framework ensures the integration of both scientific and technical expertise and socio-political acceptability into complex decisions. Each category contributes a unique perspective:

• Academia offers insights on innovation and scientific robustness,
• Industry ensures feasibility and market relevance,
• Government provides policy and regulatory insights,
• Civil society reflects the societal response and public acceptability.

This classification also allows for the weighting of stakeholder perspectives through group-specific weightings (perception weights), enhancing the inclusivity of the decision-making process [36,37].

A dedicated internal consultation among all NAVGREEN researchers led to the estimation of perception weights for the four stakeholder groups, w_q. Participants completed AHP pairwise comparison questionnaires. Only responses with a Consistency Ratio (CR) < 0.1 were accepted. The resulting normalized weights are presented in

Table 3. Estimation of stakeholders’ groups’ perception weights.

Group (q)	Stakeholder Category	Weight (wq)
1	Academia	0.210
2	Industry	0.341
3	Government	0.169
4	Civil Society	0.281

.

A structured AHP questionnaire was distributed to a representative sample of selected experts from each stakeholder category.

In particular, the AHP questionnaire was distributed to 198 invited stakeholders, from whom 51 responses were collected (overall response rate 25.8%). Respondents were classified according to the quadruple-helix stakeholder groups to capture diverse perspectives: academia (40 invited; 9 responses), industry (63; 15), government/public administration (35; 10), and civil society (60; 17). Expert selection followed predefined purposive criteria, including demonstrated domain expertise relevant to the evaluated fuels and technologies, willingness and availability to contribute through a structured elicitation process, and the intention to ensure balanced representation across stakeholder categories and viewpoints.

To ensure internal consistency of pairwise comparisons, AHP responses were screened using the consistency ratio (CR), retaining only questionnaires meeting the standard acceptability threshold (CR < 0.1). Differences across stakeholder perspectives were not forced into a single consensus judgment. Instead, stakeholder-group views were preserved and combined through the quadruple-helix aggregation logic to obtain the final criterion weights used in the illustrative case study.

Separate evaluations were conducted for cargo ships and passenger ships due to different technical and operational profiles. The responses provided the relative importance (weight) of each criterion (w_iq) for cargo ships and for passenger ships within each stakeholder group, as shown in

Table 4. Relative weights of criteria (w_iq) for cargo ships.

Criterion	Academia	Industry	Government	Society
Cost	0.17	0.15	0.15	0.17
Technological Maturity	0.17	0.13	0.13	0.13
Safety and Regulation	0.39	0.43	0.45	0.38
Carbon Footprint	0.17	0.14	0.20	0.26
Social Acceptability	0.10	0.15	0.07	0.06

and

Table 5. Relative weights of criteria (w_iq) for passenger ships.

Criterion	Academia	Industry	Government	Society
Cost	0.10	0.15	0.14	0.21
Technological Maturity	0.14	0.15	0.12	0.12
Safety and Regulation	0.46	0.40	0.46	0.38
Carbon Footprint	0.18	0.14	0.18	0.19
Social Acceptability	0.13	0.15	0.11	0.10

, respectively.

The final weight of each criterion g_i was computed using a weighted aggregation formula:

$${\mathrm{w}}_{\mathrm{i}}={\sum }_{\mathrm{q}=1}^{\mathrm{z}}{(\mathrm{w}}_{\mathrm{q}}\mathrm{\ast }{\mathrm{w}}_{\mathrm{i}\mathrm{q}});\hspace{1em}0\le {\mathrm{w}}_{\mathrm{i}}\text{≤} 1;\hspace{1em}{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}}=1;\hspace{1em}\mathrm{i}=\mathrm{1,2},\dots ,\mathrm{n}(\mathrm{n}\in \mathrm{N})$$

(12)

where ${\mathrm{w}}_{\mathrm{i}\mathrm{q}}$ denotes the relative weight (i.e., the perceived importance) assigned by the stakeholder group $\mathrm{q}$ of the quadruple helix to the criterion ${\mathrm{g}}_{\mathrm{i}}$, for $\mathrm{i}=\mathrm{1,2},\dots ,\mathrm{n}\left(\mathrm{n}∈\mathbb{N}\right)$, as obtained by applying the Analytic Hierarchy Process (AHP) (in the present case $\mathrm{n}=5$). Separate weight sets are elicited for cargo and passenger ships, as reported in Table 4 and Table 5, respectively. It holds that:

$$0\le {\mathrm{w}}_{\mathrm{i}\mathrm{q}}\le 1\mathrm{and}{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}\mathrm{q}}=1.$$

(13)

${\mathrm{w}}_{\mathrm{q}}$ denotes the perception weight assigned to the stakeholder group $\mathrm{q}$ of the quadruple helix, for $\mathrm{q}=\mathrm{1,2},\dots ,\mathrm{z}\left(\mathrm{z}∈\mathbb{N}\right)$, (in the present z = 4), estimated for cargo or passenger ships, respectively. It holds that:

$$0\le {\mathrm{w}}_{\mathrm{q}}\le 1\mathrm{and}{\sum }_{\mathrm{q}=1}^{\mathrm{z}}{\mathrm{w}}_{\mathrm{q}}=1$$

(14)

By applying the above equations and using the corresponding values of ${\mathrm{w}}_{\mathrm{q}}$ and ${\mathrm{w}}_{\mathrm{i}\mathrm{q}}$, the estimated global criterion weights ${\mathrm{w}}_{\mathrm{i}}$ for the five criteria are obtained, as presented in

Table 6. Final weights (w_i) for cargo and passenger ships.

i	Criterion (gi)	Final Weight (wi)
		Cargo Ships	Passenger Ships
1	Cost	0.16	0.15
2	Technological Maturity	0.14	0.13
3	Safety and Regulation	0.41	0.42
4	Carbon Footprint	0.19	0.17
5	Social Acceptability	0.10	0.13

for cargo and passenger ships.

It is recognized that the perceived importance of evaluation criteria may vary over time, by ship type, or according to the decision maker’s context. Therefore, the framework is designed to be adaptable. Stakeholders (e.g., shipowners) may override default weights and input their own preferences using AHP-based surveys provided within the toolkit.

3.2. Case Study

This section reports the results of applying the proposed decision-support methodology to a case study involving a cargo ship. The case study is intended to demonstrate the practical implementation of the MAUT-based model and to demonstrate how AI-supported performance estimation can populate the performance matrix under data scarcity and uncertainty. The selected ship type and energy solutions should be understood as illustrative (case study) within a framework that is, in principle, scalable to any vessel type/size and any available energy solution or combination thereof.

The evaluated vessel is an Ultramax bulk carrier, representative of the midsize dry bulk segment that is widely used in global trade. In the absence of shipyard-specific data disclosure in the case study summary, the vessel is parameterized using standard Ultramax characteristics commonly reported in industry practice. The main particulars and operational assumptions of the Ultramax bulk carrier used in the illustrative case study are summarized in

Table 7. Ultramax bulk carrier case study inputs.

Item	Value/Assumption
Vessel type/segment	Ultramax bulk carrier (midsize dry bulk segment)
Deadweight (DWT)	~65,000 DWT (representative range)
Length overall (LOA)	~200 m
Beam	~32 m
Design draft	~13 m
Service speed	~14 kn
Main engine power (order of magnitude)	~10 MW
Basis of vessel specification	Representative Ultramax characteristics (no shipyard-specific disclosure available in the case-study summary)
Role of the case study	Illustrative implementation to demonstrate operability/workflow of the framework (not intended as universal validation for all ship types)

. These representative inputs are sufficient for energy demand, emissions, and cost estimations at the level required by a comparative MAUT exercise.

Five alternative energy solutions were assessed: LNG Internal Combustion Engine (ICE), Dual-fuel Methanol (85%)/Diesel (15%) ICE, conventional HFO ICE, HFO ICE with 30% Kite Assistance, and Nuclear Power Plant. These five alternatives were selected as a compact, illustrative set from the broader taxonomy to span distinct and policy-relevant decarbonization pathways. The set includes a conventional baseline (HFO ICE), a near-term transitional mainstream option with established deployment (LNG ICE), an emerging low-/lower-carbon fuel pathway with increasing uptake (dual-fuel methanol/diesel), an enabling retrofit/efficiency measure representing incremental adoption and hybridization (HFO ICE + kite assistance), and a frontier zero-carbon pathway (nuclear power plant) included to stress-test feasibility constraints related to maturity, safety and regulatory acceptance and social acceptability. This selection ensures coverage of both mainstream and cutting-edge technological directions while keeping the illustrative application concise. The evaluation was performed across the five selected criteria (cost, technological maturity/TRL, safety and regulatory compatibility, carbon footprint, and social acceptability), using the cargo ship weight set applied in the study (w_i = 0.16, 0.14, 0.41, 0.19, 0.10).

A key implementation challenge in maritime decarbonization assessments is that empirical, harmonized datasets across all criteria and all fuel/technology pathways are often incomplete, inconsistent across sources, or not directly transferable across vessel types. To address this challenge, the proposed methodology foresees the use of AI-enabled performance estimators to generate the criterion performance values u_ij for each alternative a_j [28]. The estimators are used to interpolate missing values, extrapolate from known reference cases to the case-study vessel, and maintain methodological consistency by producing comparable outputs aligned with the selected performance indicators and the MAUT utility functions.

In operational terms, the AI layer produces raw indicator values (e.g., total cost of ownership, life-cycle emission intensity, qualitative-to-quantitative safety scores), which are subsequently normalized through the criterion-specific utility functions defined in Section 2.2. The result is a normalized performance matrix e_ij that can be aggregated by MAUT. The AI-supported estimation approach can be summarized by the following criterion:

Cost: Cost indicators are estimated through a techno-economic performance estimator that combines engineering parametrization (installed power, energy demand profile, fuel storage/handling requirements) with market inputs (fuel price scenarios, CAPEX ranges for engines/conversion, and OPEX components). Supervised regression models (e.g., tree-based regressors) can be trained on reference techno-economic datasets to predict a consistent cost metric for each pathway (e.g., normalized TCO or energy delivered cost), which is then mapped to utility using a cost-type normalization (lower cost → higher utility).

Technological Maturity (TRL): TRL is estimated using evidence-based scoring anchored on documented deployment, certification status, supplier availability, and operational references. In practice, this can be supported by semi-automated literature mining and rule-based classification to assign a TRL-consistent value per option, while the output is treated as a benefit-type criterion (higher TRL corresponds to higher utility).

Safety and Regulatory Compatibility: Safety performance is estimated using expert-system logic that encodes fuel properties and regulatory readiness (e.g., flammability/toxicity, storage pressure/temperature, bunkering complexity, and alignment with existing or emerging class/IMO frameworks). Fuzzy inference can be used to translate qualitative judgments and risk thresholds into a bounded quantitative score. Then the resulting value is normalized as a benefit-type criterion to reflect safer and more regulation-ready solutions.

Carbon Footprint: Carbon footprint indicators are estimated under a life cycle perspective (well-to-wake), combining emission factors from established LCA inventories with vessel energy/demand models. Machine-learning regression can be used to estimate emission intensity for the case study vessel by linking operational and technical features to well-to-wake emission outcomes, producing an emission metric that is normalized as a cost-type criterion (lower carbon footprint corresponds to higher utility).

Social Acceptance: Social acceptance is estimated through structured stakeholder perception inputs and expert-coded proxies (e.g., perceived safety, community acceptance, and operational familiarity). AI can support the conversion of qualitative responses into consistent quantitative scores through fuzzy expert systems and/or NLP-based processing of feedback, yielding a normalized benefit-type score (higher acceptance corresponds to higher utility).

Table 8 reports the normalized criteria performance estimators e_ic and the resulting MAUT utility scores E_c for the five energy solutions (including combined solutions). The ranking is obtained through the additive MAUT aggregation:

$${\mathrm{E}}_{\mathrm{c}}={\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}}{\mathrm{e}}_{\mathrm{i}\mathrm{c}}, (\mathrm{i} = 1,2,3,4,5),$$

(15)

using the weights w_i in the case study (for cargo ships as provided by Table 6). Notably, safety and regulatory compatibility carry the highest weight (w₃ = 0.41), implying that differences in safety-related utility have a dominant effect on the final ranking.

Table 8. Normalized criteria utilities and overall MAUT scores for an Ultramax bulk carrier (case study).

		Criteria Performance Indicators ( ${\mathbf{e}}_{\mathit{i}\mathit{c}}$)
c	Alternative	Cost (i = 1)	Technological Maturity (i = 2)	Safety and Regulatory Compatibility (i = 3)	Carbon Footprint (i = 4)	Social Acceptability (i = 5)	Utility Function (Ec)
1	LNG Engine	0.75	1.00	1.00	0.40	0.60	0.81
2	Dual-fuel Methanol (85%)/Diesel (15%)	0.50	0.50	1.00	0.60	1.00	0.77
3	HFO Engine	1.00	1.00	1.00	0.00	0.20	0.73
4	HFO + 30% Kite	0.88	0.50	0.50	0.50	0.60	0.57
5	Nuclear Power	0.25	0.00	0.00	1.00	0.00	0.23
	Criteria Weights (wi)	0.16	0.14	0.41	0.19	0.10

Table 8. Normalized criteria utilities and overall MAUT scores for an Ultramax bulk carrier (case study).

		Criteria Performance Indicators ( ${\mathbf{e}}_{\mathit{i}\mathit{c}}$)
c	Alternative	Cost (i = 1)	Technological Maturity (i = 2)	Safety and Regulatory Compatibility (i = 3)	Carbon Footprint (i = 4)	Social Acceptability (i = 5)	Utility Function (Ec)
1	LNG Engine	0.75	1.00	1.00	0.40	0.60	0.81
2	Dual-fuel Methanol (85%)/Diesel (15%)	0.50	0.50	1.00	0.60	1.00	0.77
3	HFO Engine	1.00	1.00	1.00	0.00	0.20	0.73
4	HFO + 30% Kite	0.88	0.50	0.50	0.50	0.60	0.57
5	Nuclear Power	0.25	0.00	0.00	1.00	0.00	0.23
	Criteria Weights (wi)	0.16	0.14	0.41	0.19	0.10

The results indicate a clear preference ordering driven by the interplay between safety/regulatory readiness, maturity, and carbon performance. The LNG engine achieves the highest overall utility (E₁ = 0.81), largely because it scores the maximum level in both TRL and safety (e₂₁ = e₃₁ = 1.00), reflecting high maturity and comparatively established regulatory pathways. Although its carbon footprint utility is moderate (e₄₁ = 0.40), the strong performance in the most heavily weighted criterion (safety) and the high TRL outweigh the carbon penalty within the applied weighting profile.

Dual-fuel methanol (85%)/diesel (15%) ranks second (E₂ = 0.77). Compared with LNG, it shows lower cost and TRL utilities (e₁₂ = e₂₂ = 0.50) but performs better in carbon footprint (e₄₂ = 0.60) and reaches the maximum social acceptability (e₅₂ = 1.00). The combination of perfect safety utility (e₃₂ = 1.00) and strong social acceptability is sufficient to compensate for its maturity and cost disadvantages.

The conventional HFO engine ranks third (E₃ = 0.73). Its high cost, TRL, and safety utilities (e₁₃ = e₂₃ = e₃₃ = 1.00) contribute substantially to the aggregate score. As expected, it receives the minimum carbon footprint utility (e₄₃ = 0.00) and low social acceptability (e₅₃ = 0.20). Nevertheless, the solution remains competitive under the applied weights because safety alone contributes a weight w₃ = 0.41 to the total utility. This illustrates how weight structures can materially shape rankings, especially when one criterion dominates.

HFO with 30% kite assistance ranks fourth (E₄ = 0.57). While kite assistance improves carbon footprint utility to e₄₄ = 0.50 compared with the baseline HFO case, the overall score is penalized by reduced safety utility (e₃₄ = 0.50) and lower TRL utility (e₂₄ = 0.50), suggesting higher perceived risk and/or lower regulatory/operational readiness. Given the high weight assigned to safety, this penalty offsets much of the carbon improvement.

Nuclear power ranks last by a large margin (E₅ = 0.23), despite achieving the maximum carbon footprint utility (e₄₅ = 1.00). The low score is explained by minimum TRL safety and social acceptability utilities (e₂₅ = e₃₅ = e₅₅ = 0.00), reflecting severe barriers related to regulatory feasibility, safety perception, and societal acceptance. This outcome demonstrates the ability of the framework to prevent ‘single-criterion dominance’, where an option performs extremely well environmentally (zero carbon footprint) but fails to satisfy feasibility constraints.

From a decision support perspective, these results should be interpreted as an illustrative outcome under the specific weight profile used for a bulk carrier in the case study. If decision makers place stronger emphasis on carbon footprint (e.g., under stricter policy or carbon pricing scenarios), the relative ranking may shift. Nevertheless, the case study confirms that the MAUT structure:

• transparently exposes trade-offs;
• allows the explicit incorporation of stakeholder priorities through weights;
• can integrate AI-estimated performance indicators in a consistent and auditable manner (the AI layer should be viewed as a complementary mechanism that enhances the practicality of the model in data-sparse contexts).

To preserve credibility and stakeholder trust, inputs and assumptions should be transparently documented and, where feasible, complemented with uncertainty ranges examined through sensitivity and robustness analysis [35].

3.3. Monte Carlo Robustness Results and Ranking Stability

This section reports robustness indicators obtained from the Monte Carlo sensitivity analysis described in Section 2.7. The baseline utility matrix and cargo-ship weight set are taken from the case-study inputs. In each Monte Carlo run, criterion weights are perturbed within ±20% of their baseline values and re-normalized to sum to one, and criterion-level performance utilities are perturbed within predefined bounds and propagated through the MAUT aggregation to obtain updated overall utility scores and a corresponding ranking for each alternative.

Across N = 10,000 runs, the baseline ranking order is reproduced in 73.29% of simulations, while at least one rank reversal occurs in 26.71% of runs. Ranking stability is therefore high overall, with most reversals concentrated among the top alternatives where baseline utilities are relatively close.

Table 9. Monte Carlo robustness indicators (N = 10,000).

Alternative	Baseline Rank	Pr (Rank = 1)	Pr (Top-2)	Mean Rank	Mean Utility	Median Utility	P05 Utility	P95 Utility
LNG Engine	1	0.787	0.997	1.216	0.789	0.790	0.744	0.831
Dual-fuel Methanol (85%)/Diesel (15%)	2	0.212	0.945	1.843	0.761	0.762	0.718	0.803
HFO Engine	3	0.002	0.058	2.941	0.704	0.705	0.654	0.752
HFO/30% Kite	4	0.000	0.000	3.999	0.570	0.570	0.516	0.624
Nuclear Power	5	0.000	0.000	5.000	0.233	0.233	0.190	0.278

reports standard robustness indicators from the Monte Carlo analysis. “Baseline rank” denotes the ordering obtained from the deterministic (baseline) evaluation. Pr (rank = 1) is the proportion of Monte Carlo runs in which an alternative attains the highest overall MAUT score (i.e., is ranked first), while Pr (in top-2) is the proportion of runs in which it appears among the two highest-ranked alternatives. Mean rank denotes the average ranking position across all runs (lower values indicate stronger overall performance). Mean utility and median utility summarize the central tendency of the overall utility score distribution for each alternative across runs, and P05 utility and P95 utility provide the 5th and 95th percentiles, respectively, indicating the dispersion (uncertainty band) of overall utility outcomes under the adopted perturbations.

As shown in Table 9, the probability of rank = 1 metric indicates that LNG remains the most robust top-ranked option under the adopted uncertainty structure, while dual-fuel methanol/diesel is the primary competitor, achieving rank-1 in approximately one-fifth of runs. By contrast, HFO becomes top-ranked only rarely and enters the top-2 in a small fraction of runs, while HFO/kite and nuclear power remain consistently dominated under the cargo-ship weight structure and the adopted performance uncertainty bounds. Overall, the Monte Carlo results confirm that the baseline ranking is generally stable, and they provide a transparent quantification of how uncertainty propagates into ranking variability, particularly among close-performing top alternatives.

4. Discussion

4.1. Policy and Decision-Making Implications

A key strength of the MAUT-based approach is that it makes the transition “logic” explicit and analytically testable. Decision-makers can perform scenario and sensitivity analyses to examine which conditions would need to change for low-carbon pathways to rank higher. For instance, increasing the weight of the carbon footprint (or introducing minimum performance thresholds) would shift the ranking toward alternatives with strong well-to-wake emissions profiles. Likewise, interventions that improve regulatory clarity, codes and standards, bunkering procedures, training requirements, and port readiness would directly increase the normalized safety and regulation utilities of emerging fuels. This transparency directly supports evidence-based policy design. Rather than debating technologies in the abstract, stakeholders can identify which criterion bottlenecks should be addressed to accelerate adoption.

The methodology is also well-suited to portfolio thinking, as it can evaluate both single technologies and combined solutions. When hybrid solutions are considered, the framework can quantify how the assumed power share of each technology affects the combined performance on cost and carbon footprint while using conservative aggregation rules for maturity, safety, and social acceptability. This capability is particularly relevant across ship types, where no single pathway is universally optimal across all voyage patterns and where incremental adoption of enabling technologies can complement the primary fuel transition.

4.2. Optional AI-Assisted Input Elicitation

Recent literature reviews confirm the growing role of AI in maritime transport for tasks such as emissions and fuel consumption prediction, energy management, and risk assessment, highlighting both the potential benefits and the need for careful validation and interpretability [28,29]. However, in alternative-fuel assessments, performance inputs are often heterogeneous and uneven in their availability across technologies, markets, and maturity levels, particularly for emerging solutions. The proposed MAUT-based framework is intentionally independent of the input generation mechanism and can be populated using validated engineering models, literature sources (e.g., LCA studies), structured expert elicitation, or other evidence-based datasets. In this study, any AI-related element is treated strictly as an optional support tool to help structure indicative inputs when consistent data are incomplete, for example, by harmonizing qualitative descriptions into comparable, criterion-aligned descriptors.

To avoid overstating its contribution, AI is not positioned as a means for validating the framework or for proving the superior accuracy of the resulting rankings. No claims are made regarding the predictive accuracy of any AI-generated estimates in the present work. Establishing reliability through systematic benchmarking against verified operational datasets and controlled experiments, particularly for criteria that combine technical, regulatory, and perceptual dimensions (e.g., safety and regulatory compatibility and social acceptability), is outside the scope of this study and is identified as future research.

In practical decision-support deployments, optional AI-assisted elicitation may be used as a transparent, auditable step to organize and document indicative inputs (together with their sources and assumptions) prior to utility mapping and MAUT aggregation, while final decision recommendations should rely on the best available verified evidence and stakeholder validation.

4.3. Limitations and Sensitivity Considerations

Several limitations should be acknowledged. First, the Section 3 reports a single illustrative case study. Accordingly, rankings should not be generalized across ship types, sizes, and operational profiles without re-estimating the indicator values and repeating the MAUT computation. Second, the criteria weights reflect a specific stakeholder elicitation exercise and, by design, will vary across contexts. The methodology addresses this by allowing weight re-elicitation for different decision settings. Nevertheless, conclusions about “best” technologies should always be interpreted as conditional on the assumed preference structure.

Third, uncertainty in performance indicators remains material. The resulting rankings are conditional on the quality, assumptions, and representativeness of the underlying inputs, irrespective of whether these inputs are derived from literature sources, validated models, structured expert elicitation, or proxy estimates. To assess the stability of the ranking under uncertainty, we performed a Monte Carlo simulation in which both criterion weights and key performance inputs were perturbed within plausible bounds. For each run, criterion weights were independently varied within ±20% of their baseline values and then re-normalized to sum to one. Performance inputs were sampled within uncertainty ranges reflecting variability in costs, maturity assumptions, carbon footprint factors, and qualitative scoring (safety and regulatory and social acceptability) translated into numerical scales. For each Monte Carlo run, utility scores were recomputed and aggregated through the additive MAUT model to produce an alternative ranking. Robustness was quantified using the probability of each alternative being ranked first, the frequency of rank reversals relative to the baseline ranking, and the distribution of overall utility scores.

The ranking produced by the framework is conditional on the assumed policy and market environment and should be interpreted accordingly. If carbon prices rise (e.g., through carbon taxation or tighter market-based measures), carbon-intensive alternatives experience higher effective operating costs, which directly affects the cost criterion through carbon-related cost components and can also strengthen the relative advantage of alternatives with lower lifecycle emissions through the carbon footprint criterion. Under such conditions, options with a lower carbon footprint may improve their overall utility and, depending on the magnitude of the carbon-price shock and the adopted weight set, may challenge or overtake transitional options.

Similarly, improvements in green-fuel infrastructure (availability of bunkering, supply-chain maturity, operational standards, and permitting experience) can shift several criteria simultaneously. Expanded infrastructure can reduce premium and logistics penalties (cost), support learning-by-doing and wider deployment (technological maturity/TRL), reduce implementation barriers via clearer procedures and standardization (safety and regulatory compatibility), and increase perceived feasibility and legitimacy (social acceptability). Importantly, the framework accommodates such dynamics by updating the relevant indicator values (and/or stakeholder weights) and re-running the MAUT aggregation. Ranking sensitivity to these changes can be explored deterministically through “what-if” parameter updates or probabilistically through the Monte Carlo robustness analysis, which quantifies ranking stability and the conditions under which rank reversals occur.

The reported scores and ranking are conditional on the assumed policy and market context and should be interpreted as a baseline scenario. The framework is explicitly scenario-ready. Thus, future regulatory assumptions can be introduced by updating the performance inputs and rerunning the MAUT aggregation. For example, a higher carbon price (or carbon tax) increases the effective cost of higher-emitting options (cost criterion) and strengthens the preference for lower-carbon pathways. Likewise, stricter fuel standards and expanded green fuel infrastructure can improve maturity, reduce regulatory friction, and increase social acceptability for emerging options. Such “what-if” scenario updates can be examined deterministically (by parameter updates) or probabilistically through the Monte Carlo robustness analysis, which quantifies ranking stability and potential rank reversals under plausible shifts in key inputs.

Finally, the additive MAUT formulation assumes compensability between criteria. While this is appropriate for many planning contexts, some decision makers may prefer non-compensatory rules (e.g., minimum safety requirements or carbon intensity caps). In such cases, the MAUT ranking can be complemented with constraints or screening thresholds before final selection.

5. Conclusions

Developed within the NAVGREEN project, this paper proposes an integrated MAUT-based decision-support framework to evaluate alternative fuels and technologies in decarbonization pathways toward zero-emission targets. The framework integrates an additive MAUT model with AHP-derived stakeholder weights. Five criteria (cost, technological maturity, safety and regulatory compatibility, carbon footprint, and social acceptability) were quantified using performance indicators, normalized through utility functions, and aggregated into a composite utility score. The resulting structure supports transparent comparison of alternative fuels and technologies and can accommodate both single and combined solutions. The framework is vessel-agnostic by design and can be populated with evidence from the literature, validated models, or structured expert elicitation.

The proposed decision-support framework is policy-adaptive by design. It can be re-run as carbon pricing tightens, compliance costs evolve, and green fuel infrastructure and standards mature. In this sense, the framework supports decision-makers by identifying the conditions under which transitional solutions may dominate (e.g., high TRL and regulatory readiness) and the thresholds at which lower-carbon options become preferred (e.g., higher effective carbon costs and improved infrastructure). This is aligned with recent literature showing that the inclusion of shipping in EU carbon pricing can materially change incentives and may alter technology and fuel choices [38].

More broadly, recent work highlights the growing availability of multi-source operational data for emissions characterization in shipping, supporting periodic updates of performance inputs within transparent MCDA frameworks as evidence and infrastructure evolve [39].

The Ultramax bulk carrier case study, incorporating five alternative energy solutions, is used as an illustrative implementation to demonstrate the operability and workflow of the framework and to highlight the sensitivity of rankings to the adopted weighting structure. Under the cargo ship weights, alternatives with high maturity and clearer safety/regulatory treatment can dominate the composite utility score even if their life cycle emissions performance is weaker, emphasizing that implementation feasibility is a central driver of near-term decisions. At the same time, the framework identifies the critical levers, policy interventions, infrastructure and standards development, as well as industrial learning required for low- and zero-carbon pathways to achieve prominence in future assessments.

Future research should expand the empirical basis of the performance database and apply the model across additional ship categories, voyage profiles, regional contexts, or even criteria. The ranking could be interpreted alongside externalities not fully represented by the five criteria, particularly local air quality impacts [40]. Methodologically, uncertainty-aware extensions, including robustness analysis of weights and indicator ranges, are recommended. Where AI-assisted input elicitation is used as an optional support mechanism to structure indicative inputs, future work should benchmark such estimates against verified operational datasets and established modeling evidence.

Concretely, future work may broaden validation by applying the framework to additional ship categories and route profiles and, where policy questions require it, extend the environmental layer beyond GHGs to include local air-pollutant indicators (e.g., NO_x, SO_x, and black carbon).

Finally, the methodology can be operationalized into a practical decision-support tool (e.g., an online interface or structured spreadsheet application), enabling users to update inputs, select stakeholder weight sets, and run scenario and Monte Carlo robustness analyses to support transparent decision-making, supporting iterative updates as technologies mature and regulatory requirements evolve [1,22].