2. Research approach

The research presented in this paper is the result of a 10-month research project in collaboration with one of the major companies worldwide operating in the ship design and construction industry, a consulting company working on modular product development, a consulting company working on meta modeling for systems engineering, and another university partner. The research can be framed into the design research methodology (DRM) by (Reference Blessing and ChakrabartiBlessing & Chakrabarti, 2009), encompassing research clarification, descriptive study, and an initial prescriptive study. In the frame of DRM, participatory action research (PAR) by (Reference Avison, Lau, Myers and NielsenAvison et al., 1999) has been the main research approach for problem identification and for the development of the intended support as part of the initial prescriptive study. Data collection happened through meetings and workshops both online and on-site at the company facilities and university premises during the whole period of the project. Besides the close collaboration with practitioners, research clarification has also been performed through a literature review and analysis of official published documents. The implementation of the proposed approach in the case study was based on modeling and simulation activities performed by the authors supported by access to industrial data sources. In cases where data were missing, AI-generated inputs were utilized, primarily using GPT-4o and o1 models. The validation and verification of the approach happened through a 6-hour prototyping and testing sessions run at the ship manufacturer facilities in the presence of a consulting company and the other university partner. The purpose of the validation activity was to receive feedback about the applicability and usability of the method, validate the results obtained in the example case, and set up a plan for the further development of the approach and its integration into larger decision support systems. The results of such phase are described in the discussion session of this paper.

3. CO₂ estimation in ship design

The estimation of CO₂ emissions in ship design is experiencing a transition from operational-centric estimations to comprehensive, design-integrated approaches. It is driven by increasing regulatory pressures and general awareness amid global decarbonization targets. Traditionally, CO₂ estimation primarily focused on operational emissions, particularly those related to fuel consumption for ship propulsion and hotel load (Reference Moreno-Gutiérrez, Pájaro-Velázquez, Amado-Sánchez, Rodríguez-Moreno, Calderay-Cayetano and Durán-GradosMoreno-Gutiérrez et al., 2019).

Life Cycle Assessment (LCA) has emerged as a critical methodology for evaluating the environmental impacts of ships across all stages of their life cycle. Early studies applied LCA to quantify emissions during ship operation, but recent efforts now encompass all other aspects of the cruise ship lifecycle. For example, (Reference Perčić, Ančić, Vladimir, Fan, He, Soares and SantosPerčić et al., 2021) conducted a comprehensive LCA of battery and diesel-powered ferries which assessed the environmental impact of battery and diesel-powered ferries across multiple life cycle stages. The study did not limit its analysis to operational emissions alone and included emissions linked to the design and construction phases as well. Another significant development in CO₂ estimation is the incorporation of DSE which assists in making decisions regarding optimal configurations. Along similar lines, (Reference Koronakos, Smirlis, Plitsos, Tzannatos, Karalis and PoutosKoronakos et al., 2023) presented a decision support system (DSS) called Optiship that compares design configurations that correspond to the three main LCA dimensions criteria, on economic, environmental, and social KPIs. However, DSS is devoted to the assessment of the ship’s operation and the end-of-life phases. The integration of LCA into the design phase is further supported using life-cycle performance assessment framework. It enables simultaneous evaluation of lifecycle cost and life-cycle emissions (Reference Gualeni, Flore, Maggioncalda and MarsanoGualeni et al., 2019).

For the estimation of scope 3 upstream CO₂ emissions, LCA-based methods are frequently employed (Reference Chatzinikolaou and VentikosChatzinikolaou & Ventikos, 2014) using bottom-up or top-down approach. Bottom-down approach starts with a higher-level view of the system and breaks it down into its components. On the other hand, the bottom-up approach starts from the sub-system or low-level view of the systems and builds up to form the overall system. (Tran & Lam, 2024) and (Jang et al., 2024) use the bottom-up approach in LCA, leveraging operational data such as ship routes, fuel consumption, and port activities to create detailed emissions inventories. These inventories often highlight the spatial and temporal distribution of emissions, emphasizing the impact of specific regions and operational modes (Yu et al., 2024). Other methodologies include predictive models based on ship speed and engine characteristics, which are compared with LCA simulations to validate their accuracy (Jang et al., 2024). Overall, the quantification of CO₂ emissions in the shipping industry is an intricate process that involves detailed modeling of various lifecycle phases and operational conditions to provide a comprehensive understanding of the environmental impact. However, the primary focus of research lies in the estimation and optimization strategies during the operational phase of the existing ships as the operation phase is typically identified as the most significant contributor to CO₂ emissions (Quang et al., 2020). However, due to the sheer scale and material intensity of cruise ships, incorporating scope 3 upstream emissions associated with the production, transportation, and processing of materials used in shipbuilding can have a significant impact on overall CO₂ emissions (Mathew & Kumar, 2024).

4. Data-driven CO₂ emission calculation in ship design: the proposed approach

4.2. Results of the initial prescriptive study: logic of the proposed approach

The approach proposed as a result of the initial prescriptive study consists of the integration of a parametric CO₂ emissions quantification model into a decision-support environment where different concept configurations are assessed, analysed, and visualized based on their economic engineering, engineering, and CO₂ emission performances. The decision support environment (DSE) is visualized in Figure 1 in the area marked with number 1. The decision support environment consists of a GUI to prompt the user to input design parameters to visualize various design configurations. Some parameters are used to set the higher-level boundary of the system such as the overall size of the cruise ship (internal enclosed volume and dimensions). That does not affect the material composition of the sub-systems. DSE runs the design of experiments and comparisons by executing the parametric CO₂ emission model and visualizes various design configurations against the baseline design for decision-making purposes. The iterative design space exploration process continues until the optimal configuration is identified, supporting informed decision-making.

Figure 1. Visual depiction of the logic of the approach applied in the demonstrative case study

The approach utilizes the LCA method for the estimation of materials emission combined with the bottom-up approach that starts from the sub-system or low-level view of the systems and builds up to form the overall system, as shown on the left side of the parametric emission model (marked with number 2). The rationale behind the bottom-up approach is that it allows more freedom in making design decisions (Reference Crespi, Galstyan and LermanCrespi et al., 2005) where there is an increasing need for modular and platform-based design.

The parametric emission models take parametric variables input from the GUI and utilize it in a bottom-up approach where the starting point is the sub-systems of a larger system which adds up to the higher-level system. In the same figure, number 3 represents the input layer. It shows the type and the stage where the input has been given to the emissions parametric model.

The parametric emission model allows the quantification of scope 3 upstream emissions using LCA and a bottom-up approach. Each subsystem is represented by one or more parametric models. As the subsystems are closely linked together. There exist material, structural, and computational dependencies which are shown in the figure through bi-directional connections among models of sub-systems. Data used in the development of regression models were sourced from publicly available literature and company reports. Where no available data existed, synthetic data was generated using a large language model-based AI model. The generated data was subjected to screening and sanity checks to ensure plausibility and accuracy. However, the results would be more accurate using verified and validated company data. To ensure flexibility, the model allowed for the seamless replacement of AI-generated data-based regression models with validated and trusted models as they become available.

The output of the models is the material composition in terms of the mass of each sub-system. The material composition is fed into models that calculate the emission based on the material type and the procurement channel. The procurement channel includes the transportation mode and the geographical zone from where the material is transported to the manufacturing site. Each configuration is then stored in the database and is called into the decision support environment to be visualized and compared for design space exploration purposes. Further decisions are made to include additional configurations as needed. The overall approach allows the modeling of indirect contributions from the value chain through a specific focus on material procurement and transportation. The CO₂ emission factors associated with producing different materials were obtained from industrial and literary databases which considers the material-specific environmental impacts.

The model estimated CO₂ emissions from transporting those materials from suppliers to the shipyard. Mode of transportation (sea, rail, air, road) and geographical distance from procurement to shipbuilding site were considered. All material and transportation emissions were stored in a central repository. This dataset was then fed into a Decision-Support Environment (DSE), allowing for the visualization and comparison of different design configurations in terms of CO₂ impact.

4.2.1. Example usecase: hotel system of cruise ship

The approach presented in Figure 1, is described in this section through its application in a case study focusing on the hotel section of a cruise ship. Cruise ships typically consist of seven sub-systems, each with unique characteristics depending on how they are classified. The example use-case places particular emphasis on the hotel system, as it is one of the most material-intensive sub-systems of the cruise ship. Figure 2 highlights the sub-systems within the hotel system, each playing an important role in designing modular solutions.

Figure 2. Hierarchical breakdown of cruise ship with a focus on the hotel system components

The application of the approach began with defining the high-level boundary of the cruise ship by defining the overall size in terms of gross tonnage (GT). GT represents a nonlinear measure of a ship’s overall internal volume. Space ratio is a measure of a ship’s comfort, calculated by dividing its GT by the number of passengers onboard. It establishes a high-level boundary upon which the bottom-up approach is developed. The overall length of the cruise ship was calculated using equations proposed by (Reference Schwarzkopf, Petrik, Matthias, Quante, Majamäki and JalkanenSchwarzkopf et al., 2021) respectively as follows:

(1) $$L = \root b \of {{{GT} \over a}}$$

For cruise ships, (Reference Schwarzkopf, Petrik, Matthias, Quante, Majamäki and JalkanenSchwarzkopf et al., 2021) set a = 0.0129 m ⁻¹ and b = 2.8 in Eq. (1), and a = 24.6 m ⁻¹ and b = 3.9. To calculate the total number of decks the height of the ship is taken as input which is often dictated by constructional and operational limits. The internal enclosed volume of the cruise ship is calculated using the following equation (Reference Eyres and BruceEyres & Bruce, 2012):

(3) $$GT = (0.2 + 0.02\log _{10} V) \times V$$

Eq. (3) requires the use of numerical methods or the use of iterative techniques to approximate internal enclosed volume (V). Volume facilitates the calculation of the number of decks and the area of Main Vertical Zones (MVZs) per deck. MVZs are vertical sections separated by bulkheads along the ship’s length, with fireproof insulation between them to meet fire safety requirements stipulated by the Safety of Life at Sea (SOLAS) regulations. Each MVZ per deck represents a distinct area that is populated with cabin rooms, stairways, elevators, corridors, and service rooms, such as HVAC rooms and other service-related areas.

For the sake of simplicity, cabin rooms were categorized into four main types, although further sub-classifications exist. The model required users to input the percentage of hotel system space allocated to cabin areas and the percentage of each type of cabin per MVZ per deck as shown in Figure 3.

Figure 3. Web-based Interactive GUI DSE form for setting up various design configurations

The total number of cabins per MVZ per deck was calculated to determine the mass of cabin room components, which depends on luxury level due to material variations. Cabin components include structural panels, doors, furniture, electronics, flooring, bathroom ceramics, and plumbing fixtures. Mass estimation used kg/m², with mass proportional to the cabin area. Balcony cabins and suites included additional calculations for balustrades and doors. The model accommodated various cabin types per MVZ per deck, outputting material composition and area data.

Corridor areas were calculated based on the cabin type they serve, as more luxurious cabin types require wider corridors. The calculation of corridor areas informs the material requirements for sidewalls, flooring, and ceilings. Flooring materials consist of leveling compounds and carpet finishes, which vary according to the cabin’s luxury level. The model computed total mass, with lighting design based on illumination needs.

Stairways within the hotel system were classified into passenger stairs and service stairs. Passenger stairs facilitate passenger mobility between decks, while service stairs support crew movements and enable emergency evacuation in the event of fire or other emergencies. The model allowed the selection of different configurations of stairs based on the requirements. The material required was calculated for stairs fabrication. Area requirements were also calculated, considering the width of stair flights and landing areas. Elevators were divided into two categories: passenger elevators and service elevators. Passenger elevators were typically grouped in banks while service elevators operated alongside service stairways to facilitate crew operations and had material determined using manufacturer data.

The HVAC system, accounting for up to 16 kg/m², included equipment and air ducts, with each MVZ requiring a separate system. Mass calculations were stored for analysis. The model’s output presented material composition in kilograms, with CO₂ emissions calculated using:

(5) $$C_m = \sum\nolimits_{i = 1}^n (M_{mi} \times EF_{mi} )$$

Where C_m is the manufacturing CO₂ emission of material, M_mi is the total amount of type i material required and EF_mi is the CO₂ emission factor of type i material. Eq. (5) allowed to calculate emissions by multiplying the material mass by the emission factor for each material type. Calculation of CO₂ emission due to transportation from specific procurement channels for each material was performed using the following equation:

(6) $$C_{tr} = \sum\nolimits_{i = 1}^n (M_{mi} \times TF_i \times D_i )$$

Where C_t is the CO₂ emission due to the transportation of the material. M_mi is the total amount of type i material to be transported, TF_i is the CO₂ emission factor of the transport mode i (sea, train, road, or a combination) and D_i represents the transportation distance of the material i based on geographic location. The model revealed that transportation emissions account for approximately 5% to 8% of the total CO₂ emissions associated with the cruise ship’s hotel system, underscoring its significant contribution to overall environmental impact. The material and transportation CO₂ emission data was stored in a central repository and fed to a decision-support environment as shown in Figure 1.

In Figure 4, four different configurations were simulated. A baseline was selected having a space ratio of 33 and for a total of 6600 passengers with typical distribution of cabin. Config 2 has the same high-level parameters, yet the cabin distribution was altered, and two-thirds of the cabin rooms were selected as balcony rooms. Similarly in Config 3 and 4, the high-level parameters were the same (space ratio 50 and for 8000 passengers), the only difference was the cabin distribution. The results showed that increasing balcony cabins resulted in more CO₂ emissions as there were no balconies for ocean view and inside cabins. So, overall material utilization increased. Similarly, the DSE visually depicted the information in the Sankey diagram for CO₂ contribution of each component of the hotel cabin systems. The DSE allowed to store multiple configurations and visually compared them against a pre-selected baseline configuration.

Figure 4. CO₂ Emissions breakdown and comparison of different configurations with baseline

5. Discussion and conclusion

The primary challenge that the approach addresses, identified in the descriptive study, is the ability to perform quick parametric calculations of CO₂ emissions. The capability is particularly valuable in the early design stages, where rapid exploration of alternatives is essential in making informed design decisions. While the approach has the potential to be further generalized, the current stage of research does not yet allow broader generalization. However, it is more utilizable for passenger-carrying ships, where there is significant design freedom and a multitude of design decisions to be made. In contrast, cargo ships are often more constrained and primarily focused on structure benefiting less from this approach. Nevertheless, similar methodologies can be employed to facilitate CO₂ emissions calculations in cargo ship design.

During the development of the approach, several workshops were conducted to perform preliminary verification. Hours-long prototyping sessions at the company’s facilities were held which included participants from industrial partners, consulting firms, and researchers. Feedback was gathered on the applicability and usability of the approach. Verification efforts were focused on assessing the logic of the approach, the consistency of the models, and the dependencies between different models. Overall, the applicability of the approach was tested, while its usability was only partially validated, as it was demonstrated on the hotel system of the cruise ship. Some of the key takeaways were:

• The proposed approach advocates for the use of AI in CO₂ estimation and integration to facilitate quick sustainability assessments. While this introduces the risk of misjudgement due to unreliable data generated by AI, this practice was introduced to provide evidence of the potential of data availability in areas where data have historically not been collected, this was done to promote actions to collect real data to eventually substitute AI-generated data.

• The balance between quick simulation and accuracy. It was concluded that during the initial design stage, rapid simulations with relatively higher granularity are quite helpful and acceptance of uncertainty is acceptable because of prioritization of speed of execution and because models could be improved iteratively along the design process.

• AI is currently present and is expected to improve significantly. This will allow to have more accurate estimates for missing data in the future. Thus, reducing uncertainty in the early design process. AI as a tool, is anticipated to be more reliable for overcoming initial design inertia.

The work contributes to the broader discourse on DSS for cruise ship design by providing a more holistic framework that includes scope 3 upstream CO₂ emission. Through it, the approach emphasizes sustainability from the early stages of design, which is important as there is an ever-growing regulatory pressure to consider lifecycle emission in decision-making. Moreover, the research encourages the use of modular and platform-based design, which is becoming a key priority in the maritime industry, However, CO₂ emission integration is still a challenge. Methods and tools in other fields are needed to be adapted to ship design. A challenge in CO₂ assessment is that rarely data are available for exploring new designs quickly and efficiently.

Although the proposed approach includes the integration of engineering and economic KPIs with CO₂ emission at the same level, however, the use case is limited to an initial prescriptive analysis and does not yet integrate all KPIs into the DSE.

The work presented in the paper shall be regarded as a first step toward a more comprehensive approach for modeling and simulations to support value-driven design space exploration in the conceptual design stage of ship design.