2.1. Reactive maintenance

Reactive maintenance is generally the least-desirable maintenance option because action is taken only after failure occurs. For critical components of naval vessels, this could mean a loss of operational availability at an unexpected moment, which is a potentially unacceptable outcome. Such an unexpected failure could have a strong negative impact on vessel safety. In addition, fixing structural problems only after they occur may lead to higher maintenance costs, related to higher shipyard fees due to urgency and separated maintenance activities to fix individual issues that (might have) been instead fixed together. For small structural problems, reactive maintenance may be acceptable, as long as the structural damage is monitored, though reactive maintenance is now generally avoided as the overarching strategy for strategic structural components.

2.2. Proactive maintenance

In contrast to reactive maintenance, a proactive solution is to determine dynamic maintenance intervals that are timed for the precise moment when needed, that is, neither too early nor too late. Proactive maintenance strategies can be divided into preventative and predictive maintenance (Tinga, Reference Tinga2013; Basri et al., Reference Basri, Abdul Razak, Ab-Samat and Kamaruddin2017):

• • Preventative maintenance predetermines maintenance tasks based on the functionality or expected lifetime of a specific part/structure (Rijsdijk et al., Reference Rijsdijk, Alves da Silveira and Tinga2020), but can still be an expensive option because it is an inherently conservative approach. Optimal maintenance intervals may be determined based on vessel usage profiles (Tinga et al., Reference Tinga, Wubben, Tiddens, Wortmann and Gaalman2021), measured loads (Tinga, Reference Tinga2010), or the overall monitored condition (Tinga and Loendersloot, Reference Tinga, Loendersloot, Lughofer and Sayed-Mouchaweh2019). Preventative maintenance activities may be performed for naval vessels if they are already in dry dock to minimize the chance of later failures while on deployment. In this case, the conservatism of preventative maintenance may be useful for maintaining structural integrity and mission readiness.

• • Predictive maintenance advances preventative maintenance by utilizing various CM systems (Tinga et al., Reference Tinga, Tiddens, Amoiralis, Politis, Cepin and Bris2017) to not only assess the present condition but also predict the remaining time to a failure. By examining the current and future system state, the individual component functions, and the relative importance (the associated risk of failure and level of resulting consequence), CM can help identify issues before they become major problems, helping to avoid time-based replacement or maintenance (Ross, Reference Ross2002). This is the most forward-thinking approach and therefore the most attractive for considering the future readiness of naval vessels.

Selection of the most appropriate maintenance strategy depends on the ambition level and the availability of knowledge, sensors, and data (feasibility) (Tiddens et al., Reference Tiddens, Braaksma and Tinga2023; da Silveira et al., Reference da Silveira, Meghoe and Tinga2024).

2.3. Condition-based maintenance

CBM is a proactive and predictive strategy where maintenance activities are triggered based on the current or future expected status of the structure or structural parts, thereby reducing the risk of unexpected failures while minimizing the need for (potentially unnecessary) repairs performed during planned maintenance intervals. CBM assesses the condition of the hull structure by combining design information, modeling approaches, sensed data about the structural state and loading condition, and reliability assessments of the structure to ensure structural integrity and therefore safety (Barltrop et al., Reference Barltrop, Hu and Hifi2010; Emmett et al., Reference Emmett, Churchman and Hooley2011). Including risk-based approaches (Yang et al., Reference Yang, Khan, Lye, Sulistiyono, Dolny and Oldford2013; Kujala et al., Reference Kujala, Goerlandt, Way, Smith, Yang, Khan and Veitch2019; Stambaugh, Reference Stambaugh2020) can optimise the design and operation of onboard sensor systems to support CM-based AIM (Chen et al., Reference Chen, Reniers, Khakzad and Yang2021). Risk-based maintenance scheduling for naval vessels can also help maintain vessel availability while reducing maintenance costs (Cullum et al., Reference Cullum, Binns, Lonsdale, Abbassi and Garaniya2018).

2.4. Planning strategies for maintenance activities

Maintenance activities for the naval vessel need to eventually be scheduled at a system level, after analyzing the condition of individual subsystems or components. Here different optimisation schemes can be applied, again based on the overall maintenance goals, such as:

• • Dynamic planning: grouping maintenance activities to save on set-up costs (Wildeman et al., Reference Wildeman, Dekker and Smit1997; Tinga and Janssen, Reference Tinga and Janssen2013; de Jonge et al., Reference de Jonge, Klingenberg, Teunter and Tinga2016),

• • Cost-based planning: comparing the cost of a solution versus the revenue the solution generates (Turkcan et al., Reference Turkcan, Akturk and Storer2007),

• • Time-based planning: looking at duration of operation, time for repair, time lost because of failure, based on, for example, dynamic programming (Dekker, Reference Dekker1995),

• • Failure-based planning: reducing failure probability by considering system deterioration, CM (de Jonge et al., Reference de Jonge, Teunter and Tinga2017), and reliability models.

Similar to the questions defining the purpose of the CBM system and the different points of interest for different stakeholders, the prioritized planning strategy will differ depending on the stakeholder and the current mission status. Again, this makes a distinction between commercial and naval vessels. For commercial vessels, there may be a different owner and operator, each with different priorities (e.g., maintaining sufficient condition for the vessel to have the maximum service life and possible re-sale value vs. maximizing operability while staying within design limits). For naval vessels, the owner is the operator, and the most important thing is mission readiness. In that sense, failure-based planning for naval CBM is the most logical choice to ensure that the vessels maintain sufficient structural integrity.

2.5. Aims for a CBM system for naval vessels

CBM is a strategy which determines the optimal maintenance intervals to maintain the needed equipment at the right time. To accomplish this, CBM systems consist of sensing hardware, data analysis procedures, data/analysis architecture, and presentation. A CBM system can be used to reduce maintenance costs, increase operational availability, increase vessel safety, optimise operational performance, and possibly extend a vessel’s service lifetime. However, some of these aims may come at the expense of others. For example, CBM can aid in service life extension, though it is important to consider that ships used past their original design life likely require more maintenance activities, possibly increasing the lifetime maintenance costs. In such cases, trade-offs will have to be considered to determine the optimal outcome. As an example, for the service life extension versus maintenance cost reduction challenge, this trade-off can be considered using a cost–benefit ratio and a net present value to optimise the duration for a service life extension (Liu et al., Reference Liu, Yang and Frangopol2021).

Naval vessels are operated with a long-term vision, informed by short-term mission needs. A naval CBM system should accommodate both perspectives, that is, allow for and support both short-term and long-term maintenance decisions. A particular aim for naval vessels is to ensure that the vessel maintains structural integrity, especially during an intense mission, so systems which indicate the short-term risk of damage and the resulting long-term fatigue damage trends are useful (Phelps and Morris, Reference Phelps and Morris2013). Vessels will almost always have some level of damage, but some damage like initial cracks may be small enough (and above the waterline) to not pose an immediate threat. But what would be problematic is if such a crack is hidden (e.g., behind insulation) and unknown to the crew. During a mission, possibly in rough weather, it would be very important to monitor whether that small crack grows significantly. An onboard global monitoring system which could indicate the start of such a crack, and therefore suggest either a more targeted monitoring campaign or visual inspection, would be therefore very useful. That is one of the motivations driving the envisioned baseline CM system introduced in Section 10. In addition, from the long-term perspective, keeping track of when fatigue life is consumed in the naval vessel lifecycle via monitoring can be useful for making later decisions about lifetime extension, especially when considering how a vessel contributes to fleet operations and missions. Extending maintenance intervals is often feasible, possibly leading to large cost reductions, but this is only possible when the condition can be assessed.

2.6. CBM system architecture definition

Achieving the CBM strategy requires monitoring and assessing the vessel structural health based on sensed and analyzed data, coupled with some decision framework to determine when maintenance activities should be triggered. This whole system should be contained within a digital framework which manages and controls the data flow, stores historical data, and manages the system configuration. Then, there should be a physical human–system interface to share the resulting information with people both onboard and onshore.

To practically work onboard naval vessels, this entire process needs to be organized within a coherent framework. As a possible framework, we propose the Machinery Information Management Open System Alliance (MIMOSA) Open System Architecture for Condition-Based Maintenance (OSA-CBM), a standard architecture specification for CBM (MIMOSA, 2010) and an integration of the functional specification of ISO-13374, Condition Monitoring and Diagnostics of Machines (ISO, 2025). The MIMOSA OSA-CBM framework defines six functional CBM blocks and provides an interface between these blocks, defining a “standardized information delivery system for condition based monitoring.” This CBM framework incorporates different layers of increasing complexity, moving from data acquisition to analysis and modeling and to diagnostic and prognostic analysis, which enables decision-support guidance based on vessel structural health, organized via the MIMOSA framework (Thurston and Lebold, Reference Thurston and Lebold2001):

1. 1. Data acquisition: data may come from sensor modules which measure onboard data or may be retrieved from public data sources or simulation databases, etc., which is then digitized.

2. 2. Data manipulation: takes input from the sensor modules and performs signal processing, such as transformations/feature extractions, data fusion, normalization, cleansing, and compression, leading to outputs such as digitally filtered sensor data, frequency spectra, virtual sensor signals, or condition features.

3. 3. State detection (condition monitoring): takes input from sensor and signal processing modules and compares condition features with expected/allowable values or operational limits. This may output some condition indicators. Here, data from the sensor and signal processing modules may be archived.

4. 4. Health (diagnostic) assessment: takes input from the CM system to determine the current health of the monitored system, based on the health history, operational status, loading history, and maintenance history. If the system health is degraded, some diagnostic record may be generated and output.

5. 5. Prognostic assessment: takes input potentially from all previous levels and projects current health into the future, based on estimates of the current health and future usage profile. Here, there may be an estimate of the remaining design life of the system. At this level, there may also be an archive of the utilized historical data.

6. 6. Advisory generation (decision support): takes input from the health assessment and prognostic layers to recommend actions (e.g., maintenance schedule, change of operations) based on the operational history (past usage and maintenance profiles), current and future operational profiles, overall system objectives, and resource constraints, and displays overall status levels (health/prognostic assessment, decision support recommendations, etc.).

We follow this framework structure in the remainder of the paper, discussing both the general concept, the foundational and state-of-the-art research supporting such blocks, and the research further needed to facilitate onboard implementation. To link these modules, we point to the MIMOSA OSA-CBM framework as a possible data architecture system.

4.1. Digital twin

Given the data from the sensors, processed through the DAcS and DAnS, a digital twin (DT) can analyze the sensed quantities to assess the vessel structural condition and health. A DT can, in general, have three pillars (Oracle, 2023):

• • Virtual twin: A virtual representation of a physical asset or a device.

• • Predictive twin: Physics-based, data-driven, or hybrid models operating on the virtual twin to predict the behaviour of the physical asset.

• • Twin projection: Integration of insights generated by the predictive twin into the business operation and processes.

The three pillars above are all necessary for implementing CBM, but simplifications are necessary to practically achieve such twins. At the virtual level, the preference would be for a realistic model of the as-built system, as opposed to the as-designed one. Considering the many uncertainties in the manufacturing and fabrication process, the as-designed system is an idealized system, while the true structural condition depends on the initial as-built state, which, in general, is never the same as the designed state (Lazakis et al., Reference Lazakis, Leira, Chen, Drumond, Lee, Jurisic, Liu, Mondoro, Pahlavan, Shi, Song, Sugimura, Jochum and Coppola2022). It is possible to define the as-built state using various NDT techniques, such as ultrasonic testing, guided waves, magnetic flux leakage, eddy current testing, or vibration-based testing (Solla and Riveiro, Reference Solla and Riveiro2016; Amirafshari et al., Reference Amirafshari, Barltrop, Bharadwaj, Wright and Oterkus2018); a detailed review on ship hull NDT is offered in Lin and Dong (Reference Lin and Dong2023).

However, building a perfect as-built virtual twin could take unreasonable effort in initial inspections and NDT. At the predictive level, the virtual twin should be combined with sensed data about the vessel state and excitation environment; for naval vessels, where operational readiness is the top priority, continuous monitoring via onboard sensors is preferred to inspections at regular intervals. These sensed data can then be implemented in a hybrid analysis and modeling framework which combines reduced-order physics-based and data-driven methods to update the predicted vessel state. While there will always be uncertainty surrounding sensed data, including possible bias, faults, and non-sensed locations, data-driven approaches are becoming more standard in analyzing structural health data. Model-based approaches are more likely to be co-implemented in hybrid schemes, rather than focusing solely on high-fidelity model-based approaches. Such a high-fidelity approach is extremely useful for detailed and accurate understanding of the system behaviour, but it comes at a high computational cost. For short-term decisions about maintenance for naval vessels, where local sensing feeds into local analysis for near real-time alerts and guidance, such high-fidelity modeling is not fully justified. This high-fidelity modeling may instead be used for onshore analysis of the vessel to make long-term maintenance decisions. Finally, the twin projection would be needed to extrapolate the vessel state to the future for making fleet maintenance choices. In that sense, we adopt the DT definition for the purpose of naval structural CBM defined by Wincott and Collette (Reference Wincott and Collette2019): a physical system of interest, one or more numerical models of the system, data sensors/fusion logic to link the model and system, and one or more decisions based on the twin’s prognostic output. For onboard, short-term decision making, simple physical models can be reasonably utilized, while longer-term decisions will require prognostics informed by more complex physical models.

DTs at different levels of detail and fidelity are already in use in the maritime field for tracking vessel performance and efficiency, voyage planning, and engine performance and emissions. A DT can be used over a ship’s lifecycle (Mauro and Kana, Reference Mauro and Kana2023) to link with risk assessments and emergency management procedures (Zio and Miqueles, Reference Zio and Miqueles2024) and can further incorporate the different stakeholders in the process, namely, shipyards, ship owners/operators, and maritime classification societies (Van Os J, 2018). A DT combining dynamic sensor data and a physical model was used to examine whole-life fatigue crack growth prognosis and validated via lab experiments (Ye et al., Reference Ye, Yang, Zhang, Meng and Wang2023). Though not developed for ship-specific structures, such a method offers interesting possibilities for naval DTs, especially as their DT was able to capture the accelerated crack growth and degradation of reliability toward fracture. Fatigue crack growth diagnosis and prognosis were also considered in a probabilistic DT in Leser et al. (Reference Leser, Warner, Leser, Bomarito, Newman and Hochhalter2020).

4.2. Structural response reconstruction in the DT

For CBM, a structural DT is of particular interest to simulate the loading experienced by the vessel and the resulting strains and displacements and infer the current structural condition. Here, we follow a strain-based load estimation process, and in this case, it is necessary to reconstruct stress fields over the whole hull structure, even when only informed by a limited number of sensed locations. The vessel structural model can then be used to convert these measured strains into internal loads. The structural model for this stress-load conversion can range in fidelity from simple beam models (a representative longitudinal beam with the ship’s global properties) to a detailed global model of the ship structure. The simple beam approach is computationally efficient and can evaluate overall responses like global bending and shear, but it cannot resolve local effects or torsional loads. The detailed model can resolve global and local effects, but it will be very computationally expensive, making it unfeasible for real-time or long-duration estimations. A combined approach can be employed, where the simple model can be used for global responses and detailed models are then used to examine local behaviour in specific areas of interest.

Originally, inverse FEM (iFEM) was implemented to estimate real-time strains, stresses, and displacements based on a finite number of strain measurements (Maniatty et al., Reference Maniatty, Zabaras and Stelson1989; Tessler and Spangler, Reference Tessler and Spangler2003; Tessler and Spangler, Reference Tessler and Spangler2005). The iFEM approach is useful in the naval context where it can be difficult to measure external loads directly. When considering load reconstructions for a ship structural DT, a combination of using direct structural calculations based on sea state reconstructions was suggested for short-term statistical assessments and iFEM for real-time response assessments (Fujikubo et al., Reference Fujikubo, Okada, Murayama, Houtani, Osawa, Iijima, Hamada, Toh, Oka, Hirakawa, Shibata, Ashida, Arima, Tanaka, Tatsumi, Takeuchi, Mitsuyuki, Mikami, Kobayashi, Komoriyama, Ma, Chen, Ochi and Miratsu2024). Matrix conversion methods can also be used to recreate the full-field global structural response from a limited number of strain measurements (De Lauzon et al., Reference De Lauzon, Benhamou and Andoniu2021; Thompson et al., Reference Thompson, Jiang and Benhamou2021; Sireta et al., Reference Sireta, van der Cammen and Storhaug2022; Sireta and Storhaug, Reference Sireta and Storhaug2022; Jagite et al., Reference Jagite, Sireta, Zilakos, Sagvolden, Hareide and Brubak2025). The reconstructions from iFEM are also useful as training data for AI models, which is the focus of much current research. A deep learning generative adversarial network was suggested for this purpose by Sun et al. (Reference Sun, Chen, Yi and Li2024) and Sun and Chen (Reference Sun and Chen2024). Modal superposition can also be used, as recently considered for time-domain stress reconstructions by Jing et al. (Reference Jing, Yuan, Guo and Tang2025). In Liu and Ren (Reference Liu and Ren2022), a coherent scheme was described to construct a vessel structural DT and incorporate measured onboard stresses. The authors later suggested a rapid acquisition method to evaluate the structural strength of the ship DT, allowing the evaluation of stresses at non-sensed locations (Liu and Ren, Reference Liu and Ren2023).

The external hydrodynamic (wave) loading on the vessel can also be reconstructed for use in hydrodynamic or structural simulations. Here, the exciting wave environment can be directly, inversely, or virtually estimated; this topic is discussed further in Section 5.4.

5.1. Fatigue

Due to the cyclic wave loading on a vessel, cracks due to fatigue may be initiated. Fatigue failure occurs in a series of stages: crack initiation, stable crack growth until a critical length, then the initiation of unstable crack growth, potentially leading to fracture. An efficient combination of monitoring can be made by combining global load measurements from the start of the ship’s lifetime (e.g., via the baseline sensor system later described in Section 10) with dedicated measurements in a more localized area when indicated; such localized measurements may be done using the methods described in the following sub-section. The former set of measurements will be able to provide an overview of the historic loads acting on the vessel. In addition, it might be possible to indicate specific areas of the hull structure which have seen increased loads to indicate further investigation. When the overall loading suggests that crack initiation and growth are imminent, dedicated monitoring of the initiation and propagation (i.e., before the onset of unstable crack growth) should start. As thoroughly described in the review by Bjørheim et al. (Reference Bjørheim, Siriwardane and Pavlou2022), NDTs to monitor fatigue can be separated into the categories of fatigue damage and fatigue crack monitoring, as discussed below.

5.2. Corrosion

Corrosion of the vessel structure can cause material loss of the hull, reducing the vessel’s strength or fatigue properties, for example, by generating corrosion pits that act as crack initiation sites. Therefore, corrosion, in general, increases the likelihood of structural failure. In addition, corrosion can also directly result in functional failure. Microbiologically Influenced Corrosion (MIC) pitting can compromise ballast tanks and cause leakage. It should be noted here that the root cause of corrosion does not necessarily have to be electrochemical: degradation of a protective coating system due to, for example, UV radiation or loss of adhesion can also lead to corrosion and subsequent mechanical failure. These parameters can act as reliable predictors for corrosion-related problems.

There are many types of corrosion, three of which are of interest for this work. General corrosion (often referred to as uniform corrosion, although it is not uniform) is mostly spread over the exposed surface, and the corrosion process over this surface occurs mostly at the same rate. In contrast, pitting corrosion is localized to a very small area, typically in the micrometer to millimeter range, resulting in cavities. MIC is affected by the activity of microorganisms, depending on the temperature and pH of the surrounding environment. Climatic conditions have a strong influence on the corrosion behaviour of marine structures and should also be considered when defining maintenance strategies (Abbas and Shafiee, Reference Abbas and Shafiee2020). The loss of structural integrity due to corrosion, for example, due to the presence of deep corrosion pits at critical locations, or a significant loss of wall thickness for structural parts is the final phase of the corrosion process. At that stage, corrective actions are drastic and can be very expensive, as typically parts of the structure need to be replaced and the ship safety is compromised. Accordingly, maintenance costs related to corrosion make up approximately one-fifth of the total maintenance costs of a ship (Herzberg et al., Reference Herzberg, Change, Daniels and O’Meara2012).

As maritime systems are so prone to corrosion, significant effort is additionally put into corrosion coating management, corrosion prevention, and mitigation. This can be achieved by inspecting the structure periodically or by ensuring that corrosion processes cannot take place at all (or occur at a reduced rate). Preventing corrosion processes from taking place at all can be achieved by corrosion management methods like cathodic protection and application of protective coatings. These methods should also be monitored over time to ensure they still work properly and to prevent the structure reaching the final phase of severe corrosion damage. Periodic inspections can be done visually but are preferably performed by CM systems.

5.3. Extreme load monitoring

Apart from the cyclic wave loading leading to fatigue and the compounding effect of corrosion damage, extreme wave loading can also lead to structural damage. The wave bending moment is a major contributor to the extreme bending moment for naval vessels (Brown, Reference Brown1999), and extreme wave impacts can lead to large whipping loads (extreme hull girder vibrations initiated by slamming wave impacts), especially for long slender vessels (Storhaug, Reference Storhaug2014). This wave impact-induced whipping can significantly increase the number of stress cycles and therefore strongly impact the resulting fatigue life of the vessel. Whipping responses can also lead to significant increases to the total midship bending moment of a vessel in a specific operating condition (Kim et al., Reference Kim, Engle and Troesch2011) and can subsequently have a strong impact on the fatigue life of a vessel, as noted by Kahl et al. (Reference Kahl, Fricke, Paetzold and von Selle2014) and Storhaug (Reference Storhaug2014, Reference Storhaug2007). However, when vessels operate primarily in moderate sea states, and especially for smaller and/or more stiff vessels, the contribution of whipping to the total fatigue accumulation may be limited. For example, it was found that whipping loads for a frigate type hull do not significantly contribute to the accumulating fatigue damage for small wave heights (less than 4 m) (Drummen et al., Reference Drummen, Schiere, Dallinga, Thornhill and Stambaugh2014). These caveats further motivate the need for environment monitoring to determine whether the vessel has encountered these whipping-prone sea states.

For high-speed vessels, slamming loads (on the bottom for monohulls and wetdeck for multi-hulls) can become significant. Slam identification using empirical mode decomposition was carried out on the HSV-2 SWIFT, a former high-speed catamaran ferry converted for use by the US Navy, by Alsalah et al. (Reference Alsalah, Holloway, Mousavi and Lavroff2021). In this monitoring approach, strain gauges at midships and bow accelerometers were used to identify slam occurrences and subsequent whipping responses. Slamming was also monitored on a different wave-piercing catamaran using strain gauges, accelerometers, a MRU, GPS, and US sensor paired with ML techniques by Shabani et al. (Reference Shabani, Jason Ali-Lavroff, Spiridon Penev and Thomas2021), though more focused for seakeeping performance and passenger comfort. Such an ML approach was also taken by Dessi et al. (Reference Dessi, Sanchez-Alayo, Shabani and Ali-Lavroff2023) to monitor and identify bottom slamming on a high-speed monohull, though this time in a model-scale experiment. Alternatively, slamming was monitored on a polar research vessel using an array of accelerometers at the waterline by Bossau and Bekker (Reference Bossau and Bekker2022). With the addition of strain gauges to the accelerometer array, accumulated extreme loading damage due to whipping was monitored by Pferdekämper and Bekker (Reference Pferdekämper and Bekker2024). Here, the use of OMA indicated that slamming-induced whipping led to increased excitation of the flexural bending modes of the vessel, increasing the fatigue damage. It was noted that the accelerometer array allows identification of the wave impacts and better understanding of the global hull deformation.

5.4. Environment (wave) monitoring

Environmental monitoring is important for both short-term decisions, to determine how rough the seas are and the resulting vessel capability in possibly degraded status, and long-term decisions by comparing the designed and the experienced environmental conditions to evaluate fatigue usage. In both cases, it is necessary to estimate the environmental parameters acting on the system, most importantly the encountered wave energy density spectrum. In general, information about the wave environment encountered by the ship is essential for analysis of the service life maintenance and sustainment efforts, since knowing the encountered wave energy density spectrum allows adjustment in fatigue life estimations when compared to the designed environment. To monitor the excitation environment consists of sensing parameters associated with the wind, wave, and current loading, where the wave loading is the most relevant for structural CBM. These parameters feed into the DAcS and DAnS described above. It is important to note that these environmental parameters are estimated from indirect measurements and are usually not directly sensed or measured, which introduces new challenges. There are multiple ways to determine the exciting wave environment, such as by direct estimation (wave radar), inverse estimation (via measured ship responses), and virtual estimation (hindcast), as discussed further below.

6.1. Role of ML and AI

ML and AI are also proving very useful by helping in signal and feature classification for different sensing methods, eventually linking monitored data with diagnostic markers. It is particularly envisioned that continuing advances in ML/AI will enable local monitoring techniques like US and AE to become more reliable for onboard damage detection and classification (Yang et al., Reference Yang, Yang, Tian, Deng, Hu, Ma, Gao, Zhang, Ma, Yang, Xu and Wu2023). While AE has been used by the petrochemical industry for many years and shows strong potential in experimental investigations for marine applications, there are still challenges to be solved for further real-world application (Laflamme et al., Reference Laflamme, Ubertini, Di Matteo, Pirrotta, Perry, Fu, Li, Wang, Hoang, Glisic, Bond, Pereira, Shu, Loh, Wang, Ding, Wang, Yu, Han, Goldfeld, Ryu, Napolitano, Moreu, Giardina and Milillo2023), especially relating to signal classification in the presence of noise and the transmission of AE signals through complicated ship structure geometries. An ongoing challenge with AE in the marine environment is signal processing for classification, something ML is well-suited for. To this end, a deep ML procedure for identification of corrosion and fatigue damage in ship structures based on AE was proposed by Karvelis et al. (Reference Karvelis, Georgoulas and Stylios2021). The authors employed feature selection first by applying a Discrete Cosine Transform on the AE signals and then reducing the signal dimension by Piecewise Aggregate Approximation. These signals were then fed into a Deep Belief Network trained on historical data for damage localization. The authors noted that for complicated ship structural geometries, such deep learning will likely be needed for AE to accurately localize damage.

Using strain as inputs, Katsoudas et al. (Reference Katsoudas, Silionis and Anyfantis2023) evaluated different classification algorithms (statistical models plus an artificial neural network) to determine which worked best for corrosion induced thickness loss. The authors found that the binary-classifier artificial neural network performed the best with the minimum amount of information, and that in general all classifiers did better for classifying local pitting corrosion than for global uniform corrosion. Deep learning has further been identified to detect and monitor material degradation (Nash et al., Reference Nash, Drummond and Birbilis2018) and can be used in combination with EN to characterise the type of process (Homborg et al., Reference Homborg, Mol and Tinga2024). Major damages, resulting in large load re-distributions across the structure, were considered by Silionis and Anyfantis (Reference Silionis and Anyfantis2022) as a statistical pattern recognition problem. The resulting damage identification framework offers a first step toward identifying different damage modes (both fatigue and corrosion) leading to structural degradation. The detection of cracks and their propagation, along with localization and sizing of cracks, were considered using a convolutional neural network and stress data by Zhu et al. (Reference Zhu, Xu, Yue, Chen, Sun, Wang, Ao and Gan2025), offering a direct example of damage diagnosis in ship structures.

7.1. Role of ML and AI

ML and AI also have a role to play in prognostic assessments, though these are more recent advances. It has historically been noted that data-driven methods are better suited for diagnostics than prognostics, since data-driven methods cannot reliably extrapolate outside of their data range (Farrar and Worden, Reference Farrar and Worden2012). Hybrid models are useful to exploit the value of CM capabilities while maintaining a link to physical (and therefore explainable) processes. Different levels of physical fidelity will then be needed based on the timeline of the damage analysis. This is especially relevant for the use of DBNs, where simple physical models can suffice for short-term prognostics, but not for longer-term damage growth and decision making. As such, Bayesian frameworks combined with neural networks have been proposed as diagnostic tools for structural health monitoring when physics-based models are not available (Cadini et al., Reference Cadini, Sbarufatti, Corbetta, Cancelliere and Giglio2019). The balance between trusting a model (of varying fidelity) and data streams has also been considered for damage prognosis by using a Kalman filter to solve the recursive Bayesian estimation problem for sizing a fatigue crack, using AE measurements as input (Rabiei and Modarres, Reference Rabiei and Modarres2013). Data-driven fatigue crack prognosis is examined by Ye et al. (Reference Ye, Yang, Zhang, Meng and Wang2023) in a DBN where model parameters are updated based on sensed data to reduce model uncertainty. The authors were able to model and predict the accelerated fatigue crack growth stage up to fracture by additionally considering loading uncertainty. The loading in the prognosis analysis was based on corrected structural modes and strain input data, reducing the load uncertainty and allowing a longer-term prognosis. Such a result confirms that prognosis is possible, but is only reliable given high-fidelity models and loading definition.

8.1. Short-term onboard versus long-term onshore decision framework

Short-term decisions for naval vessels are mainly related to determining if a vessel can complete its mission, even with known structural damage. As noted above, DBNs with simple physical models can be useful for short-term prognostics and therefore short-term decision making (Zhang and Collette, Reference Zhang and Collette2022). Practically speaking, the difference in the analysis between the short-term and long-term decision making will be based on the fidelity of the physical models, the defined decision intervals, and the damage definition (growth of a single crack vs. the remaining fatigue lifetime of the ship). Short-term decisions will be taken by crew onboard, reacting to locally sensed and analyzed data about individual damages which may affect mission capabilities. For these decisions, a damage-tolerant mission planning view can be taken, as illustrated by Karve et al. (Reference Karve, Guo, Kapusuzoglu, Mahadevan and Haile2020). There, the authors use a model-based Bayesian damage diagnosis model paired with probabilistic damage prognosis to eventually support intelligent mission planning to minimize growth of existing damage. This could be a very relevant approach for dealing with existing damage during on-going missions, but it may require high-fidelity models to forecast up to critical damage states.

Longer-term decisions concerning the remaining useful life of the vessel and the connection with inspection and maintenance decisions will be performed by engineers onshore analyzing measured data and comparing the experienced and designed environments. Generally, a direct “data-to-decision” chain is found in decision-oriented structural health monitoring literature. Such approaches are increasingly based on the value of information (Silionis and Anyfantis, Reference Silionis and Anyfantis2025), value of structural health monitoring information (Kamariotis et al., Reference Kamariotis, Chatzi and Straub2023, Reference Kamariotis, Chatzi, Straub, Dervilis, Goebel, Hughes, Lombaert, Papadimitriou, Papakonstantinou, Pozzi, Todd and Worden2024), dynamic Bayesian networks (Zhang et al., Reference Zhang, Qin, Lu, Liu and Faber2023), and (partially observable) Markov decision processes because these concepts and frameworks inherently connect sensed information to sequential decisions related to inspection, maintenance, and operations. Markov Decision Processes and Partially Observable Markov Decision Processes can help in the global optimisation needed for CBM planning. To deal with such a complex system like a ship, deep reinforcement learning is being utilized with Partially Observable Markov Decision Processes to optimise integrity management planning given many dimensions and constraints (Morato et al., Reference Morato, Andriotis, Papakonstantinou and Rigo2023).

A good example of linking digital twin-enabled virtual monitoring data to inspection-timing decisions can be found in Li and Brennan (Reference Li and Brennan2024). The approach illustrates how a CM data pipeline, specifically from a virtual monitoring system, can feed into data-driven forecasting, fatigue reliability, and finally inspection planning, making a clear connection from the digital “data” domain back to the physical vessel. Including physically monitored data in such a scheme can further help reduce uncertainty and conservatism, thereby elongating inspection intervals, as noted by Hageman et al. (Reference Hageman, van der Meulen, Rouhan and Kaminski2022). Another DT approach for fatigue damage monitoring and prognosis was offered in VanDerHorn et al. (Reference VanDerHorn, Wang and Mahadevan2022) using a fully virtual monitoring system. Uncertainties within the fatigue damage prediction were still to be further addressed, such as the finite element and hydrodynamic models and SN curve parameters, and the link from fatigue prediction to an explicit decision model was not fully formalized, but the model nonetheless was demonstrated on an in-use containership.

A probabilistic framework for risk-informed decision making for vessel maintenance due to fatigue damage was presented in Dong and Frangopol (Reference Dong and Frangopol2016), where the risk for both inspected and uninspected fatigue details was examined. Utility theory was employed by Sabatino and Frangopol (Reference Sabatino and Frangopol2017) to link risk allowances with CM to develop a cost-effective CM program, applied to the HSV-2-Swift naval vessel discussed previously in Section 5.3 about extreme load monitoring. Life cycle performance of a vessel relating to reliability and system redundancy were considered by Okasha et al. (Reference Okasha, Frangopol and Decò2010), where CM was employed within Bayesian updating of the lifetime reliability assessment. Both the effects of corrosion and fatigue damage on vessel structural integrity, plus the inclusion of random extreme loads, were examined by Akpan et al. (Reference Akpan, Koko, Ayyub and Dunbar2002); such risk assessments are integral in the decision frameworks employed within structural CBM. The joint effects of corrosion and fatigue were also examined by Dong and Frangopol (Reference Dong and Frangopol2015) to determine risk-based optimum maintenance intervals, though not informed by CM but by probabilistic assessments of damage formation and growth. Evaluation of structural status by either inspection or CM was considered by Han et al. (Reference Han, Yang and Frangopol2019) to develop a probabilistic life cycle maintenance framework to manage ship structural damage due to fatigue and corrosion. A structured decision making process has also been introduced to make maintenance plans and “optimal decisions” in the face of uncertainty related to the vessel loading and structural response (Nichols et al., Reference Nichols, Fackler, Pacifici, Murphy and Nichols2014), leading to extensions of vessel fatigue life. Vessel fatigue lifetime has been estimated based on in-service measurements combined with maintenance data to allow for informed decision making (Magoga et al., Reference Magoga, Aksu, Cannon, Ojeda and Thomas2019), and calculation procedures have been suggested to include stress monitoring in real-time vessel decision support (Nielsen et al., Reference Nielsen, Jensen, Pedersen and Ito2011). As another decision framework, Risk and Total Ownership Cost was examined by Stambaugh (Reference Stambaugh2020) to compare the possible risk due to the failure modes of fatigue cracking and corrosion.

9.1. Royal Norwegian Navy

The Royal Norwegian Navy was an early adopter of CM systems through research programs running at the Norwegian Defence Research Establishment from the mid-1990s to about 2010, focusing on CM of the new Skjold and Alta/Oksøy class surface effect ships (Pran et al., Reference Pran, Johnson, Jensen, Hegstad, Sagvolden, Farsund, Chang, Malsawma, Wang, Claus, William and Spillman2000, Reference Pran, Havsgård, Sagvolden, Farsund and Wang2002; Sagvolden et al., Reference Sagvolden, Pran, Vines, Torkildsen and Wang2002; Torkildsen et al., Reference Torkildsen, Grovlen, Skaugen, Wang, Jensen, Pran and Sagvolden2005).

The Skjold series corvette was designed based on data collected on a pre-series vessel during sea-trials in 1999 using the CM system. The system was based on 20 FBG sensors distributed over four fibers, each fiber with either three or seven sensors (Jensen et al., Reference Jensen, Taby, Pran, Sagvolden and Wang2001), as shown in Figure 1. These sensors were placed using the hull FE model for optimal measurement of the major hull girder loads, and additional sensors were placed to monitor key local loads. Sensors A, B, C, and D on the midship section, as shown in Figure 2, were used to estimate the vessel’s hogging/sagging bending, horizontal bending, torsion, vertical shear force, and longitudinal compression force. Sensor arrays E and F monitored split moments between the twin hulls and rosettes at the wetdeck (locations G, J, and K) were meant to measure wetdeck slamming (Pran et al., Reference Pran, Johnson, Jensen, Hegstad, Sagvolden, Farsund, Chang, Malsawma, Wang, Claus, William and Spillman2000). During trials, the vessel was also outfitted with a Seatex MRU (Kongsberg seatex MRU-6 motion sensor, 2025), MirosSM048 “Seabat” RadarAltimeter (Miros, 2025), and Trimble DSMPro DGPS (Trimble, 2025).

Figure 1.

Sketch of FBG sensor placement on Skjold (Wang et al., Reference Wang, Pran, Sagvolden, Havsgård, Jensen, Johnson and Vohra2001).

Figure 2.

Midships cross-section of Skjold, with sensor placement (Wang et al., Reference Wang, Pran, Sagvolden, Havsgård, Jensen, Johnson and Vohra2001).

The system was further developed for the Alta/Øksøy class of mine countermeasure vessels with a view to providing risk-based operator guidance and CBM tools. Using load distributions and the recent load spectrum, the operator was provided guidance on the current overloading risk with alarm and warning levels set by the ship owner’s policy. In addition, a data-driven tool based on vessel motions and fleet history provided additional guidance parameters. The onshore CBM tool focused on providing information on load histories and local/global overloading events, which helped guide the inspection of the sandwich hull. A prototype of the system was installed in 2008 and has been in operation on the series since 2013.

9.2. United States Coast Guard

The United States Coast Guard (USCG) has adopted a fleet-wide CM program for the 154 ft Sentinel-class fast response cutters, using a system consisting of three unidirectional strain gauges and two to three multi-directional gauges (rosettes). The systems are in operation on all 50 vessels in active service, providing information on fatigue development and load spectra in the form of an automatically generated weekly data transfer, containing a pdf report as well as summary data files.

Other monitoring campaigns have been done on the USCG vessels based on their “Fatigue Life Assessment Program” (Stambaugh et al., Reference Stambaugh, Drummen, Cleary, Sheinberg and Kaminski2014). As part of the project, a hull structure monitoring system was installed on board two USCG National Security Cutters (NSC) (Drummen et al., Reference Drummen, Schiere, Dallinga, Thornhill and Stambaugh2014; Hageman et al., Reference Hageman, Drummen, Stambaugh, Dupau, Herel, Derbanne, Schiere, Shin and Kim2014; Stambaugh et al., Reference Stambaugh, Drummen, Cleary, Sheinberg and Kaminski2014; Drummen et al., Reference Drummen, Rogers, Benhamou, Hageman and Stambaugh2019). The system was designed as the minimum system to quantify fatigue life and extreme loads of the hull structure of this class and support long-term structural maintenance decisions. The instrumentation consisted of three LBSGs, three unidirectional strain gauges, one MRU, and one DGPS. Measurements from the three LBSGs were used to infer the vertical and horizontal bending moments. The overall position of the LBSGs and local strain gauges is presented in Figure 3. The number and positions of LBSGs were optimised using hydro-structural software.

Figure 3.

Locations of the strain gauges on NSC (Drummen et al., Reference Drummen, Rogers, Benhamou, Hageman and Stambaugh2019).

9.3. Sealift service containership equivalent

Naval sealift services, although not at the forefront of warfare operations, are of vital importance to their support. Vessels dedicated to this mission are varied in types, sizes, and functions; many navies use civil merchant fleet services alongside their own sealift fleet for cost-optimised operations when risk of attack is not an issue. Large ships (200–300 m in length) such as dry cargo, tankers, and containerships work in the sealift fleet to perform prepositioning and seabasing. A notable example of this is the vessel MV Maj. Bernard F. Fisher (T-AK-4396), a containership operated by the U.S. Military Sealift Command. Equivalently, the authors had access to a CM campaign of a 260 m ship of the same type (and therefore with similar structural design and behaviour): the ZIM Luanda containership.

In the Research & Development project FIBRESHIP (Fibreship, 2019), the structural health of the ZIM Luanda containership was monitored in the navigation route from Valencia, Spain to Halifax, Canada (about 3300 nm). The sensor network was composed of an environmental tracker, inclinometers, strain gauges, and an IMU, as shown in Figure 4. The data provided by the sensors were recorded and processed by a DAcS. The main objective was to constantly track the structural integrity of the ship in real time, guaranteeing efficiency and safety throughout the transatlantic voyage. It is important for any CM system to target the critical properties of the ship structure to provide meaningful results to downstream consumers, such as CBM policies or procedures. For this vessel, due to its length, the CM system targeted structural deformation (torsional and bending moments) to assess integrity and structural failures. The results were correlated to sea state and simultaneously measured weather conditions (via GPS location and hindcast) to validate the hydro-structural numerical model developed in the project. The sensors in this monitoring campaign included:

• • Environmental tracking system: This system combines data from the Airmar 220WX WeatherStation (220WX weatherstation® instrument, 2025) (wind speed, wind direction, vessel location) with wave information obtained via hindcast at specific GPS points. Note that if radar is available, this system directly receives and processes the data in real-time; otherwise, the system uses GPS to retrieve sea state information from public databases and then merges all the data together. It provided sea waves’ direction, height and periods, and wind direction and speed.

• • Biaxial inclinometers: Five sensors distributed along the ZIM Luanda ship were installed. These sensors provided real-time information on the ship’s rotational movements in the X and Y axes which were fundamental for assessing the hull’s hogging, sagging, and torsional deformations.

• • Strain gauges: Installed at the mid-section, where bending stresses were presumed to be highest, these gauges measured the local deformations, using a targeted configuration to focus on axial deformations.

• • Relative displacement sensors: Here, eddy current displacement sensors (TQ-402) (Data sheet, 2025) measured strain by correlating hull displacement with changes in the magnetic field. These sensors were tested as an alternative method of capturing local deformations, correlating them with strain gauge data, thus ensuring the reliability of the monitoring system. The sensors were mounted facing the hull on a rod, maintaining a small gap to detect any hull movement at the sensor tip through changes in the magnetic field.

• • Inertial measurement unit: With its six degrees of freedom, the IMU captured the vessel’s motions (roll, pitch, yaw, surge, sway, and heave) and essential data for correlating the external loads (environmental factors) with the ship’s structural responses.

Figure 4.

Description for sensor positioning in ZIM Luanda Containership (Fibreship, 2019; Fuentes, Reference Fuentes2019).

The monitoring results were able to provide the hogging, sagging, bending, and torsional deformations of the containership at different operational and weather conditions, showing that they were well below the safety limits recommended by classification societies for regular sailing conditions, though they also showed that in case of extremely rough navigation conditions the bending moments could reach some values above allowable limits. Outputs from a CM system like the one deployed here could also be used for assessing ship-wide and local structural health indicators and fatigue in the long term.

9.4. Lessons learned from monitoring campaigns

All three monitoring campaigns presented above offer valuable lessons learned for implementing structural CBM onboard naval vessels. The monitoring campaign from the Royal Norwegian Navy indicated the potential in using FBG for real-time structural health monitoring (Wang et al., Reference Wang, Pran, Sagvolden, Havsgård, Jensen, Johnson and Vohra2001), as FBG can be a good choice for harsh or corrosive environments. For many sensors, a FBG system can be multiplexed, reducing the overall cost per sensor. By continuously analyzing the fiber strain signals, changes in the vessel stiffness can be analyzed, indicating the possibility for damage detection. The FBG system can also indicate static hull deformations, and with enough sensors, it can detect hull vibrations and monitor noise (i.e., the vessel acoustic signature). Using a modular design for the fiber optic sensing system was also indicated to be useful because then the modules can be adapted for different monitoring applications. Global forces were determined from some of the measured strains, allowing a clear comparison between the rule loads used in the vessel design and the experienced loads, giving a clear resulting safety factor. Such information can then be further used to optimise vessel weight and strength for future vessels of the same series (Pran et al., Reference Pran, Johnson, Jensen, Hegstad, Sagvolden, Farsund, Chang, Malsawma, Wang, Claus, William and Spillman2000).

Based on the USCG monitoring campaign, a minimum system to quantify fatigue and extreme loads and support long-term structural maintenance decisions was presented in Drummen et al. (Reference Drummen, Rogers, Benhamou, Hageman and Stambaugh2019). The instrumentation consisted of three LBSGs, two strain gauges, a MRU, and a GPS. The LBSGs were used to infer cross-sectional load effects like the vertical bending moment, as they cannot be measured directly. Measured strains were combined with the first few global vibration modes to obtain section loads (Schiere et al., Reference Schiere, Bosman, Derbanne, Stambaugh and Drummen2017). A similar approach was employed in Andoniu et al. (Reference Andoniu, De Lauzon and Lamaison2019) but used selected wave load cases instead of vibration modes; improved accuracy was achieved for the bending moments using this approach. The number of LBSGs to be used depends on the accuracy that is needed at a particular location on the ship. The system as described in Drummen et al. (Reference Drummen, Rogers, Benhamou, Hageman and Stambaugh2019) was not able to provide information about the actual cracks appearing on board; due to insulation and other internal structures on board, these might be difficult to detect. Therefore, an AE monitoring system was suggested to supplement the monitoring system. Strain gauge measurements from the early lifetime of the ship can then be used to pin-point a localized area in a later stage of the ship’s lifetime where fatigue cracks are expected. These insights form the major inspiration for the baseline sensor system suggested in Section 10.

From the ZIM Luanda campaign, it was found that it is crucial to incorporate the CM system during the vessel’s construction phase, akin to essential services like electrical and ventilation systems. This early integration can prevent the challenges faced with post-construction installation, such as adapting ship structure and routing system wires through complex ship sections, which proved to be a significant difficulty for the ZIM Luanda. The monitoring campaign findings further suggested that relative displacement sensors, despite their higher cost and potential installation challenges, offer advantages such as robustness, versatility, and ease of maintenance. They have been demonstrated to be as sensitive and accurate as traditional strain gauges. However, their cost and complexity of installation could limit widespread use. Additionally, the importance of sensor location was highlighted, with the mid-section being particularly crucial for measuring traction and compression deformations. The inability to use wireless sensor systems within steel ships due to the Faraday cage effect presented a substantial challenge, necessitating the use of specially designed cables that were immune to magnetic interference.

9.5. Other monitoring paradigms

Beyond the lessons learned from the specific monitoring campaigns discussed above, we can also look to other ship structure monitoring activities for inspiration. Virtual monitoring, beyond just for estimating the environmental conditions, offers some advantages: namely that the costs and effort involved in setting up and maintaining an analysis using hindcast are significantly lower than for a physically monitored system. But the accuracy of virtual monitoring is generally lower than that of a traditional monitoring system. DGPS data were utilized with wave hindcast to test a virtual monitoring approach in Thompson (Reference Thompson2020). This work found that the virtually calculated stress spectra were in good agreement with measured stress spectra from a naval vessel monitoring campaign. It was further found in Hageman and Thompson (Reference Hageman and Thompson2022) that physical monitoring resulted in more accurate fatigue estimates than virtual monitoring via wave hindcast, especially in storm conditions. While virtual monitoring does have its benefits, physical monitoring gives more accurate and reliable information for individual vessels. The field would benefit from further targeted research determining which quantities need to be physically sensed versus which can relatively reasonably be virtually sensed, and how this affects eventual prognostic assessments, to facilitate the uptake of CBM for naval vessels.

When a fleet of vessels is concerned, an efficient combination of physical and virtual monitoring can be done, as it is possible to physically monitor a single vessel within a fleet and extrapolate insights to other virtually monitored vessels within the fleet. Such fleet monitoring has already been implemented to improve wave environment estimations, based on combining information from multiple monitored vessels in a common area (Nielsen et al., Reference Nielsen, Brodtkorb and Sørensen2019; Mounet et al., Reference Mounet, Nielsen, Brodtkorb, Tannuri and de Mello2022, Reference Mounet, Chen, Nielsen, Brodtkorb, Pillai, Ashton and Steele2023). But of course, the variance of the operational area of the fleet is extremely important when deciding if fleet monitoring is possible. Using spectral fatigue analysis to compare the accumulated fatigue damage of 10 naval vessels within a fleet, Thompson (Reference Thompson2018) found significant influence of the operational area on the resulting damage. However, when considering a mini-fleet focused on a single area, the fatigue damage across vessels varied little. The implications of partial virtual monitoring of a fleet of vessels operating in different areas and prevailing environmental conditions were further explored by Hageman et al. (Reference Hageman, Drummen, Thompson and Stambaugh2022). In general, fleet monitoring, where at least some vessels are physically monitored, can be useful for the long-term decisions needed for fleet maintenance decisions but may be less useful for onboard reactions by crew for mission-readiness.

Monitoring campaigns can also be conducted just for short periods, theoretically under harsh conditions, to assess the current fatigue life of the vessel (Johnson et al., Reference Johnson, Lynch and Collette2018). However, it is important to note that the short-term monitoring needs to begin at or near the beginning of the vessel’s lifetime, unless there is also enough information about the past operational profile to correlate operational conditions with fatigue damage. It is also possible to combine virtual and physical monitoring by using a short-term perspective with a conversion matrix method (Thompson et al., Reference Thompson, Jiang and Benhamou2021). Related to the idea of short-term monitoring is that of linking monitored and unmonitored states. In this way, monitoring of the structural health in certain sea conditions has been used to infer the structural response in “unobserved states” using the power spectral density of measured structural health monitoring data (Mondoro et al., Reference Mondoro, Soliman and Frangopol2016; Zhu and Collette, Reference Zhu and Collette2017). Though such short-term monitoring can save costs, it is envisioned that life-long monitoring is more useful for tracking the vessel structural health and making informed decisions on maintenance.

Another notable monitoring approach is the “digital healthcare engineering” paradigm for marine assets (Sindi et al., Reference Sindi, Kim, Yang, Thomas and Paik2024; Kim et al., Reference Kim, Xie and Paik2025; Kim and Paik, Reference Kim and Paik2025) based on on-site monitoring, ship-to-shore data transmission, advanced analytics on a DT, AI-driven diagnostics, and predictive health analysis for maintenance planning. Developed for aging ships and offshore structures, this approach was particularly applied to an aging containership, and such a pipeline could also be interesting for naval vessels, especially the pillars of AI-driven diagnostics and predictive health analysis for maintenance planning. Some aspects might need to be adapted to address short-term decision making required for naval vessels, as digital healthcare engineering relies on high-fidelity simulations on detailed digital twins for diagnostics and prognostics. This high-fidelity digital twin and simulation module can be useful for long-term naval prognostics, where it has already been noted that simple models cannot accurately model longer-term damage growth.