AI in Composite Overwrapped Pressure Vessels: A Review and Advanced Roadmap from Materials Design to Predictive Maintenance

Abstract

The integration of Artificial Intelligence (AI) into the design, manufacturing, and lifecycle management of Composite Overwrapped Pressure Vessels (COPVs) is transforming the pathway toward autonomous and adaptive composite systems. This paper presents a comprehensive review and roadmap for AI-enabled COPVs development, bridging materials design, process optimisation, and predictive maintenance. The study synthesises over a decade of research on data-driven composite manufacturing, combining technology intelligence, PESTEL-SWOT environmental assessment, and cross-sectoral analysis of industrial and academic advances. A unified workflow is proposed to illustrate AI integration across the COPVs lifecycle, highlighting data feedback loops for continuous optimisation through digital twins and intelligent process control. Structural Health Monitoring (SHM) plays a central role in this ecosystem by providing real-time high-fidelity data on damage evolution and environmental interactions in COPVs. Through embedded sensing technologies such as fibre optic sensors and acoustic emission systems, SHM enhances digital twin fidelity, supports AI-based anomaly detection, and strengthens model validation in safety-critical hydrogen storage applications. Critical challenges are identified, including limited hydrogen-exposure datasets, lack of real-time adaptability, explainability in safety-critical design, and sustainability of AI-intensive workflows. These challenges highlight the need for tighter SHM-AI integration to enable reliable condition assessment and prognostics under multi-physics loading conditions. Based on these findings, the paper outlines actionable research directions to enable reliable, transparent, and sustainable AI adoption in composite manufacturing under the Industry 4.0 and hydrogen-economy paradigms.

1. Introduction

Composite Overwrapped Pressure Vessels (COPVs) are critical components for high-pressure hydrogen storage and transport, ensuring containment integrity, gas impermeability, and resistance to embrittlement. They exist in several configurations (Figure 1) and must sustain complex operating conditions, including pressure fluctuations, thermal gradients, and long-term environmental exposure. The performance and reliability of these vessels depend on advanced composite materials and tightly controlled manufacturing processes to guarantee efficiency, safety, and durability.

COPVs play a central role in the hydrogen value chain for mobility, aerospace, and energy applications [1]. Despite their advantages, COPVs manufacturing remains a complex multi-stage process encompassing liner preparation, filament winding, curing, machining, and non-destructive testing (NDT). Each step is sensitive to variations in parameters such as fibre tension, winding angle, curing temperature, and resin flow behaviour, which affect the final mechanical performance [2,3,4]. Defects like fibre misalignment, voids, or porosity can lead to reduced burst pressure and increased variability in performance. These challenges emphasise the need for predictive and adaptive manufacturing strategies.

Structural Health Monitoring (SHM) has become a foundational enabler for artificial intelligence (AI) in the context of storage tanks and other critical infrastructure, because it provides the high-quality, high-frequency, and multi-modal data streams required for robust model training and inference [5]. In tank systems where degradation mechanisms such as corrosion, fatigue cracking, settlement-induced distortion, and weld defects evolve over long time horizons, sensor networks (e.g., acoustic emission, guided waves, strain gauges, fibre optic sensors) transform latent physical processes into measurable features suitable for data-driven analytics. As demonstrated in the broader SHM literature [6,7], the transition from traditional threshold-based diagnostics to pattern recognition frameworks has been pivotal in operationalising AI for damage detection, localisation, and prognosis. More recent work integrating deep learning and feature learning methods into SHM pipelines [8,9] shows that AI performance is strongly contingent on the quality of sensing architectures, data labelling strategies, and physics-informed feature engineering elements that are central in tank monitoring studies. In this sense, SHM is not merely a data acquisition layer for AI; it defines the observability, interpretability, and ultimately the reliability of AI-driven decision support in safety-critical tank infrastructure.

The Industry 4.0 paradigm offers opportunities for intelligent manufacturing through the integration of the Internet of Things (IoT), Artificial Intelligence (AI), and Digital Twin technologies [10,11]. However, the COPVs sector still faces major barriers to digital transformation: fragmented data ecosystems, reliance on expert heuristics, and the absence of unified frameworks for AI integration. Despite the abundance of process and inspection data from filament tension records and curing profiles to acoustic emission logs most information remains siloed and underexploited [12], limiting traceability, cross-process optimisation, and data-driven design-to-production pipelines.

Recent advances in sensing technologies and Structural Health Monitoring (SHM) methodologies have significantly strengthened the integration of artificial intelligence (AI) in the assessment of storage tanks and related infrastructure. Contemporary studies emphasise the synergistic coupling of high-resolution sensor networks, physics-based modelling, and data-driven algorithms to enhance damage detectability, diagnostic reliability, and prognostic capability [13,14,15,16,17,18,19,20]. These works collectively demonstrate that AI performance in SHM applications is fundamentally conditioned by sensing architecture design, signal quality, feature extraction strategies, and the incorporation of mechanical and material domain knowledge. In tank systems where degradation processes such as corrosion, cracking, settlement, and weld defects evolve gradually and interact non-linearly, the fusion of numerical modelling, experimental sensing, and machine learning enables improved anomaly detection, damage localisation, and remaining useful life estimation. Consequently, SHM provides not only the empirical data backbone for AI deployment but also the physical interpretability and validation framework required for trustworthy decision support in safety-critical storage infrastructure.

This paper presents a comprehensive review and roadmap for AI-enabled COPVs development, encompassing materials design, structural optimisation, manufacturing, and predictive maintenance. The study examines technological, methodological, and organisational enablers of AI adoption and identifies key barriers. The framework aligns data-driven decision-making with domain knowledge and physical constraints to support reliable, efficient, and sustainable COPVs manufacturing under the Industry 4.0 paradigm [21,22].

The literature indicates that AI is increasingly used across four interconnected stages of the COPVs lifecycle: (i) material discovery and optimisation, leveraging machine learning and generative algorithms; (ii) design and simulation optimisation, integrating AI with finite element modelling, generative design, and uncertainty quantification; (iii) manufacturing and process control, employing AI for defect detection, winding path optimisation, curing cycle prediction, and process parameter tuning; and (iv) predictive maintenance and structural health monitoring (SHM), enabling early damage detection and life prediction through data-driven sensor fusion and digital twins. Figure 2 illustrates the interconnected concepts and research focus areas commonly discussed in the literature on hydrogen storage tanks.

This review synthesises recent scientific and industrial progress in these four domains, identifies technical and data limitations, and constructs a forward-looking AI roadmap for COPVs lifecycle management. By bridging materials science, data analytics, and process intelligence, this review provides a conceptual framework for explainable and sustainable AI integration in hydrogen storage.

2. Review Methodology and Data Sources

This review adopts a multidisciplinary approach integrating academic research, industrial data, and strategic market intelligence to evaluate AI applications in COPVs manufacturing and design. Peer-reviewed literature, technical reports, patents, and market analyses were systematically examined to provide both qualitative and quantitative insights into trends, adoption, and technological maturity.

To complement academic sources, real-time technology and market monitoring were conducted using the Talkwalker platform [23] and a tailored AI thesaurus of domain-specific terms. A database of 375 companies active in hydrogen tanks and materials enabled tracking of emerging players and trends. Insights were consolidated through a collaborative dashboard integrating AI-based analytics for SWOT (Strengths, Weaknesses, Opportunities, and Threats) profiling, market sizing, and technology readiness assessment.

The market of hydrogen storage tanks and transportation is divided into several segments based on application, including Vehicles, Hydrogen Transportation Trailers, Stationary Storage, Railways and Marine. Vehicles represent the largest share of the hydrogen storage tank and transportation market in 2025, accounting for $94.2\%$. Type-IV tanks dominate the hydrogen tank material market (41.7% in 2024, projected 50% by 2030) due to their lightweight, high-strength composite construction and remain the fastest-growing segment driven by automotive adoption [24]. Aligned with the market needs for 2025–2026, our R&D development efforts are currently focused on Level-IV tanks dedicated to the automotive sector.

AI is rapidly transforming the hydrogen tanks industry by optimising various stages of the product lifecycle, from design and material selection to manufacturing and operational maintenance. AI algorithms can analyse vast datasets to simulate material behaviour under extreme conditions, enabling engineers to design lighter, stronger, and more cost-effective tanks. This includes predictive modelling for composite winding patterns, stress distribution, and thermal management, significantly accelerating research and development cycles and enhancing tank performance and safety [25,26]. Among others, Market research publishers such as Consegic Business Intelligence, The Business Research Company, and MarketsandMarkets consider that the hydrogen tank market is entering a phase of significant transformation, driven in particular by the increasing integration of AI throughout the entire process chain.

2.1. Overview of Methodology

A multi-stage methodology was implemented, as illustrated in Figure 3, including identification of technology domains and industrial stakeholders, literature and market data collection, quantitative and qualitative analysis, and synthesis of trends and gaps.

Note that the literature search and selection were conducted using a total of 453 records identified from Scopus, Questel-Orbit patents, and Talkwalker. After screening and eligibility assessment, 77 documents comprising 66 scientific articles, 4 patents, and 7 Talkwalker sources were retained for qualitative synthesis.

2.2. PESTEL Analysis of COPVs

To assess external factors shaping COPVs development, a PESTEL (Political, Economic, Social, Technological, Environmental, and Legal) analysis was conducted.

Political: EU regulations under the European Green Deal target ${\mathrm{CO}}_2$ emission reductions of 45% by 2030 and 90% by 2040. Hydrogen Europe projects at least 50,000 hydrogen trucks in operation in the EU-27 by 2030 [27]. Government incentives and funding for green hydrogen infrastructure remain key drivers for fuel cell vehicle adoption [28,29,30,31].

Economic: COPVs rely on high-performance materials such as carbon fibre composites, which represent up to 60% of tank costs. Manufacturing complexity and limited supplier bases increase vulnerability to price fluctuations, while competition from alternative storage technologies affects market adoption.

Social: Safety concerns related to high-pressure hydrogen storage, public trust, skilled workforce requirements, and infrastructure investment influence COPVs acceptance and deployment.

Technological: Advances focus on improving strength-to-weight ratios, fatigue life, hydrogen impermeability, and real-time monitoring via embedded sensors. Challenges include scalable manufacturing, high-pressure durability, self-healing materials, and integration with EVs, fuel cells, and renewable energy systems.

Environmental: COPVs production consumes high energy, involves non-recyclable materials, and can generate long-term waste. Managing material degradation from hydrogen, UV (ultraviolet), moisture, and chemicals is crucial for sustainability.

Regulatory: The upcoming ISO/AWI 11119-2 [32,33] revises safety, material categories, manufacturing methods, and design factors. The standard emphasises data-driven design, digital monitoring, AI/FEA/CFD (for artificial intelligence, finite element analysis and computational fluid dynamics respectively) integration, and full traceability, reinforcing compliance and smart manufacturing practices.

The PESTEL analysis shows that external pressures (tightening regulations, growing market demand for high-performance hydrogen systems, rising safety expectations, and rapid technological advances) make AI not only beneficial but strategically essential for COPV manufacturing.

2.3. Sectorial SWOT Analysis

Strengths, weaknesses, opportunities, and threats related to AI adoption in COPVs manufacturing were systematically evaluated. Key points are summarised in the main text, while full quantitative and visual details, including company-level benchmarking, are provided in Appendix A.

2.4. Summary of the Approach

This methodology integrates multiple perspectives: technology readiness, market adoption, regulatory landscape, and industrial capabilities. Combining qualitative insights with quantitative data provides a structured and replicable framework for AI applications in COPVs. Detailed data supporting the analyses are provided in Appendix A.

3. Analytical Framework for AI Deployment in COPVs

This section synthesises insights from literature, market intelligence, and PESTEL/SWOT analyses to provide a detailed analytical framework for AI applications in COPVs, focusing on materials discovery, design optimisation, intelligent manufacturing, and predictive maintenance.

3.1. AI-Enhanced Material Discovery and Optimisation

Recent studies demonstrate the growing role of AI in accelerating material discovery and optimisation for COPV applications. Saharudin et al. [34] applied large language models to assess mechanical and environmental trade-off in composite materials for Type-V pressure vessels. Their analysis shows that while Carbon T-700/Epoxy provides superior mechanical strength, Basalt/Epoxy offers a more sustainable alternative with reduced environmental impact. In parallel, Osman et al. [35] used machine learning techniques to screen porous carbon-based adsorbents, identifying high-performance candidates that improve hydrogen storage safety and efficiency. Qiao et al. [36] integrated AI methods with Metal–Organic Frameworks (MOFs) to optimise hydrogen storage under ambient conditions, contributing to emissions reduction and carbon neutrality targets. At Khalifa University, Lemaoui et al. [37] combined molecular modelling with neural networks to predict thermal conductivity in green solvents, supporting the design of advanced ionic and eutectic solvents for COPVs-related systems. Similarly, Madirisha et al. [38] employed interpretable machine learning models (CatBoost and XGBoost) to predict storage performance in Al-PILCs, identifying key drivers such as pressure and temperature and thereby reducing the experimental workload and development time. When integrated with Density Functional Theory (DFT), such AI-assisted property-prediction tools further accelerate high-throughput virtual screening. Collectively, these AI-driven methodologies enable rapid evaluation of novel fibre–resin composites and liner materials, optimising critical parameters such as the strength-to-weight ratio, hydrogen permeability, and long-term durability [39,40,41,42].

Despite significant advances in AI-assisted material screening and the proliferation of predictive models for hydrogen-storage performance, current approaches still face important limitations in capturing the complex multi-scale interactions that govern COPV materials. Many existing tools remain constrained by narrow training datasets, limited domain transferability, and insufficient modelling of emerging fibre–matrix systems or innovative liner architectures.

These observations highlight the need for a more integrated and physically informed approach to AI-driven material development for COPVs. Building on the limitations identified in current methods, the present study proposes a research roadmap for future directions, outlining how high-fidelity simulations (such as multi-physics FEA, molecular dynamics, or DFT) could be coupled with data-driven prediction models to improve robustness, enhance generalisation, and accelerate the discovery of next-generation COPV materials.

3.2. AI-Driven Design and Structural Optimisation

In industry, companies such as OPmobility [43], in collaboration with Neural Concept [44], have deployed 3D generative AI platforms to simulate multi-physics behaviours in hydrogen tanks. This integration enables rapid design iterations, enhances robustness, and has shorter development cycles, positioning AI-augmented design as a key differentiator for Original Equipment Manufacturers (OEMs) and Tier 1 suppliers. Recent patent filings further demonstrate the accelerating adoption of AI in COPVs design and simulation. Fan et al. [45] patented an ML-based framework for optimising composite layup angles to meet target strength. The patent discloses a design method for the filament winding process of a Type IV hydrogen storage container that prioritises lightweight construction. It provides a structured process for planning and optimising the winding sequence, angles, and layer distribution of composite reinforcement (typically carbon fibre) around a polymer liner so that the resulting pressure vessel meets strength and performance targets with minimised weight. The method integrates consideration of mechanical requirements and manufacturing parameters to guide process design and improve the overall efficiency and structural performance of the hydrogen storage container. The Hefei General Machinery Research Institute patented an AI-assisted system to refine winding sequences and layer counts for lightweight tanks [46]. The patent discloses a design methodology for the composite layering of Type IV high-pressure hydrogen storage cylinders that jointly evaluates both the structural strength and fatigue life. The method uses a finite element model to assess the strength and fatigue performance of an initial ply plan, identifies fatigue-sensitive regions, and then adjusts local winding angles and layer distribution without changing the total layer count to improve the fatigue resistance while maintaining the burst strength. Iterative verification and trial testing ensure the optimised layering meets the strength and fatigue requirements for 70 MPa service conditions. Fang et al. [47] combined simulation and ML to assess liner stress amplification due to dense defects, improving structural reliability.

AI has also been widely applied to optimise COPVs’ geometry and structural layout. Van Bavel et al. [48] achieved a 27.3% reduction in layup thickness via reliability-based optimisation. Coskun et al. and Li et al. [49,50] used meta-heuristics and surrogate models to minimise interlaminar stresses and manage hybrid uncertainties. Neural networks coupled with genetic algorithms further reduced the vessel mass without compromising the strength [51].

For performance prediction, Ma et al. [52] applied Artificial Neural Networks (ANNs) to forecast hydrogen storage capacity and thermal behaviour, achieving close agreement with the experimental data. Winding pattern optimisation has benefited from genetic algorithms [53], Machine Learning with Feature Engineering (ML-FE) frameworks [54], and ANN-GA hybrids [55], improving stacking sequences and dome geometries. Evolutionary algorithms also enhanced material efficiency and asymmetric winding strategies [56,57]. AI has significantly improved thermal analysis as well. Bhattacharjee et al. [58] achieved an R² of 0.9975 for temperature rise prediction during fast filling. The study investigates how varying inlet diameters of a hydrogen gas storage tank affect internal temperature rise during fast filling using computational fluid dynamics (CFD) simulations. Different inlet sizes (e.g., 20 mm and 25 mm) were modelled to analyse spatial temperature distributions, and the results show that larger inlet diameters tend to produce higher temperature increases during rapid fill. An artificial neural network trained with the neurofit technique was then used to predict the temperature rise with high accuracy, demonstrating that the ANN can reliably capture the relationship between inlet diameter and thermal response during fast filling. Patel et al. [59] integrated CFD and ML to identify thermally hazardous scenarios and enhance operational safety. A very recent review paper highlighted the integration of AI with traditional modelling approaches for the design, optimisation, and performance prediction of COPVs [13], emphasising AI–mechanics hybrids for improving failure prediction, burst pressure estimation, and uncertainty-aware material optimisation.

Combining the Finite Element Method (FEM) and AI provides rapid, accurate predictions that bridge simulation and experiment. Ding et al. [60] used surrogate models for mass reduction, where they optimised the helical winding angle and bandwidth of high-pressure hydrogen storage vessels using a surrogate-based optimisation framework, enabling the efficient identification of winding parameters that improve structural performance while reducing the computational cost. Wang et al. [61] applied hybrid AI-FEM schemes for failure prediction, with similar results reported in [62]. Surrogate models (ANNs, Support Vector Regression (SVR), Gaussian Process Regression (GPR)) support fast exploration of design spaces and are often paired with evolutionary algorithms for multi-objective optimisation. AI also supports uncertainty quantification and adaptive learning [63]; the authors conducted uncertainty quantification and global sensitivity analysis for burst pressure in Type-IV hydrogen composite pressure vessels, using Latin Hypercube Sampling and sparse Polynomial Chaos Expansion to model variability from material and geometric uncertainties, showing that fibre tensile strength and ply thickness dominate burst pressure variability, thus guiding more reliable vessel design. For failure analysis, Shang et al. [64] developed a predictive framework combining regression models and virtual sample generation via the Markov Chain Monte Carlo (MCMC) method. Generative AI models have further enhanced the simulation efficiency. Zhuang et al. [65] proposed a generative network-based method to rapidly compute the stress distribution in the composite winding layers of high-pressure hydrogen storage tanks, replacing or augmenting traditional finite-element analysis with a trained deep learning model to accelerate stress predictions while maintaining acceptable accuracy compared to conventional simulation-based approaches. Hong [66] applied transfer learning to reduce reliance on full-scale simulations. Klepp et al. [67] accelerated analysis of hydrogen adsorption using AI-enhanced models. They developed a hybrid modelling framework that combines physics-based CFD of hydrogen flow and adsorption in tanks filled with activated carbon with machine learning-derived source terms, enabling more computationally efficient prediction of temperature and concentration distributions during charging/discharging than traditional simulations. At the system level, hybrid AI and molecular dynamics approaches, such as Perturbed Chain-Statistical Associating Fluid Theory PC-SAFT-based coupling [68], extend AI’s role to underground storage and complex hydrogen environments demonstrating a broad impact across the hydrogen value chain.

Research on AI-driven COPV design and structural optimisation has produced a wide spectrum of surrogate-, metaheuristic-, and hybrid FEM-AI approaches; however, their limited ability to fully capture multi-objective trade-offs, uncertainty propagation, and multi-scale failure mechanisms reveals the need for a broader and more integrative perspective. Building on these insights, this study proposes a research roadmap that outlines how physics-informed simulation and AI-driven optimisation can be unified into a coherent modelling strategy. Such an approach aims to more reliably address the inherently multi-objective and uncertainty-driven nature of COPV design while overcoming the limitations of existing, narrowly focused tools.

3.3. Intelligent Manufacturing and Process Control

AI technologies increasingly support process modelling, quality prediction, curing-cycle monitoring, tension control, and early defect detection in COPV manufacturing. Hybrid approaches that combine AI with physics-based simulations, along with the integration of sensor data accelerate feedback loops and reduce dependence on destructive testing. A notable example is the smart hydrogen tank developed by Arkema Piezotech (France), designed to enable real-time structural health monitoring and impact detection. This system incorporates a lightweight network of printed piezoelectric sensors directly onto the COPVs surface, capable of capturing acoustic emissions—often referred to as “screams”—generated by stress events such as fibre rupture or matrix cracking. Leveraging printed electronics, the sensor network minimises wiring complexity while ensuring continuous reliable monitoring during operation and transport. This innovation directly contributes to overcoming key safety challenges associated with high-pressure hydrogen storage. In manufacturing, Ma et al. and Pardhi et al. [69,70] demonstrated how AI can enhance lifecycle optimisation and overall hydrogen vehicle system performance. Recent studies have also highlighted AI’s effectiveness in predicting critical manufacturing outcomes. Santos et al. [71] employed Gaussian Process Regression (GPR) and Kernel Principal Component Analysis combined with Lasso regression (KPCA-Lasso) to estimate COPVs’ burst pressure using manufacturing parameters. This predictive approach enhances early-stage quality control and may reduce overly conservative safety margins without compromising the structural integrity, thereby improving both the design efficiency and manufacturing reliability.

Although numerous studies have examined AI-enabled monitoring and developed models grounded in data-driven prediction of manufacturing quality and structural integrity, these contributions still fall short in addressing the full multi-stage complexity of COPV behaviour (from material variability through processing, assembly, and in-service degradation). Current approaches remain largely compartmentalised, focusing on isolated steps rather than providing an integrated end-to-end representation of the tank’s lifecycle. This limitation motivates the need for a unified modelling perspective that couples process data, physics-based simulation, and predictive analytics. Such an approach would support more reliable, efficient, and digitally driven COPV design and manufacturing by bridging the gaps between material behaviour, process parameters, and long-term structural performance.

3.4. Predictive Maintenance

AI-enabled structural health monitoring (SHM) leverages sensor data to detect early-stage damage, predict fatigue life, and optimise inspection schedules [5]. Digital twin-driven predictive models further enable performance-based maintenance and Remaining Useful Life (RUL) estimation. By analysing complex operational data, AI enables real-time monitoring, fault detection, and lifespan prediction, improving safety, reducing costs, and minimising downtime. Industrial developments illustrate this trend. Airbus patented a machine learning-based method to monitor COPVs’ filling levels through resonance-frequency analysis [45]. Rolls-Royce developed an AI-driven diagnostic system that interprets sensor data to assess aircraft fuel tanks and support condition-based replacement. Henan Nayu New Material deployed an AI-based leak detection framework for solid-state hydrogen tanks, using deep acoustic analysis. AI also enhances autonomous State Grid Intelligent Technology, and Hui Z. [72] demonstrated cloud-based algorithms capable of detecting defects and optimizing inspection routes for hydrogen-powered UAVs.

Collectively, these efforts reflect a growing movement toward embedding AI within COPV-integrated platforms for real-time diagnostics, decision support, and system control. AI is reshaping non-destructive testing (NDT) and SHM. Figure 4 presents the annual publication trends for (i) AI-enabled SHM in COPVs/${\mathrm{H}}_2$ tanks and (ii) AI applications in Type IV/V COPVs derived from the bibliographic queries defined in Section 2. The figure compares annual publication counts (2017–2026) for two queries: AI-SHM in COPVs/${\mathrm{H}}_2$ tanks and AI in Type IV/V COPVs. Both topics grow slowly up to 2020 and then increase sharply from 2021 onward, with the steepest rise in 2024–2025 (peaking at 126 for AI-SHM and 189 for Type IV/V AI in 2025).

As an example of online condition monitoring for pressure vessels, Gong et al. [73] propose a quasi-distributed FBG sensing network, whose placement is guided by stress-concentration analysis, combined with a vibration-correction algorithm; their approach enables reliable leak detection during vacuum extraction with approximate localisation and an initial assessment of leak severity, illustrating how instrumentation design and signal pre-processing form critical upstream building blocks for AI-enabled diagnostics and predictive maintenance. Aramburu et al. [74] systematically reviews non-destructive testing (NDT) methods applied to polymer composite pressure vessels, evaluating techniques such as ultrasonic testing, radiography, thermography, acoustic emission, and visual inspection for detecting defects (voids, delamination, improper impregnation) and monitoring structural health; it highlights the importance of advanced sensors, data analysis, and AI-driven tools to improve the reliability, predictive maintenance, and safety in composite vessel manufacture and service. Charmi et al. [75] used guided-wave data and deep learning to isolate damage features under fatigue loading. Jiang et al. [76] achieved up to 98% accuracy in burst pressure prediction using acoustic emission data and TabNet models. Advanced sensing and modelling approaches further enrich predictive maintenance capabilities. Fibre-optic neural sensor networks [77], convolutional neural networks (CNNs) [78], and AI-driven digital twins [79] are increasingly used for real-time monitoring. Karapanagiotis et al. [80,81] integrated distributed fibre-optic sensing with machine learning to enable continuous tracking of damage evolution. Similarly, Souza et al. [77] and Maurin et al. [82] demonstrated how embedded optical fibres combined with AI support early defect detection and precise strain monitoring. Quarssis et al. [83] employed XGBoost for real-time mechanical performance prediction. Hao et al. [84] applied ML to improve the long-term performance prediction of fibre-reinforced plastics, while Kadri et al. [85] used Bidirectional Long Short-Term Memory (Bi-LSTM) networks to model COPV degradation. Hu et al. [86] examined failure risks under extreme conditions using AI-enhanced evaluations.

Detection of microcracks and hydrogen-induced damage has been advanced using deep learning and clustering methods. Hua et al. and Ren et al. [87,88] used acoustic emission signals transformed into Mel spectrograms and deep learning classifiers to detect and classify damage in filament-wound CFRP composite structures, demonstrating that spectrogram-based deep models can effectively identify damage types and improve structural health monitoring accuracy compared to conventional signal analysis. Qiu et al. [89] demonstrated early hydrogen defect detection using CNNs trained on emission data. El Moutaouakil et al. [90] classified ultrasonic guided-wave responses under variable pressures. Yang et al. [91] used Gaussian Mixture Models with AE-rate theory to characterise damage evolution, while Patil et al. [92] explored IoT-AI integration to enhance hydrogen storage safety. Chen et al. [93] demonstrated that AI-enabled predictive maintenance is becoming a strategic enabler of national competitiveness within hydrogen infrastructure.

Taken together, these developments mark a paradigm shift: AI is no longer limited to data analysis, it is enabling self-aware adaptive COPVs systems. The convergence of smart sensing, machine learning, and cloud-based intelligence is laying the foundation for intelligent hydrogen storage and a safer more efficient energy ecosystem.

Despite substantial progress in AI-enabled SHM and data-driven predictive models, existing approaches do not yet unify multi-sensor data, physics-based behaviour, and lifecycle uncertainty into a cohesive deployable predictive-maintenance framework for COPVs. Current methods remain highly specialised (targeting burst-pressure prediction, defect detection, or fatigue assessment independently) without providing an integrated understanding of in-service degradation under diverse operational conditions. To address this gap, we propose a comprehensive modelling strategy that couples heterogeneous sensing data, machine-learning inference, and physics-informed digital twins. Such a unified approach would enable reliable real-time decision-making for next-generation hydrogen COPV systems, supporting safer operation, improved asset management, and lifecycle-aware maintenance strategies.

Figure 5 illustrates the integrated development framework adopted for filament-wound hydrogen storage tanks. The workflow encompasses (a) the modelling and simulation of filament winding stacking sequences, (b) finite element analyses of the stacked composite structure, (c) the real filament winding manufacturing process, and (d) the resulting hydrogen tanks produced via the filament winding route. Artificial intelligence is embedded across these stages as a cross-disciplinary enabler, supporting design optimisation, performance prediction, process control, and data-driven feedback between modelling and manufacturing.

4. Roadmap for AI Deployment in COPVs Production

A phased, lifecycle-oriented roadmap for AI-driven COPVs production is proposed (

Table 1. AI Deployment Roadmap for COPVs: Four-Stage Overview.

Stage	Current AI Applications	Proposed Roadmap/Key Actions
AI-enhanced Material Discovery and Optimisation	ML-DL for property prediction, virtual screening, high-throughput materials design	Standardised, multi-scale material databases; AI + physics coupling for fibre-matrix and liner interactions; multi-objective optimisation integrating strength, permeability, sustainability; closed-loop virtual-experimental workflows; digital material prototyping; integration of DFT/MD/FEA with data-driven models for improved generalisation; focus on Type IV/V tanks, lightweight architectures, and hydrogen compatibility and high-strength composite
AI-driven Design and Structural Optimisation	Surrogate modelling; evolutionary algorithms; uncertainty quantification	Physics-informed and hybrid AI–mechanics modelling; Adaptive surrogate models for multi-objective design (mass, cost, safety); integrated design–manufacturing–SHM platforms enabling lifecycle-aware optimisation; improved uncertainty propagation and design reliability
Intelligent Manufacturing and Process Control	Process modelling; quality prediction; defect detection; limited digital twin use	Unified, multi-source datasets (material, process, sensor data); hybrid AI–physics process models for curing, winding, and consolidation; closed-loop real-time control systems (vision, tension, temperature); predictive process management to reduce waste and improve reproducibility
Predictive Maintenance	Sensor-based SHM; early damage detection; fatigue prediction	Smart multiplexed, interoperable sensor networks e.g., quasi-distributed FBG monitoring for online leakage diagnosis with stress-concentration-guided sensor placement and vibration correction; multimodal AI integrating acoustic, vibration, strain, thermal, and pressure data; physics-informed digital twins for anomaly detection and remaining useful life (RUL) forecasting; life-cycle uncertainty modelling for in-service degradation

), integrating materials, design, manufacturing, and predictive maintenance strategies. Scenario planning accounts for uncertainties in policy, market adoption, and material availability.

Data Challenges and Standardisation Needs

The scarcity and heterogeneity of datasets, particularly hydrogen-exposed composite materials, constrain AI model robustness. Fragmented proprietary and laboratory data prevent benchmarking and reproducibility. Establishing shared standardised data repositories under international coordination is critical for model validation and transfer learning.

5. Discussion and Future Perspectives

AI integration in COPVs manufacturing improves efficiency, safety, and sustainability. Operational advantages include process optimisation, quality assurance, accelerated R&D, and data-driven decision-making. Infrastructure upgrades, workforce upskilling, and strategic collaborations are required for effective deployment. These improvements are expected to reduce development cycles, minimise defect rates, and support regulatory compliance, making AI a key enabler for the industrial-scale adoption of hydrogen storage systems.

5.1. Barriers

Integrating AI throughout the hydrogen-tank value chain faces several significant barriers. First, the lack of high-quality and standardised datasets, especially for failure modes and long-term degradation, limits model accuracy. Sensor data such as acoustic emissions or optical-fibre signals often contain substantial noise, making defect detection and lifetime prediction difficult. Many AI models still struggle to generalise to new materials, geometries, and operating conditions, reducing their reliability in real industrial environments. Workflows remain fragmented, with design, manufacturing, simulation, and SHM systems operating in isolation rather than through unified data pipelines. Real-time integration of sensors, edge computing, and manufacturing equipment is still challenging due to heterogeneous hardware and communication protocols. Deep learning models also suffer from low interpretability, raising concerns for certification, safety, and regulatory approval. In manufacturing, the limited digitalisation of filament winding and curing processes slows down AI deployment. At the system level, cybersecurity and data-sharing constraints hinder the development of connected digital twins. Finally, the absence of standardised protocols for AI validation and SHM certification remains a major obstacle to large-scale industrial adoption. Addressing these barriers requires coordinated efforts in data curation, robust sensor calibration, model interpretability studies, and harmonised workflow integration, which together can improve trustworthiness and scalability of AI in hydrogen tank applications.

5.2. Regulatory and Standardisation Challenges

Despite the availability of multiple international standards, significant standardisation challenges remain for deploying AI across the hydrogen-tank value chain. Existing AI frameworks such as ISO/IEC 22989 [94] (AI concepts), ISO/IEC 23894 [95] (AI risk), ISO/TR 24028 [96] (AI trustworthiness), and ISO/IEC 38507 [97] (AI governance) provide high-level guidance but lack sector-specific procedures for certifying AI models in safety-critical COPVs applications. Hydrogen and pressure-vessel standards including ISO 11119 [32] (composite cylinders), ISO 15869 [98] (onboard hydrogen), ISO 19881 [99] (pressure relief), ISO 16111 [100] (hydrogen containers), and ISO 21009 [101] (cryogenic vessels) do not yet incorporate requirements for AI-enabled monitoring, digital twins, or predictive maintenance. As a result, fragmented data formats, heterogeneous sensing practices, and the absence of standardised AI validation pipelines complicate certification, slow industrial adoption, and create uncertainty for Original Equipment Manufacturers (OEMs) and regulators. Bridging this gap will require developing AI-tailored standardisation guidelines, linking digital twin outputs with certification protocols, and establishing reproducible validation procedures that demonstrate safety and reliability for end users.

5.3. Opportunities and Strategic Recommendations

AI-enabled hydrogen technologies present significant opportunities for all stakeholders across the hydrogen value chain to position themselves as leaders in materials, manufacturing, and SHM innovation. Organisations can accelerate progress by developing open high-quality datasets and reference benchmarks on composite behaviour, defect signatures, and lifecycle degradation, enabling robust and transferable model training. Implementing hybrid AI-FEM toolchains and digital-twin platforms will support faster design cycles and promote deeper collaboration among OEMs, suppliers, research institutions, and technology developers. Partnerships with sensor manufacturers and industrial production lines can create real-time data-rich test beds for validating AI-driven process monitoring and early defect detection. Stakeholders can also drive pre-standardisation activities by linking existing hydrogen vessel regulations (ISO 11119 [32], ISO 15869 [98]) with emerging AI governance frameworks (ISO/IEC 23894 [95], ISO/TR 24028 [96]), helping to shape future certification pathways. Strategic investment in explainable AI, trustworthy analytics, and secure digital infrastructures will strengthen user confidence and support broader adoption. Finally, by fostering multidisciplinary expertise and providing shared pilot facilities or demonstration environments, organisations can reduce implementation barriers, de-risk innovation, and contribute to safer, lighter, and increasingly intelligent hydrogen storage solutions. These opportunities collectively highlight AI as a central pillar that connects materials selection, structural design, process monitoring, and lifecycle management, ensuring rational and evidence-based advancement of hydrogen tank technologies.

5.4. Future Research Directions

Future research should expand beyond classical Type-IV tank modelling toward hybrid physics–AI methodologies that couple high-fidelity FEM simulations with machine-learning predictors for burst pressure, damage progression, and remaining-useful-life estimation. A critical enabler of this vision is the creation of curated high-quality datasets and benchmarking protocols, addressing a major industry gap and enabling reproducible comparable AI models for defect detection and lifecycle prediction. By embedding explainable AI and uncertainty quantification into the workflow, we can help bridge the divide between rapid AI innovation and the certification requirements that govern safety-critical composite pressure vessels. Finally, engaging in pre-standardisation activities and demonstrating smart self-aware tank concepts will establish the team as a leading reference for the next generation of AI-driven hydrogen safety technologies. Moreover, AI integration should be systematically linked across design, materials optimisation, simulation, SHM, and predictive maintenance, providing evidence-based justification for each proposed research direction. This approach will ensure that future developments are not only innovative but also grounded in rational reproducible methodologies that can accelerate adoption in industrial hydrogen storage applications.

6. Conclusions

This review shows that Artificial Intelligence is rapidly reshaping the entire lifecycle of Composite Overwrapped Pressure Vessels (COPVs), providing new capabilities for material discovery, structural design, process control, and in-service monitoring. AI-based material informatics accelerates the identification of optimal composite architectures, while hybrid physics–AI models enhance reliability in design, improve failure predictions, and enable multi-objective optimisation for lightweight and high-performance hydrogen storage systems. In manufacturing, digital twins, machine vision, and sensor fusion improve defect detection, winding accuracy, and cure management, supporting more consistent and certifiable production. During operation, AI-enabled structural health monitoring (combining acoustic emissions, fibre-optic sensing, and prognostics) supports earlier detection of degradation and transition toward predictive maintenance. However, the integration of AI into safety-critical COPVs applications remains constrained by fragmented datasets, limited model interpretability, cybersecurity risks, and regulatory frameworks that do not yet address digital twins or data-driven verification. Current standards offer partial guidance but require significant evolution to accommodate trustworthy AI in hydrogen storage systems.

These gaps create opportunities for coordinated action by Research and Technology Organisations, industry partners, and standardisation bodies. Priorities include building shared high-quality datasets, developing validated physics-informed AI models, establishing sensor-integrated test beds, and advancing pre-standardisation efforts for AI-based inspection and SHM. Strengthened knowledge-management practices and interoperable ontologies will further support reproducibility and industrial deployment. Overall, AI emerges not as an auxiliary tool but as a foundational technology that enables safer, lighter, more efficient, and more intelligent COPVs. Its responsible and standardised integration will be essential for scaling hydrogen technologies across mobility, aerospace, and energy infrastructures.

Beyond incremental improvements, AI provides a unifying digital backbone across the entire engineering chain of hydrogen tanks. In the design phase, generative and surrogate-based optimisation frameworks enable rapid exploration of stacking sequences, winding patterns, and geometrical configurations under multi-physics constraints, significantly reducing development cycles. At the materials level, data-driven approaches facilitate the tailoring of resin systems, nanomodified matrices, and fibre architectures to meet coupled requirements of strength, permeability resistance, and durability under high-pressure hydrogen environments. Within simulation workflows, AI-enhanced reduced-order models and physics-informed neural networks enable faster yet reliable prediction of nonlinear damage evolution, burst pressure, and fatigue life, thereby complementing conventional finite element analyses. When integrated into digital twin architectures, these models create a continuous feedback loop between virtual prototypes and operational tanks. In service, advanced SHM systems leveraging machine learning for pattern recognition, anomaly detection, and remaining useful life estimation transform maintenance strategies from scheduled inspection to fully predictive and condition-based paradigms. Collectively, these advances position AI as a catalyst for accelerating certification pathways, increasing confidence in lightweight designs and supporting the safe large-scale deployment of hydrogen storage technologies in mobility, aerospace, and stationary energy systems.