Artificial intelligence and digital technologies in railway systems: a systematic literature review using the TCCM framework

Artificial intelligence (AI) and digital technologies have strengthened the transformation of railways by enhancing efficiency, safety, sustainability and operational resilience of the systems. It enables predictive maintenance and condition-based maintenance (Vatakov, Pencheva, & Dimitrova, 2022; Rohovyi, Samsonkin, Yurchenko, & Bureika, 2025). Railways play a significant role in a country's economic development, be it passenger mobility or freight logistics. There remains a continuous pressure on operators to enhance operational performance, optimisation of asset utilisation, reducing lifecycle costs and enhance service reliability (Efimova, Baboshin, Igolnikov, & Dmitrieva, 2020). Thus, leading to the integration of intelligent technologies such as AI, machine learning (ML), deep learning (DL), Internet of Things (IoT), digital twins (DT) and cyber-physical systems (CPS) in the railway systems, serves as enablers of digital transformation (Vatakov et al., 2022; Rodríguez-Hernández, Crespo-Márquez, Sánchez-Herguedas, & González-Prida, 2025; Kour et al., 2022).

The relevance of AI in literature on digital transformation in railway systems is increasing, which remains scattered and fragmented. The studies on the themes around infrastructure monitoring, freight logistics, signalling systems and railway operations have developed independently.

The Theory–Context–Characteristics–Methodology (TCCM) framework offers a structured and systematic approach for synthesising a body of literature in any domain by interrogating the theoretical underpinnings, contextual settings, characteristic dimensions and methodological approaches adopted in the reviewed studies (Paul, Parthasarathy, & Gupta, 2017). TCCM framework has been applied in this study to the reviewed literature, expanding the horizon of understanding beyond thematic clustering.

To map the literature on AI and digital transformation in Railway infrastructure, 33 articles have been identified and included from the Scopus, Web of Science and Google Scholar databases, covering the period 1999–2026. The primary objectives of the study are: (1) mapping the theoretical foundations underlying AI applications in railway systems; (2) identifying the contextual settings and geographies in which these applications have been studied; (3) characterising the technological dimensions, application domains and publication trends and (4) analysing methodological approaches, their rigour and associated limitations. The review culminates in an integrated TCCM synthesis, identification of research gaps and a forward-looking research agenda.

The following are the objectives of the study, which are to be fulfilled using the systematic literature review (SLR).

The current study systematically evaluates the literature in the domain. SLRs are established approaches to conduct reviews in a structured manner (Kochar, Kalsie, & Deka, 2025). It involves synthesising evidence to appraise and understand a specific topic in any domain (Randles & Finnegan, 2023). SLRs involve systematically identifying, analysing and interpreting the available literature related to a particular topic or specific research questions. The study followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, followed by TCCM approach, to conduct a systematic literature search and review, which is detailed below. The bibliometric analysis of the data extracted from Scopus has been done to understand the macro-level emergence of the research area. It is followed by a TCCM-based SLR on a few selected papers.

The TCCM framework for developing domain-based review articles was propounded by Rosado-Serrano, Paul, and Dikova (2018). Theories (T) constitute the different theoretical underpinnings and paradigms used to explain the inter-relationships between the constructs. Contexts (C) are the circumstances that shape the research setting. Characteristics (C) refer to the elements of a construct and their relationship with other variables of interest. Methods (M) encompass details of samples, measurement scales, research designs and analytical tools (Paul, Khatri, & Kaur Duggal, 2024). TCCM helps map the theoretical and empirical breadth of a research field, allowing critical evaluation of the theoretical underpinnings, methodological approaches, characteristics and various contexts (Paul et al., 2024).

To fulfil the objectives of the study, the articles are extracted from databases such as Scopus and Web of Science to conduct a comprehensive search. The search term (“digital transformation” OR digitalisation OR digitisation OR “industry 4.0” OR “smart logistics”) AND (“artificial intelligence” OR “machine learning” OR “deep learning” OR “predictive analytics”) AND (“rail freight” OR “railway freight” OR “rail logistics” OR “rail transport”) AND (cargo OR logistics OR transportation OR “supply chain”). No Filters were applied to the bibliographic data collected from Scopus and Web of Science owing to the limited number of articles on the research topic using the key string, except for the English language. Bibliographic data of a total of 121 studies were extracted from Scopus, and 17 articles were extracted from Web of Science in the initial round of extraction. 15 articles were extracted from Google Scholar. Initially, all types of articles were assessed in preliminary screening, including final papers, conference papers, reviews and book chapters. After the extraction of the bibliographic data from the databases, erroneous and incomplete records were identified and deleted. Fifteen records were found to be duplicates in the two separate databases of Scopus and Web of Science. 116 articles were screened for relevance. And 33 articles were selected for full-fledged assessment to fulfil the objectives of the study. The PRISMA-based flow diagram is shown in Figure 1, explaining the identification, screening and inclusion of the articles.

The current study adopts a hybrid SLR design, which combines the TCCM framework with bibliometric analysis. Both approaches complement each other's findings. First, bibliometric analysis was used to examine the macro-level development of the literature on AI and digital technologies in railway systems. It was conducted on the bibliographic data extracted from Scopus, consisting of 121 entries. Secondly, the TCCM framework is applied to the 33 articles selected after adopting the PRISMA methodology.

These 33 studies were comprehensively assessed across different parameters. The information related to (1) Title (2) Objectives of the study (3) Key findings of the study (4) Theory used (5) Sector (6) Application (7) Geography (8) Technology (9) Purpose (10) Outcome (11) Methodology (12) Data sources (13) Limitation (14) Study Type was extracted in an Excel sheet. This specific information was later used to fulfil the objective of the study and apply the TCCM framework to the articles extracted.

Figure 2 below illustrates the publication trends on AI and digital technologies in railway systems based on bibliographic data extracted from the Scopus database. It suggests that the early and emerging phase of publication is from 2019 to 2022, with a single publication in 2019, 3 in 2020, 7 in 2021 and 9 in 2022. This indicates rising scholarly attention during this period. Year 2023 saw a major acceleration, with publications rising sharply to 24. 2024–2025 suggest consolidation and peak activity in publications, which remained high with 25 articles in 2024 and 32 articles in 2025, suggesting that AI-enabled railway research has moved beyond an emerging topic into a substantially established research stream. In 2026, 20 publications were published, which is still high for the year and suggests the area is still emerging and attracting substantial scholarly attention. Overall, this area of AI-enabled research in railway systems demonstrates a strong positive trajectory, with more than 80% of publications being done in 2023 to 2026.

Citation is widely acknowledged as an essential statistic for assessing the impact of an article and a journal in the scientific community. Total citation (TC) as a variable has been used in the study to look for the influential and impactful authors' sources, and countries. The types of publication sources reveal that literature is distributed across diverse sources (journals and conference proceedings). The reviewed literature covers articles published in peer-reviewed journals and conference proceedings, reflecting the interdisciplinary nature of railway AI across multiple domains. This study also includes conference proceedings in its literature review. It suggests that the field of research is in its nascent stage, which is evolving with emerging technologies such as edge AI and DT. Table 1 summarises the top sources, authors and countries.

Bibliometric Data of 121 articles extracted from the Scopus database were analysed to conduct thematic mapping across the data collected. The co-occurrence keyword network was prepared using all keywords (Minimum 5 occurrences) (Figure 3). Based on co-occurrence analysis, it can be identified that railroad transportation (143 key strength) and railroads (137 key strength) are highly occurring keywords, followed by digital transformation, with another prominent keyword being Industry 4.0. DT is also the most visible technology-based keywords in the literature. Other highly visible keywords include logistics and freight transformation, transportation, information management, sustainability and sustainable development, IoT, AI and supply chain. These keywords are clustered in four major themes. Cluster 1 brings together keywords like freight transportation, logistics, supply-chain management, Industry 4.0/5.0, IoT, circular economy, investment, competitiveness and sustainable development. Thus, it indicates an emergence of a value-chain-oriented research stream connecting digital technologies, freight efficiency and sustainable transportation. Cluster 2, being a more technology-oriented cluster, links AI, ML, DT and digitalisation with railway and rail transportation. It reflects the technological foundation of intelligent railway systems across the literature. Cluster 3 links keywords like automation, traffic control, information management, safety, intelligent systems and cybersecurity, highlighting the operational and risk-management dimensions of railway digitalisation. Cluster 4, connects digital transformation and digitisation with rail transportation and sustainability, representing a broader transitional theme linking technological transformation with sustainable railway development.

The thematic map was prepared using keyword co-occurrence analysis in Biblioshiny (Figure 4). It helps identify the maturity and importance of the themes in the bibliometric dataset. The X-axis represents the degree of relevance or centrality, and the Y-axis represents the degree of development or density. Higher centrality indicates greater importance of the research area, and higher density indicates that the theme is well developed and specialised within the literature. The upper-right quadrant shows two clusters as the motor themes, first being the digital theme digitalisation and ML cluster, which shows that substantial work on AI-enabled digital railway systems has been done. This cluster is a strongly developed cluster and core to the technology stream within railway AI research. Secondly, the position of sustainable development, logistics and freight transport management indicates it is relatively well connected to the broader railway research network and internally cohesive within the bibliometric dataset. The position of clusters as the motor themes indicates the structural importance and interconnectedness of these keywords within the research network.

The lower-right quadrant shows basic themes with high centrality and low density. These themes are foundational and still in development. The largest bubble of railroads, railway transportation and rails indicates that the theme is core to the literature and it connects with almost every other theme. However, the research remains fragmented and not highly specialised for this cluster. The second bubble of digital transformation, information management and sustainability also shows importance as an emerging foundational theme within the literature. The upper-left quadrant of niche themes indicates low centrality and high density of particular clusters or themes. The cybersecurity intelligence systems cluster indicates that researchers are actively working on this theme. However, it remains isolated from the mainstream railway AI research. The fourth quadrant in the lower left represents themes with low centrality and low density as emerging or declining themes. The port railway station transport is a very small cluster, which is not enough to interpret. Still, these clusters can be interpreted as being underdeveloped areas that require further scholarly attention. The green cluster, which is near the centre, with the words Industry 4.0, IoT and Investments, explains the theme as being moderately developed and moderately connected, which can be interpreted as a transitional theme across the literature.

Thematic evolution analysis (Figure 5) shows how the research fields have progressed over time within the area of AI-enabled systems. The analysis is based only on bibliometric data from 2019. The first period from 2019 to 2023 is when the research is primarily centred on the themes around digital transformation, Industry 4.0, rail transportation and railway operations. These broad themes evolved towards information management, maintenance and digitalisation, which are more application-oriented, by 2024. However, by 2025, these themes integrated towards advanced digital technologies within railway operations, with themes around intelligent systems, decision-making and sustainability. The themes in 2026 highlight the growing prominence of themes like DT, sustainable development, information management and change management, which reflects that themes have transitioned from isolated digital technologies towards more intelligent and integrated railway ecosystems.

TCCM framework is adopted in the study to conduct the review and fulfil the objectives of the study. The TCCM approach develops understanding of the literature through theoretical foundations, application contexts, key characteristics and methodological approaches.

Theoretical foundation in the literature is dominated by technology-driven fields, with AI and related domains forming the core basis across the literature. The prominent theories contributing to the theoretical landscape on digital transformation in railway infrastructure are discussed below. These are not explicit theories but form the conceptual base of the literature.

AI and ML form a strong base of the theoretical perspective in the literature on the topic (Tang et al., 2022; Bris-Peñalver, Verdecia-Peña, & Alonso, 2026; Phusakulkajorn et al., 2023). The system learns patterns in the data to make intelligent decisions, without explicit programming (Kushwaha, Kumar, & Harsha, 2025). In the railway context, ML algorithms include neural networks, DL models and vector machines (Tang et al., 2022; Phusakulkajorn et al., 2023). Applications include predictive maintenance, anomaly detection and operational optimisation, further enhancing system reliability (Karim, Dersin, Galar, Kumar, & Jarl, 2021; Tang et al., 2022; Thompson et al., 2025; Kushwaha et al., 2025; Meira et al., 2023).

DL is used to test non-linear relationships in large datasets by extending ML by utilising multi-layered neural networks present in it (Quan, Wang, Baihang, & Ziwen, 2025; Bris-Peñalver et al., 2026; Wang, Ji, Xu, Su, & Zhang, 2025). It supports infrastructure inspection, defect detection, environmental risk prediction and object detection, as well as time-series forecasting for equipment failures (Zhang et al., 2026; Quan et al., 2025).

DT enable predictive insights, scenario simulation and improved decision-making (Ghaboura, Ferdousi, Laamarti, Yang, & El Saddik, 2023; Kushwaha et al., 2025; Dong, Sun, Ding, & Yin, 2025). DT is based on the idea of creating a virtual (digital) replica or model of a physical system that is continuously updated with real-time data (Thompson et al., 2025; Kushwaha et al., 2025; Krmac & Djordjevic, 2024) DT have three main components: a physical entity, a digital model and data connections, which are connected in a continuous feedback loop (Alvarado et al., 2021; Kushwaha et al., 2025; Thompson et al., 2025). In the railway context, it helps monitor infrastructure, predict failures, simulate operations and optimise performance (Alvarado et al., 2021; Thompson et al., 2025; Kushwaha et al., 2025; Vatakov et al., 2022).

ITS and CPS integrate the theoretical aspects of computational intelligence with physical transportation systems (Nisandzic & Martinovic, 2022; Ghaboura et al., 2023; Bris-Peñalver et al., 2026). Application of these systems and frameworks in railways enhances safety, capacity, control and operational efficiency (Pappaterra, Flammini, Vittorini, & Bešinović, 2021; Tsyganov, 2021; Nisandzic & Martinovic, 2022; Ficzere, 2023; Thompson et al., 2025). The theoretical aspects of ITS and CPS explain how these systems facilitate real-time monitoring and automated decision-making (Nisandzic & Martinovic, 2022; Vatakov et al., 2022; Nicodeme, 2024; Bris-Peñalver et al., 2026).

Railways generate large-scale data or big data, which can be in the form of sensor data, operational logs and environmental data (Krmac & Djordjevic, 2024; Sarp, Kuzlu, Jovanovic, Polat, & Guler, 2024). The availability of sensor data, operational logs and environmental data enables advanced analytics for optimisation and forecasting in railways (Bauer, Grobler-Dębska, & Woźniak, 2024; Vatakov et al., 2022; Sarp et al., 2024). Studies like Pappaterra et al. (2021) and Bauer et al. (2024) highlight the importance of centralised data systems and open datasets for enabling AI applications in the literature.

Some studies have implicitly adopted a socio-technical perspective, discussing the dynamics between technological systems and organisational factors within the literature (Thompson et al., 2025; El Moussaoui & El Moussaoui, 2026). However, this perspective is present across the literature implicitly (Phusakulkajorn et al., 2023; El Moussaoui & El Moussaoui, 2026). The limited explicit use of such theories indicates a gap in the literature, where organisational and behavioural dimensions can be explored (Sarp et al., 2024; El Moussaoui & El Moussaoui, 2026).

The articles reviewed indicate that certain domains in the literature on AI applications in railway systems have been studied predominantly in the context of infrastructure, maintenance and operations (Tang et al., 2022; Bešinović et al., 2022; Pappaterra et al., 2021; Phusakulkajorn et al., 2023; Bris-Peñalver et al., 2026). However, there is limited attention towards freight logistics and supply chain integration (du Plessis, Gerber, Goedhals-Gerber, & van Eeden, 2025). Railway infrastructure is the most focused sub-context within the literature, followed by rolling stock maintenance and operations management (Tang et al., 2022; Kushwaha et al., 2025; Thompson et al., 2025; Phusakulkajorn et al., 2023). Sector-wise, railway infrastructure management emerged as the most studied context, which includes track monitoring, asset health management and predictive maintenance. It is followed by applications in operations, including scheduling, signalling and traffic control. Multimodal supply chains and railway freight are important areas within the research; however, these domains remain underrepresented in the literature despite its strategic importance (Vidal, Caiado, Scavarda, Ivson, & Garza-Reyes, 2022; El Moussaoui & El Moussaoui, 2026; du Plessis et al., 2025).

Railway systems comprise both passenger and freight, serving as the primary application environment for goods transportation and individual mobility services (Sarp et al., 2024; El Moussaoui & El Moussaoui, 2026). As a sub-theme, infrastructure and asset management form a critical context across the literature (Alvarado et al., 2021; Tang et al., 2022; Bris-Peñalver et al., 2026). The context focuses on predictive maintenance, condition monitoring and lifecycle optimisation, which helps improve reliability and further reduce operational costs (Alvarado et al., 2021; Thompson et al., 2025; Sarp et al., 2024; Bris-Peñalver et al., 2026). Railway operations and control, as a sub-domain, consist of scheduling, traffic management, signalling and real-time decision-making (Tang et al., 2022; Nisandzic & Martinovic, 2022; Bešinović et al., 2022; Krmac & Djordjevic, 2024). The logistics and supply chain as a context across the literature remains underexplored, though important.

Railway systems, as smart transportation systems, are integrated, intelligent mobility ecosystems at a broader level (Bešinović et al., 2022; Laiton-Bonadiez, Branch-Bedoya, Zapata-Cortes, Paipa-Sanabria, & Arango-Serna, 2022; Sarp et al., 2024). These ecosystems enhance connectivity and interoperability by using IoT, real-time data and digital platforms across transport modes (Vatakov et al., 2022; Sarp et al., 2024; Thompson et al., 2025; Laiton-Bonadiez et al., 2022). Overall, this entire framework operates within the context of an Industry 4.0 environment. Industry 4.0 environment is characterised by digitalisation, automation, CPS and data-driven decision-making, which provides a technological backbone enabling the convergence of AI, DT and advanced analytics across all railway contexts.

The geographic distribution of the literature shows alignment with technologically advanced countries (Kawasaki, 2020; Zhang et al., 2026; Kushwaha et al., 2025). Literature reflects a strong concentration of studies conducted in Europe, China and those regions that are comparatively technologically advanced (Zhang et al., 2026; Kushwaha et al., 2025). These regions actively support digital transformation initiatives and smart mobility programs (Dong et al., 2025; Sarp et al., 2024). There is a lesser representation of emerging economies within the literature, being present only in a few studies where AI adoption is examined in the context of developing railway systems (El Moussaoui & El Moussaoui, 2026). China's dominance across the literature is consistent with its position as the world's largest and most technologically advanced rail network (Dong et al., 2025; Kushwaha et al., 2025; Zhang et al., 2026).

Characteristics refer to the elements of a construct and their relationships with other variables of interest (Paul et al., 2024). Reviewed literature reveals several key characteristics that help map the literature on AI and digital transformation in the railway context.

Predictive maintenance is the most dominant characteristic emerging from the reviewed literature (Tang et al., 2022; Krmac & Djordjevic, 2024; Bris-Peñalver et al., 2026). AI and ML have been increasingly used to predict equipment failures, monitor infrastructure status and optimise maintenance schedules using historical and real-time operational data (Alvarado et al., 2021; Tang et al., 2022; Bris-Peñalver et al., 2026). Furthermore, DL and neural networks enable railway systems to move beyond reactive maintenance strategies towards predictive and prescriptive maintenance paradigms (Tang et al., 2022; Bris-Peñalver et al., 2026). Real-time monitoring systems enhance the intelligence of surveillance systems using IoT devices, sensors, computer vision and DT (Kushwaha et al., 2025; Thompson et al., 2025; Bris-Peñalver et al., 2026).

Another dominant characteristic identified is the reliance on data-driven decision-making (Sarp et al., 2024; Thompson et al., 2025; Bris-Peñalver et al., 2026). Railway systems rely on operational and environmental data for decision-making when monitoring tracks, trains and signalling systems. The literature highlights integrating intelligent surveillance systems with AI algorithms for automated inspection and anomaly detection (De Santis, Arnò, Martino, & Rizzi, 2022; Ghaboura et al., 2023; Bris-Peñalver et al., 2026). AI systems and ML algorithms help railways leverage large datasets, sensor networks, operational logs and traffic information, enhancing the accuracy of decisions (Tang et al., 2022; Sarp et al., 2024). Railways generate big data, which is captured through centralised data systems that enhance and accelerate the development of intelligent decision support systems across the railway's infrastructure (Bauer et al., 2024).

Multi-criteria optimisation as a technique has been highlighted in several studies (Geng, Zhang, Zhu, & Zhong, 1999; Tang et al., 2022; Vidal et al., 2022; Song et al., 2026). It helps solve logistics-related decisions where conflicts arise among cost efficiency, safety and sustainability. Multi-criteria optimisation incorporates approaches such as fuzzy multi-criteria decision-making (MCDM), optimisation algorithms, genetic algorithms and hybrid AI models, which the literature discusses in relation to optimising freight allocation, transport chain performance and inventory management (Geng et al., 1999; Vidal et al., 2022; Tang et al., 2022; Song et al., 2026).

Automation improves efficiency and reliability while reducing human intervention and the scope of operational errors in the processes (Nisandzic & Martinovic, 2022). AI and CPS enhance automation capabilities of the railway systems in signalling, train control, scheduling, autonomous inspection and operational management (Nisandzic & Martinovic, 2022; Sarp et al., 2024).

Sensor data, operational data, maintenance records, environmental information and simulation outputs are types of heterogenous data generated by railways (Bauer et al., 2024; Neumann et al., 2026; Sarp et al., 2024; Bris-Peñalver et al., 2026). Such data can be integrated and centralised, leading to its usage in supporting holistic railway management (Alvarado et al., 2021; Neumann et al., 2026). Sustainability and operational efficiency have been discussed in the literature multiple times (Alvarado et al., 2021; Thompson et al., 2025; Sarp et al., 2024; Rohovyi et al., 2025; Bris-Peñalver et al., 2026). Sustainable dimensions include reduction in energy consumption, optimisation of resource utilisation, minimisation of delays and improved asset lifespan. The interconnectedness between AI, IoT, DT and CPS operates collaboratively in modern railway systems (Krmac & Djordjevic, 2024; Thompson et al., 2025; Sarp et al., 2024; Bris-Peñalver et al., 2026). Overall, the identified characteristics indicate that railway digital transformation is evolving beyond individual AI applications towards integrated, intelligent transportation systems that enhance predictive maintenance and autonomy and make railways more sustainable.

Research Methods involve a combination of theoretical, empirical, simulation-based and riveted data-driven approaches used in the literature (Tang et al., 2022; Kushwaha et al., 2025; Ahmadov, Manafov, Karimov, & Mustafayeva, 2025; Thompson et al., 2025; Bris-Peñalver et al., 2026). However, a significant portion of the studies adopt SLR as a methodology to synthesise existing literature. These studies mainly focus on areas such as predictive maintenance, DT, Industry 4.0 technologies and AI (Tang et al., 2022; Ghaboura et al., 2023; Krmac & Djordjevic, 2024; Kushwaha et al., 2025; Thompson et al., 2025; Sarp et al., 2024; Bris-Peñalver et al., 2026). Some studies have employed conceptual and theoretical frameworks to explain the role of AI and digital transformation in railway systems (Ghaboura et al., 2023; Kushwaha et al., 2025; Thompson et al., 2025; Sarp et al., 2024; Bris-Peñalver et al., 2026). These frameworks explain the interactions among AI, IoT, DT and CPS. Conceptual studies across the literature relate smart railway systems, digital transformation strategies and intelligent transport ecosystems (Laiton-Bonadiez et al., 2022; Sarp et al., 2024; Thompson et al., 2025; Bris-Peñalver et al., 2026). ML and DL technologies have been empirically tested with large-scale railway datasets for predictive maintenance, anomaly detection, infrastructure monitoring and operational optimisation. Simulation and optimisation methodologies, as components of the reviewed literature, are used particularly in studies related to railway operation, set scheduling, signalling systems and logistics management.

Multiple studies have used different types of datasets, including primary as well as secondary datasets and bibliometric data of a large number of articles (Pappaterra et al., 2021; Tang et al., 2022; Laiton-Bonadiez et al., 2022; Zhang et al., 2026). Primary datasets include real-world sensor and operational data being predominantly used in empirical studies. Sensor modalities include vibration sensors, acoustic sensors, inertial measurement units, GPS, Light Detection and Ranging (LiDAR) systems, thermal cameras and pressure sensors (Tang et al., 2022; Neumann et al., 2026; Sarp et al., 2024; Bris-Peñalver et al., 2026; Zhang et al., 2026). Secondary data sets are compiled from company reports, national-level statistics and operational records, which have been used in different case studies.

The major limitation for the empirical studies remains the scarcity of the datasets available in the context of AI applications in railway datasets. Most existing public datasets focus on infrastructure monitoring (track geometry, visual inspection) and traffic data, with severe under-representation of freight operations, passenger flow and energy management data (Pappaterra et al., 2021; Tang et al., 2022; Krmac & Djordjevic, 2024; El Moussaoui & El Moussaoui, 2026).

This section integrates the TCCM summary to identify recurring configurations across four dimensions. The integration helps to examine how particular theoretical perspectives are associated with specific railway application domains, geographical contexts and methodological approaches. Table 2 below maps the intersections, revealing dominant research trajectories and fragmentation within the literature.

The integrated TCCM synthesis reveals three broad trajectories in the railway AI literature. The technology-integration trajectory is centred around DT, CPS and Industry 4.0/5.0. It is predominantly associated with maintenance, asset management and infrastructure applications, with systematic reviews, conceptual frameworks and modelling approaches being applied in these themes. The algorithmic trajectory is centred on AI, ML, DL, explainable AI and prognostics, and is more strongly oriented towards predictive maintenance, condition monitoring, fault detection and other data-intensive applications, with greater reliance on experimental and model-development approaches. Third, a decision-optimisation trajectory involving MCDM, fuzzy logic and intelligent transport systems is more visible in freight, logistics, scheduling and operational decision-making and is primarily supported by modelling, simulation and case-based approaches. Importantly, freight and logistics do not constitute a single mature TCCM configuration. Although it appears across several theoretical perspectives – including DT, MCDM, AI/ML and supply-chain or socio-technical perspectives – the associated methodologies range from modelling and simulation to experimental, survey and review approaches. This theoretical and methodological dispersion suggests that freight logistics remains not only relatively underexplored but also fragmented, with limited cumulative development around a consistent theoretical and methodological foundation. Overall, the integrated TCCM analysis (Figure 6) shows that the evolution of railway AI research is shaped by distinct TCCM configurations rather than a unified research paradigm, highlighting opportunities for future studies to connect emerging AI technologies with explicit theoretical foundations and context-specific empirical validation.

AI research in railway systems has primarily evolved through technology-centric themes. DT, ML, CPS and Industry 4.0/5.0 have been initial prominent themes focusing upon technological capability, system integration, monitoring and automation. Three trajectories emerge from combined bibliometric and TCCM findings. These are technology integration through DT, CPS and Industry 4.0/5.0; algorithmic applications involving AI, ML and DL; and decision-optimisation using MCDM, fuzzy logic and intelligent transport systems. This suggests that the literature has established considerable technological relevance but remains comparatively limited in explaining the organisational, institutional and strategic mechanisms through which AI creates sustained value for railway organisations. Thus, future research needs to move beyond demonstrating technical feasibility towards examining how AI capabilities are developed, governed and translated into operational, organisational and sustainability outcomes. There exists contextual and methodological fragmentation in the literature on AI and digital transformation in Railways. Freight and logistics emerge as highly connected themes in the bibliometric network, particularly through their association with sustainability, supply-chain management, Industry 4.0 and digital transformation; however, the TCCM analysis indicates that research in this area remains dispersed across different theoretical perspectives, technologies and methodological approaches. Similarly, the geographical concentration of studies limits the generalisability of existing findings, while the scarcity of longitudinal and multi-context empirical research makes it difficult to assess the sustained effects of AI adoption. These findings highlight the need for comparative research across different railway systems, particularly emerging and resource-constrained economies, where institutional arrangements, infrastructure capabilities and data availability may shape technology adoption differently. Overall, the literature would benefit from greater integration of technological, organisational and contextual perspectives, supported by longitudinal, mixed-method and empirically validated research designs. Such an approach can advance the field from a predominantly technology-centred understanding of AI towards a more comprehensive explanation of intelligent, resilient and sustainable railway transformation.

The results of the study demonstrate that this thematic expansion lacks coherence and theoretical foundation, limiting an integrated understanding of how AI-enabled technologies create value across railway systems. The identified gaps and corresponding future research have been discussed below.

The reviewed literature is strongly oriented towards technological perspectives, while organisational and socio-technical perspectives remain largely implicit within the literature and theoretical grounding. Future research should focus on examining how and under what organisational, institutional and contextual conditions AI creates value in railway systems rather than asking what AI technologies can do.

The fragmentation of freight and logistics research is a major gap identified in the review. Future studies should therefore focus on investigating AI-enabled fleet optimisation, multimodal transportation, railway inventory management, last-mile logistics, stakeholder coordination and capacity allocation, combining the two fragmented sets of literature within AI-enabled railway systems.

The findings indicate that research is concentrated in technologically advanced geographies. Future studies should focus on examining railway AI adoption in emerging and developing economies.

This gap pertains to the concerns in the methodological approach across the literature. Longitudinal research remains scarce. Therefore, future research should employ longitudinal panel, quasi-experimental or mixed-method research designs to examine AI implementation over time.

Research areas such as passenger experience, revenue management, demand forecasting and autonomous railway operations are comparatively less developed. Therefore, future research should examine how AI can create value across different operational and commercial functions.

The literature is increasingly moving from isolated technologies to interconnected digital ecosystems. This evolution shows the growing prominence of DT, information management, sustainability and change management in the later period of the literature. Therefore, future research should focus on examining technology configurations rather than individual technologies in isolation. This will help in a broader understanding of integrated AI-enabled railway ecosystems (see Table 3).

This study systematically reviews the literature on AI and digital transformation in railway systems using thematic analysis of Bibliometric data followed by the TCCM framework. The study followed PRISMA guidelines to extract articles for inclusion in the review. Bibliometric data of 121 articles extracted from the Scopus database were used to conduct thematic analysis, following which 33 studies published between 1999 and 2026 were included in this TCCM-based analysis from the Scopus and Web of Science databases. The article aims to enhance the understanding of the evolution, current state and future directions of the AI-enabled railway research domain.

Digital transformation, Industry 4.0 and DT constitute the dominant keywords across the literature. The thematic evolution suggests a gradual shift from traditional railway digitalisation towards interconnected railway ecosystems, reflecting the increasing integration of AI, automation, logistics and data-driven decision-making. Further, the application of TCCM on the literature reveals that it is driven by technological perspectives with DT, CPS and Industry 4.0/5.0 primarily associated with maintenance, asset management and infrastructure serving as the primary theoretical foundations in the domain. The study further highlights that freight and logistics, despite increasing thematic importance, remain theoretically and methodologically fragmented. Emerging economies and several application areas beyond maintenance and infrastructure are also comparatively underrepresented.

There is a balanced mix of methodologies being adopted in the literature; however, the empirical evidence remains fragmented due to scarcity of comprehensive railway datasets, which pose a major constraint in advancing AI research and deployment.

Based on the findings of the study, an AI-enabled digital transformation framework has been prepared below in Figure 7.

The figure above explains how AI and digital technologies are applied in railway systems, generating positive outcomes and benefits for multiple stakeholders. The key enablers play an important role in shaping the overall outcomes of the digital transformation. Contextual factors moderate the entire framework. The overall benefits are intelligent and autonomous railway ecosystems, enhanced asset Utilisation, sustainable and resilient railways, and economic and environmental benefits. From the perspective of value, the entire discussion can be summarised as “the technological development enhances the capability of the system, which further enhances the operational performance, creating strategic value across the ecosystem for multiple stakeholders”.

Despite the significant progress in the literature, several important gaps remain. The literature is heavily skewed towards technical and engineering-oriented studies, and it gives limited attention to the organisational, managerial and socio-technical aspects that explain how AI creates value within railway organisations. Within the literature, supply chain management and freight logistics remain underexplored despite their strategic importance. Studies also concentrate on technologically advanced economies, limiting the generalisability of the findings to emerging economies.

This study contributes to the literature by reviewing fragmented studies using a structured TCCM-based synthesis of AI applications in railway systems. The findings identify prevailing themes, methodological trends and emerging trends in the domain. Overall, future research should strengthen organisational and socio-technical theorisation, expand freight and emerging-economy research, adopt longitudinal and mixed-method approaches, and examine integrated technology configurations.