Bibliometric Analysis of Research Progress in Tunnel Safety Based on CiteSpace

Summary

This study presents a bibliometric analysis of tunnel safety research, using Web of Science data and the CiteSpace tool to identify publication trends, collaboration, and research hotspots. The analysis reveals a focus on fire and traffic safety, shifting toward intelligent management and multi-hazard resilience, with fragmented collaboration.

Abstract

Due to the ever-increasing complexity of tunnel spaces and the diversification of associated risks, a comprehensive understanding of tunnel safety research is essential. This study employs CiteSpace to conduct a bibliometric analysis of publications on tunnel safety in the Web of Science Core Collection. The results show that the number of publications has increased steadily, with a significant surge after 2017. Central South University is the leading contributor, and Tunnelling and Underground Space Technology stands out as an important journal. The research is highly interdisciplinary, encompassing underground engineering, fire safety engineering and related disciplines, although collaboration networks among authors and institutions remain fragmented, reflecting limited synergy. Keyword analysis shows that research hotspots focus on tunnel fire safety and traffic accident analysis, with a transition from fundamental theory and risk assessment toward intelligent and system-oriented management. The field is shifting from traditional static analysis to dynamic operational safety analysis, multi-hazard coupled risk assessment, and system resilience evaluation. However, challenges remain in integrating digital technologies and transdisciplinary collaboration. Unlike traditional narrative reviews that often rely on fragmented or static assessments, this study provides a reproducible bibliometric framework for dynamically mapping the evolution of tunnel safety research. It offers methodologically structured insights for identifying emerging trends and guiding future interdisciplinary investigations.

Introduction

As a critical component of modern transportation and infrastructure networks, tunnels have increased continuously in scale and complexity worldwide. However, the unique and confined environment within tunnels presents significant safety challenges, including structural cracking¹, fire risks², traffic hazards³, operational safety concerns⁴, and issues related to whole-lifecycle safety monitoring⁵. Historical data consistently show that incidents within tunnels, especially fires, frequently lead to severe casualties and property damage⁶. These incidents not only damage the vehicles and the tunnel structure itself, but also reveal the inadequacies of existing safety response strategies⁷. Therefore, in-depth research on tunnel safety is not merely fundamental for ensuring the long-term stable operation of infrastructure, but also necessary for safeguarding public safety, optimizing emergency management systems⁸, promoting sustainable development⁹, and improving societal well-being.

Early research on tunnel safety can be traced back to the 1990s. The research topics mainly focused on fundamental issues such as the accumulation of flammable gases and personnel evacuation in tunnels. Subsequently, the field experienced significant progress, with numerous scholars conducting extensive and valuable research. The research scope gradually expanded to include issues such as tunnel accident mechanisms, fire protection system design, construction safety control, and structural design optimization. Since the beginning of the 21^st century, particularly over the past two decades, with the rapid advancement of digitalization and intelligent technologies, the research scope further evolved toward frontier areas such as structural stability monitoring^10,11, accident risk prediction, tunnel lighting systems, driver behavior, and driver information interaction systems. Existing research on tunnel safety can be broadly categorized into three main aspects.

First, some studies focus on the transmission mechanisms of fire disaster chains under multi-factor coupling conditions¹². These studies employed methods such as statistical accident analysis¹³, case-based retrospective studies, and probabilistic risk assessment (PRA) to investigate accident causation, risk evaluation¹⁴, fire evolution mechanisms, evacuation simulation¹⁵, and structural safety monitoring. Second, a number of studies focus on the enhancement of proactive prevention and control capabilities in dynamic environments. Many scholars have utilized computational fluid dynamics (CFD) simulations¹⁶, BIM-GIS integrated modeling, and multi-source sensor network technologies to analyze fire smoke propagation, visibility attenuation, dynamic evacuation path optimization, and health monitoring and early warning during tunnel operation¹⁷. These studies have further optimized tunnel ventilation and smoke exhaust systems¹⁸, improved intelligent lighting configurations¹⁹, and advanced the exploration of human behavior modeling and digital twin-driven closed-loop management. Third, certain studies primarily focus on the application of advanced technologies and the pursuit of sustainable tunnel development. Some scholars have employed digital and intelligent technologies such as Virtual Reality (VR)²⁰and Building Information Modeling (BIM)²¹for tunnel blasting²², underground void detection, and excavation support. These studies also explore the coordinated development of environmental, social, and economic factors during tunnel engineering processes, with a focus on sustainability²³. In addition, some scholars have applied bibliometric methods to review research on ground settlement in tunneling²⁴, conduct visual analyses of infrastructure inspection²⁵, and examine the relationship between tunnel lighting and low-carbon development²⁶.

Tunnel safety has evolved into a comprehensive and interdisciplinary research field^6,26. However, systematic and integrated studies are still lacking, and the evolution of the knowledge structure and identification of emerging trends in this field are still insufficiently clear, making it difficult to determine future research directions. Narrative reviews are valuable for interpreting mechanisms, comparing engineering practices, and synthesizing expert knowledge. However, they are often limited by the subjectivity of literature selection and by their difficulty in quantitatively tracing the temporal evolution of large research fields. Bibliometric analysis complements narrative review by providing transparent retrieval criteria, repeatable data-processing procedures, and quantitative indicators of collaboration, centrality, clustering, and burst evolution. In this study, CiteSpace was used not as a substitute for expert interpretation but as a visualization and quantitative-mapping tool to identify the intellectual structure and emerging themes of tunnel safety research. The scope of this study was limited to tunnel safety. This scope included fire safety, smoke control, evacuation, traffic safety, driver behavior, tunnel lighting, ventilation, environmental control, monitoring, emergency management, risk assessment, and system resilience during tunnel operation. Studies primarily concerned with tunnel excavation safety, shield-tunnel construction prediction, blasting, ground settlement, construction support, and lining-crack diagnosis were excluded unless they explicitly addressed operational safety outcomes. These exclusions were applied because construction-stage geotechnical risk and operation-stage safety management differ substantially in mechanisms, data sources, evaluation indicators, and engineering interventions^27,28.

To systematically clarify the characteristics and trends of research in the field of tunnel safety, this study conducts a bibliometric analysis using the CiteSpace visualization tool based on data from the Web of Science Core Collection (WoSCC), excluding Chinese-language publications and various informal research materials. The analysis covers publication output characteristics, major journals, collaboration networks among core authors and institutions, the distribution and clustering of research keywords over time, and emerging frontier directions²⁸. In contrast to conventional review approaches, this study establishes a quantitative, reproducible, and visualization-based bibliometric protocol. It integrates multi-dimensional metrics, such as keyword burst detection, timeline clustering, and co-occurrence centrality analysis. Using these metrics, the study constructs a comprehensive and dynamic knowledge framework that systematically clarifies the field’s intellectual structure and thematic evolution. This study has two primary analytical objectives. The first objective is to construct a comprehensive and dynamic knowledge framework using the CiteSpace visualization tool, thereby overcoming the fragmented and static nature of traditional review assessments. The second is to investigate whether tunnel safety research has undergone identifiable phase transitions, specifically a shift from conventional fundamental theories and static risk analysis toward an intelligent, systematic, and multi‑hazard coupled dynamic safety management system.

Protocol

This study conducted a systematic bibliometric analysis to identify and visualize research trends in the field of tunnel safety from 1992 to 2026.

Literature search and data collection
Access the Web of Science Core Collection (WoSCC) database. The search strategy was performed using the following topic-specific query: TS = [(“tunnel” and “tunnel safety”)]. After the query was entered, the “Search” button was clicked to execute the literature search. The results were filtered by selecting “Articles”, “Review Articles”, and “Early Access Publications” as the document types. A total of 15861 articles and review articles published between 1992 and 2026 were collected. This study focuses on core areas such as tunnel fire safety, road safety, traffic safety, and personnel evacuation, aiming to review the research progress and hotspots in overall tunnel safety. Therefore, publications related to excavation safety of tunnels, safety prediction of shield tunnels, and safety assessment of tunnel lining cracks were excluded. Following this screening process, 428 articles were retained for further analysis. In the “Record Content” section, “Full Record and Cited References” was selected, and the filtered dataset was exported as “Plain Text” files. Each record was verified to contain complete citation and co-authorship information. The detailed screening process is presented in a flow diagram (Figure 1).

Data preprocessing
The plain text dataset exported from WoSCC was imported into a spreadsheet using the Data | From Text/CSV function. Tab and space characters were selected as delimiters to separate the text into columns. The resulting spreadsheet was used to filter, sort, and summarize the records for subsequent analyses. Records were aggregated by publication year to calculate annual and cumulative publication counts. The Journal (SO) column was filtered to retain only journals with three or more articles. Institutions were sorted by frequency and centrality to identify the top 10 institutions, and keywords were sorted by frequency and centrality to identify the top 20 keywords. Keyword clusters were sorted by size, and the six largest clusters were retained. Detailed criteria and procedures for each operation were described in the corresponding analytical subsections.

For the analysis of the number evolution of published literature, the Publication Year (PY) column was filtered in the spreadsheet software, and the values were converted to numeric format. The records were then aggregated by publication year to calculate both annual and cumulative publication counts. A publication trend graph was generated by constructing a bar chart for annual output and overlaying a line chart for the cumulative total (Figure 2). For the analysis of distribution patterns of publication journals, the Journal (SO) column was isolated, and the number of records for each journal was calculated. Publication frequency was calculated for each journal between 1992 and 2026 by grouping records according to journal title and sorting the results in descending order. Journals with three or more articles were retained, resulting in 28 journals, and the findings were presented in a summarized table (Table 1).

Authors’ collaboration analysis
The WoSCC plain text dataset was imported and configured in CiteSpace using the Data | Import | Web of Science command, and the input and output directories were specified as appropriate. Following an in-depth review of the retrieved literature, publications from 1992 to 2004 were found to account for a relatively small proportion of the dataset. More importantly, the research content and themes of these publications were dispersed and showed weak relevance to tunnel safety. Incorporating these papers into the bibliometric analysis would have introduced excessive redundancy and obscured the identification of research hotspots and evolving trends. In addition, the literature retrieved in March 2026 could not fully represent the annual publication output for 2026. Consequently, the CiteSpace analysis was restricted to literature published between 2005 and 2025. The time span was set from 2005 to 2025 based on publication relevance, using annual time slicing with a time slice of 1 year.

CiteSpace was configured with author as the node type. The selection criterion was set to the g-index (k = 15). Term sources were set to Title, Abstract, Author Keywords (DE), and Keywords Plus (ID). Under the Pruning menu, both the Pathfinder and Pruning the merged network options were selected to simplify the network structure. The analysis was executed by clicking Go to produce the author collaboration network visualization. Font sizes and color schemes were adjusted to improve readability (Figure 3).

Institutional collaboration and top 10 institutions analysis
The node type was set to institution in CiteSpace. The selection criterion was set to the g-index (k = 25). Term sources were set to Title, Abstract, Author Keywords (DE), and Keywords Plus (ID). Under the Pruning menu, the Pathfinder and Pruning the merged network options were selected to reduce network complexity.

The analysis was run by pressing Go to generate the institutional collaboration map. Typography and color settings were adjusted to improve interpretability (Figure 4). The institutional collaboration data were saved using Output | Network Summary as HTML, CSV, RIS | Save as CSV. The CSV file was then opened in Microsoft Excel. The “Freq”, “Centrality”, and “Label” columns were combined, and a descending sort was applied using Sort & Filter | Descending. The top 10 institutions were extracted and assembled into a final table (Table 2).

Keyword co-occurrence and top 20 keywords analysis
In CiteSpace, the node type was set to keyword. The g-index was set to k = 17. Term sources were set to Title, Abstract, Author Keywords (DE), and Keywords Plus (ID). Under the Pruning menu, the Pathfinder and Pruning the merged network options were selected to trim the network. The analysis was initiated by clicking Go to obtain the keyword co-occurrence map. Font and color attributes were modified to improve visual readability (Figure 5). The keyword co-occurrence dataset was exported through Output | Network Summary as HTML, CSV, RIS | Save as CSV. The CSV file was opened in Microsoft Excel. The “Freq”, “Centrality”, and “Label” columns were combined, and a descending sort was performed using Sort & Filter | Descending. The top 20 keywords were identified and organized into a final table (Table 3).

Keyword cluster and top 6 clusters analysis
The node category was set to keyword in CiteSpace. The g-index was set to k = 17. Term sources were set to Title, Abstract, Author Keywords (DE), and Keywords Plus (ID). Under the Pruning menu, the Pathfinder and Pruning the merged network options were selected to refine the network. Keyword clustering analysis was performed by clicking Go. Cluster labels were extracted using the log-likelihood ratio (LLR) algorithm. Only clusters with a silhouette score greater than 0.7 were retained. The clustering map was generated using Cluster Label Optimization | K: Keywords, and the display settings were adjusted for optimal readability (Figure 6). Cluster details were extracted via Clusters | View Cluster Content and organized into a Microsoft Excel worksheet. The data table was opened, and the “Size” column was selected and sorted using Sort & Filter > Descending. The results were compiled into the final keyword clustering table (Table 4).

Keyword timeline cluster analysis
The node type was kept as a keyword in CiteSpace. The g-index was set to k = 17. Term sources were set to Title, Abstract, Author Keywords (DE), and Keywords Plus (ID). Under the Pruning menu, the Pathfinder and Pruning the merged network options were selected to prune the network. Keyword clustering analysis was executed by clicking Go. Cluster labels were extracted using the same LLR algorithm and silhouette score threshold (> 0.7). Timeline View was selected to generate the keyword timeline clustering map. Font and color settings were adjusted to enhance readability (Figure 7).

Keyword bursting analysis
The node type was kept as a keyword in CiteSpace. The g-index was set to k = 17. Term sources were set to Title, Abstract, Author Keywords (DE), and Keywords Plus (ID). Under the Pruning menu, the Pathfinder and Pruning the merged network options were selected for network pruning. Keyword analysis was executed by clicking Go. The Burstness option was selected, the parameter γ was set to 0.5, and the minimum burst duration was set to 2 years to perform the keyword burst detection analysis. The data were refreshed to generate the list of the top 25 keywords with the strongest citation bursts (Figure 8).

Results

Publication output and growth trends
The initial search retrieved 15861 articles and reviews related to tunnel safety published between 1992 and 2026 from the Web of Science Core Collection (as of 26 March 2026). After manually excluding publications unrelated to the research focus (e.g., studies on tunnel excavation safety, tunnel safety prediction, and tunnel lining crack safety assessment), 428 articles and reviews focusing on tunnel safety were retained for bibliometric analysis. This refined dat...

Discussion

This bibliometric analysis provides a structured overview of research trends in tunnel safety. By integrating multi-dimensional metrics including keyword burst detection, timeline clustering, and co-occurrence centrality analysis, this study constructs a dynamic knowledge framework. This framework clarifies the field’s intellectual structure and thematic evolution. Unlike traditional narrative reviews that often rely on fragmented or static assessments, this quantitative and visualization-based protocol offers a re...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
CiteSpace	Chaomei Chen	https://citespace.podia.com/	Bibliometric visualization software used to generate collaboration networks, keyword co-occurrence maps, cluster maps, timeline views, and burst-detection outputs.
Microsoft Excel	Microsoft Corporation	https://www.microsoft.com/excel	Spreadsheet software used for data cleaning, sorting, frequency calculations, and table preparation.
Web of Science Core Collection	Clarivate	https://webofscience.clarivate.cn/wos/woscc/basic-search	Bibliographic database used to retrieve the records included in the bibliometric analysis.