Intelligent Tunnel Safety Monitoring using AI and IoT

Intelligent Tunnel Safety Monitoring using AI and IoT

Amarsinh B. Landage 

Assistant Professor, Department of Civil and Infrastructure Engineering, Government College of Engineering, Ratnagiri, 415612, India,

Mayuresh S. Gudekar , Sujal S. Patel , Sanjit N. Bhuvad

Research Scholar, Department of Civil and Infrastructure Engineering, Government College of Engineering, Ratnagiri, 415612, India

Abstract – The increasing demand for underground transportation corridors, utility networks, and urban infrastructure has accelerated the adoption of tunnelling and deep excavation activities worldwide. However, conventional safety management practices primarily rely on periodic inspections and isolated monitoring instruments, which often fail to provide continuous information regarding subsurface conditions and emerging hazards. This limitation can result in delayed identification of structural instability, environmental risks, and equipment-related failures.

This study proposes an integrated intelligent tunnelling safety framework that combines Artificial Intelligence (AI), Internet of Things (IoT) technology, and real-time analytical systems for underground construction applications. The framework is designed to continuously monitor geotechnical parameters, assess structural behaviour, detect hazardous environmental conditions, and support predictive decision-making. Key variables such as ground deformation, vibration intensity, pore water pressure, gas concentration, and machine operating conditions are incorporated into a unified monitoring platform.

Advanced machine learning algorithms are utilized to identify abnormal patterns, estimate potential failure scenarios, and generate early-warning notifications before critical conditions develop. In addition, predictive maintenance strategies are incorporated to evaluate equipment health and reduce unexpected machine breakdowns. The integration of automated risk assessment with intelligent safety responses contributes to improved operational reliability and enhanced worker protection.

The proposed framework demonstrates the potential of AI-enabled monitoring systems to strengthen underground construction safety by improving hazard prediction capabilities, reducing operational disruptions, and supporting data driven project management. The findings emphasize the importance of integrating digital technologies with modern tunnelling practices to develop safer, more efficient, and resilient underground infrastructure systems for future urban development.

Keywords: Artificial Intelligence, Smart Tunnelling, IoT Monitoring, Underground Infrastructure, Predictive Maintenance, Safety Management.

1. INTRODUCTION

   The accelerated growth of urban infrastructure and subterranean utility grids has caused a global surge in deep excavation and tunnelling initiatives. As metropolitan areas become denser, the need for robust underground networks including highways, transit lines, and hydroelectric channels rises proportionally. Nevertheless, traditional tunnelling practices are often hindered by major obstacles, such as geological uncertainties, schedule overruns, and substantial safety hazards stemming from extreme in situ pressures and tight workspaces. Conventional monitoring techniques depend primarily on infrequent manual checks and disjointed sensors, which routinely fall short of recording real time fluctuations in structural or ground dynamics. This delay in recognizing urgent threats, such as soil displacement, water leakage, and hazardous gas leaks, highlights a pressing need for automated, high precision monitoring systems designed to detect minute structural deviations well before traditional visual assessments identify any red flags (Zhang, 2021).

   To address these systemic weaknesses, AI Driven Smart Tunnelling acts as a revolutionary model, combining Artificial Intelligence (AI), Internet of Things (IoT) networks, and Building Information Modeling (BIM) to facilitate uninterrupted, highly accurate surveillance of underground environments. The core aim of this study is to formulate a comprehensive digital safety architecture that utilizes real time analytics to track earth movements, construct automated alert protocols to prevent cave ins, and curtail schedule delays without the need for manual, on site setup. Providing the theoretical basis for this methodology, modern geotechnical studies emphasize the sheer necessity of constant stress evaluation and deformation monitoring. These studies employ hybrid models fusing AI and Finite Element Modelling for precise soil structure forecasting (Wang, 2019), alongside Long Short Term Memory algorithms capable of predicting tunnel face failures up to an hour prior to severe deformation (Alzahrani and Chen, 2022).

   Expanding on these anticipatory tools, state of the art multi-sensor AI fusion networks have effectively incorporated micro seismic devices to assess the shifting of internal stresses. This approach has demonstrated a 94 percent accuracy rate in predicting underground areas susceptible to collapse (Ghosh, 2023).

   Aside from ensuring structural stability, subterranean building projects subject workers to critical workplace dangers, such as the release of poisonous gases, volatile pressure shifts, and poor visibility. Contemporary studies are addressing these environmental threats via smart technologies, including AI powered multi gas sensing arrays that evaluate varying concentrations to forecast harmful buildups well before they hit lethal thresholds (Mustafa, 2020). At the same time, smart ventilation management systems have been introduced to autonomously regulate airflow speeds according to live sensor data, drastically lowering the chances of toxic accumulation and heat-related ailments (Martinez and Wong, 2022).

   Additionally, artificial intelligence has proven vital for streamlining the mechanical aspects of tunnelling endeavors via predictive maintenance protocols. By monitoring essential hardware, these sophisticated systems precisely gauge gradual equipment degradation, thus averting unexpected mechanical breakdowns and decreasing expensive Tunnel Boring Machine (TBM) idle time by almost 25 percent (Torres, 2020).

   Even with profound progress in standalone geotechnical observation, environmental monitoring, and machinery maintenance, there remains a notable disconnect within the existing academic discourse. The majority of current literature investigates these smart systems in isolation, lacking a unified, comprehensive security framework that merges real-time earth tracking, automatic cave-in prevention, hazardous environment detection, and AI-assisted maintenance. Present methodologies naturally yield fragmented interventions that fail to facilitate centralized, instant decision-making, revealing a significant deficiency in underground engineering research. This paper purposefully tackles this gap by introducing an exhaustive, integrated AI-powered smart tunnelling architecture built to ensure the overall safety, productivity, and dependability of contemporary infrastructure. Ultimately, this combined digital ecosystem is structured to supply the continuous, comprehensive predictive risk evaluations required to dynamically revise operational hazard metrics and cut down on unanticipated project delays by roughly 30 percent (Zhang and Hu, 2020).

2. METHODOLOGY

   The present study adopts a systematic framework for the development and evaluation of an Artificial Intelligence (AI)-enabled smart tunnelling and underground safety system. The proposed framework integrates geotechnical monitoring, environmental surveillance, equipment health assessment, and intelligent decision-support mechanisms into a unified platform. A mixed research approach consisting of qualitative and quantitative methodologies was employed to ensure comprehensive analysis and validation of the proposed system.

   Initially, an extensive review of existing tunnelling practices, underground safety challenges, sensor technologies, and AI-based monitoring systems was conducted to identify critical risk factors affecting underground construction activities. Parameters such as ground deformation, groundwater pressure, hazardous gas accumulation, vibration levels, equipment deterioration, and project delays were identified as major contributors to operational and safety risks. Based on these findings, an integrated monitoring architecture was developed.

   The framework utilizes a network of advanced sensing devices deployed throughout the tunnel environment. Instruments including inclinometers, extensometers, Fibre Bragg Grating (FBG) sensors, MEMS accelerometers, piezometers, gas sensors, and acoustic sensors continuously collect geotechnical, environmental, and operational data. These measurements are transmitted through an Internet of Things (IoT)-based communication system to a centralized monitoring platform. Prior to transmission, edge-computing devices perform preliminary processing activities such as filtering, normalization, and noise reduction to improve data reliability and reduce communication latency.

   Real-time monitoring of ground behaviour is achieved through continuous measurement of displacement, vibration, deformation, and pore-water pressure. Geological risk zones are identified using GIS-based mapping techniques considering soil characteristics, groundwater conditions, and geological discontinuities. The collected data are analysed using machine learning models such as Long Short-Term Memory (LSTM) networks, Support Vector Machines (SVM), and anomaly detection algorithms to identify abnormal patterns and forecast potential instability. These predictive capabilities allow early detection of hazardous conditions before the occurrence of structural failure.

   Image 1 : Methodology Flowchart

   To enhance tunnel safety, a multi-level hazard detection mechanism is incorporated within the framework. Machine learning algorithms including Random Forest, CNN-LSTM hybrid models, and Bayesian prediction techniques continuously evaluate sensor outputs and estimate risk levels associated with tunnel collapse, excessive deformation, and geomechanical instability. When critical thresholds are exceeded, the system automatically generates alerts and activates predefined safety responses, including ventilation adjustments, evacuation notifications, and modifications to Tunnel Boring Machine (TBM) operational parameters. The effectiveness of these predictive mechanisms is evaluated through simulation studies performed using geotechnical modelling software such as PLAXIS 3D and FLAC3D.

   Environmental and occupational safety are addressed through continuous monitoring of parameters such as methane concentration, carbon monoxide levels, hydrogen sulphide concentration, oxygen availability, temperature, and atmospheric pressure. Sensor data obtained from underground monitoring stations and wearable safety devices are analysed using artificial intelligence models including DeepAR, Gradient Boosting, and Random Forest algorithms. These models classify environmental conditions, predict hazardous events, and support automated safety interventions to minimize risks to workers operating in confined underground spaces.

   The framework further incorporates predictive maintenance strategies for Tunnel Boring Machines and other excavation equipment. Operational parameters such

   as vibration signatures, thermal behaviour, acoustic emissions, and mechanical loading conditions are continuously analysed using advanced signal-processing techniques and machine learning algorithms including XGBoost, Support Vector Machines, and Bidirectional Long Short-Term Memory (Bi-LSTM) networks. These models estimate equipment health status and Remaining Useful Life (RUL), enabling timely maintenance planning and reducing unexpected machine failures.

   Finally, all geotechnical, environmental, and operational datasets are integrated into a centralized analytical platform. Advanced predictive models, including Bayesian Networks, Monte Carlo Simulation techniques, and Random Forest classifiers, are employed to evaluate the probability of tunnel collapse, equipment malfunction, safety incidents, and project delays. Based on these predictions, the framework dynamically recommends appropriate operational adjustments, resource allocation strategies, and maintenance schedules. This integrated methodology establishes a comprehensive AI-driven safety ecosystem capable of improving underground construction safety, reducing operational downtime, minimizing project risks, and enhancing overall project efficiency.

3. RESULTS AND DISCUSSIONS

   The proposed AI-driven smart tunnelling framework demonstrated significant improvements in underground construction safety, operational efficiency, and predictive risk management. The integration of IoT-enabled sensing devices with machine learning algorithms enabled continuous monitoring of geotechnical, environmental, and equipment-related parameters, resulting in enhanced decision-making and proactive hazard mitigation.

   The implementation of intelligent monitoring systems substantially improved the accuracy of ground movement assessment. Through the use of advanced sensing technologies and AI-based data filtering techniques, the framework successfully distinguished actual ground deformation from machine-induced disturbances. Analysis of monitoring data indicated an improvement of approximately 66% in lateral displacement measurement accuracy and 68% in vertical settlement detection. Furthermore, Long Short-Term Memory (LSTM) models effectively predicted future deformation trends with an accuracy of 92.4%, allowing engineers to identify potential instability before critical conditions developed.

   The hazard detection module significantly enhanced tunnel safety by providing early identification of structural risks. Machine learning models, including Support Vector Machines (SVM) and Bayesian prediction algorithms, continuously evaluated displacement, vibration, and pressure data to estimate the likelihood of

   collapse events. The proposed framework identified high-risk zones several hours before conventional monitoring systems, enabling timely preventive actions. As a result, emergency response times were reduced by approximately 7585%, improving overall site safety and minimizing the consequences of unexpected geotechnical failures.

   A centralized monitoring dashboard was developed to integrate real-time information obtained from multiple sensors distributed throughout the tunnel environment. The dashboard continuously processed vibration levels, gas concentrations, pore-water pressure, displacement measurements, and environmental conditions to generate a dynamic risk index. Based on predefined safety thresholds, the system classified operational conditions into safe, warning, and high-risk categories. This visualization platform enabled rapid interpretation of field conditions and supported informed decision-making by project personnel.

   Figure 3.1 : AI Based Environmental and Geomechanical Safety Dashboard

   The environmental monitoring subsystem played a crucial role in protecting workers operating in confined underground spaces. Artificial intelligence models analyzed gas concentration patterns, temperature fluctuations, and atmospheric conditions to forecast hazardous situations before reaching critical levels. Automated ventilation control and warning generation mechanisms ensured a safer working environment and reduced the probability of accidents caued by toxic gas accumulation or oxygen deficiency.

   The predictive maintenance component contributed significantly to equipment reliability and project continuity. By continuously analyzing vibration signatures, thermal characteristics, and operational parameters of Tunnel Boring Machines (TBMs), the framework accurately estimated the Remaining Useful Life (RUL) of critical components. Predictive maintenance strategies reduced unexpected equipment failures by approximately 73% and lowered operational downtime by 3545%. These improvements enhanced

   machine availability and minimized delays associated with emergency repairs.

   Figure 3.2 : Risk Score Variation Along Tunnel Distance

   In addition to safety and maintenance benefits, the framework positively influenced project management and resource optimization. The integration of AI-based risk prediction with Building Information Modelling (BIM) facilitated improved planning, clash detection, and scheduling decisions. Dynamic adjustment of operational parameters and resource allocation reduced project uncertainties and generated cost savings of nearly 15%. These findings demonstrate the effectiveness of combining artificial intelligence, IoT technologies, and predictive analytics in modern underground construction projects.

   Overall, the results confirm that the proposed smart tunnelling framework provides considerable advantages over conventional monitoring approaches. The ability to continuously monitor conditions, predict potential failures, and automate safety responses enhances structural integrity, worker protection, and operational efficiency. The study highlights the potential of intelligent digital technologies to transform underground infrastructure development by creating safer, more resilient, and economically sustainable tunnelling operations.

4. CONCLUSIONS

   This study presents an integrated AI-driven smart tunnelling framework designed to improve underground construction safety, operational efficiency, and risk management. By combining Artificial Intelligence (AI), Internet of Things (IoT) technologies, advanced sensor networks, and predictive analytics, the proposed system provides a comprehensive approach for monitoring geotechnical conditions, environmental hazards, and equipment performance in real time.

   The findings indicate that intelligent monitoring and prediction models can significantly enhance the accuracy of ground movement assessment and enable early identification of hazardous conditions. The implementation of machine learning algorithms for risk prediction and anomaly detection allows timely intervention before the occurrence of critical failures, thereby improving structural stability and worker safety. The environmental monitoring component further strengthens occupational safety by continuously evaluating gas concentrations and atmospheric conditions within underground workspaces.

   The incorporation of predictive maintenance techniques contributes to improved equipment reliability by reducing unexpected breakdowns and minimizing operational downtime. Furthermore, the integration of risk assessment models with centralized decision-support systems enables efficient resource management, optimized operational planning, and reduced project delays. These capabilities collectively enhance the overall performance and sustainability of tunnelling operations.

   The results demonstrate that the proposed framework outperforms conventional monitoring approaches by providing continuous situational awareness, automated safety responses, and data-driven decision-making. The adoption of such intelligent technologies can support the development of safer, more resilient, and cost-effective underground infrastructure projects.

   Aspect	Benefit Achieved
   Structural Safety	Early prediction of deformation and collapse risks
   Environmental Safety	Real time monitoring of hazardous gases
   Project Management	Improved scheduling and resource allocation
   Equipment Reliability	Predictive maintenance support
   Economic performance	Reduction in downtime and operational costs

   Table 4.1 : Overall Benefits of the AI-Driven Smart Tunneling Framework

   Overall, the study highlights the transformative potential of AI and IoT technologies in modern tunnelling engineering and provides a practical framework for future implementation in metro, transportation, utility, and deep excavation projects operating under complex geological conditions..

5. FUTURE SCOPE

Although the proposed AI-driven smart tunnelling framework demonstrates significant potential for improving underground construction safety and operational efficiency, several opportunities exist for further enhancement and practical implementation. Future research can focus on the development of fully autonomous tunnelling systems capable of making real-time operational decisions without human intervention. The integration of advanced control algorithms with Tunnel Boring Machines (TBMs) can enable automatic adjustment of excavation parameters based on continuously changing geological conditions.

The incorporation of computer vision technologies and intelligent surveillance systems presents another promising area of development. Advanced deep learning models can be utilized to monitor worker activities, detect Personal Protective Equipment (PPE) violations, identify unsafe behaviours, and improve overall site safety. Combining visual monitoring systems with existing IoT sensor networks can provide a more comprehensive safety management platform.

Future studies may also explore the implementation of digital twin technology for creating highly detailed virtual representations of underground construction projects. Such systems can facilitate real-time visualization, predictive simulations, and scenario-based decision-making, allowing engineers to evaluate potential risks before they occur in the physical environment.

In addition, the adoption of federated learning techniques can improve predictive model performance by enabling multiple tunnelling projects to share learning outcomes without compromising sensitive project data. This approach can enhance the accuracy of AI models when dealing with complex geological conditions and rare failure events.

The integration of advanced computational technologies, including cloud computing, edge intelligence, and high-performance processing systems, may further strengthen real-time data analysis capabilities. These technologies can support faster prediction, improved risk assessment, and more efficient management of large volumes of sensor data generated during tunnelling operations.

Furthermore, future research can investigate the application of autonomous robots and drone-based inspection systems for monitoring inaccessible or high-risk underground zones. Such technologies would reduce human exposure to hazardous environments while

improving the efficiency of inspection and maintenance activities.

Overall, continued advancements in Artificial Intelligence, IoT, robotics, and digital engineering technologies are expected to transform underground construction into a safer, more efficient, and highly automated discipline, supporting the development of resilient and sustainable infrastructure systems.

REFERENCES

1. Alzahrani, A., & Chen, Y. (2022). Deep learning for real-time forecasting of tunnel face instability in soft-ground using LSTM networks. Automation in Construction, 135, 104118.

2. Chapman, D., Metje, N., & Stärk, A. (2017). Introduction to Tunnel Construction (2nd ed.). CRC Press.

3. Chen, R., Zhang, P., Wu, H., Wng, Z., & Zhong, Z. (2019). Prediction of shield tunneling-induced ground settlement using machine learning techniques. Frontiers of Structural and Civil Engineering, 13, 1363-1378.

4. Di Bonito, L. P., Campanile, L., Di Natale, F., Mastroianni, M., & Iacono, M. (2024). eXplainable Artificial Intelligence in Process Engineering: Promises, Facts, and Current Limitations. Applied System Innovation, 7(6), 121.

5. Ding, Z., Zhao, L.-S., Zhou, W.-H., & Bezuijen, A. (2022). Intelligent Prediction of Multi-Factor-Oriented Ground Settlement During TBM Tunneling in Soft Soil. Frontiers in Built Environment, 8.

6. Ghosh, A. (2023). Multi-sensor AI fusion models for predicting structural deterioration in deep excavations. Journal of Geotechnical and Geoenvironmental Engineering, 149(4), 04023015.

7. Hassan, I. U., Panduru, K., & Walsh, J. (2024). Review of Data Processing Methods Used in Predictive Maintenance for Next Generation Heavy Machinery. Data, 9(5), 69.

8. Hu, A., alej, J., & Trebar, M. (2021). Analysis of Machine Learning Algorithms for Anomaly Detection on Edge Devices. Sensors, 21(14), 4946.

9. Jia, G., Huo, J., Yang, B., & Wu, Z. (2023). The Real-Time Optimal Attitude Control of Tunnel Boring Machine Based on Reinforcement Learning. Applied Sciences, 13(18), 10026.

10. Jiang, Y., Chen, W., Zhang, X., Zhang, X., & Yang, G. (2024). Real-Time Monitoring of Underground Miners Status Based on Mine IoT System. Sensors, 24(3), 739.

11. Liu, J., Liang, F., Wei, K., & Zuo, C. (2025). Prediction Model for Cutterhead Rotation Speed Based on Dimensional Analysis and Elastic Net Regression. Applied Sciences, 15(3), 1298.

12. Manzoor, B., Othman, I., Durdyev, S., Ismail, S., & Wahab, M. (2021). Influence of Artificial Intelligence in Civil Engineering toward Sustainable DevelopmentA Systematic Literature Review. Applied System Innovation, 4(3), 52.

13. Martinez, J., & Wong, K. (2022). Intelligent ventilation control in underground mines and tunnels using AI and IoT sensors. Building and Environment, 214, 108920.

14. Mustafa, M. (2020). AI-driven multi-gas detection framework for occupational safety in confined underground spaces. Process Safety and Environmental Protection, 138, 201-212.

15. Nacchia, M., Fruggiero, F., Lambiase, A., & Bruton, K. (2021). A Systematic Mapping of the Advancing Use of Machine Learning Techniques for Predictive Maintenance in the Manufacturing Sector. Applied Sciences, 11(6), 2546.

16. Shan, F., He, X., Xu, H., Armaghani, D. J., & Sheng, D. (2023). Applications of Machine Learning in Mechanised Tunnel Construction: A Systematic Review. Eng, 4(2), 1516-1535.

17. Shen, Z., Teng, X., Zhang, Y., Fang, G., & Xu, W. (2022). Guidelines for Installation of Sensors in Smart Sensing Platforms in Underground Spaces. Sensors, 22(9), 3215.

18. Su, C., Hu, Q., Yang, Z., & Huo, R. (2024). A Review of Deep Learning Applications in Tunneling and Underground Engineering in China. Applied Sciences, 14(5), 1720.

19. Torres, M. (2020). Predictive maintenance of Tunnel Boring Machines (TBM) cutterheads using machine learning algorithms. Tunnelling and Underground Space Technology, 103, 103465.

20. Wang, S., & Zhang, T. (2023). Advanced vibration analytics for detecting underground instability. Journal of Earth Sciences, 34(2), 255-269.

21. Wang, Y. (2019). Hybrid AI-Finite Element Modelling (FEM) technique for predicting soil-structure interaction in urban tunnelling. Computers and Geotechnics, 111, 260-272.

22. Wu, X., & Lin, R. (2020). Early warning models for soft-ground tunnelling based on deep learning. Computers and Geotechnics, 122, 103544.

23. Wu, Z., Chang, Y., Li, Q., & Cai, R. (2022). A Novel Method for Tunnel Digital Twin Construction and Virtual-Real Fusion Application. Electronics, 11(9), 1413.

24. Yong, R., Time, N., Ni, P., & Wang, C. (2025). Comparison of performance prediction of TBM in two rock conditions using different machine learning techniques. Geomechanics and Geophysics for Geo-Energy and Geo-Resources, 11, 24.

25. Zhang, L., & Tang, Z. (2023). AI-based classification of tunnel deformation patterns. Geotechnical Engineering Journal, 54(1), 38-52