Industrial facilities such as oil and gas plants, chemical processing sites, power stations, and manufacturing environments require continuous inspection to ensure operational safety, equipment reliability, and early hazard detection. Traditional inspection methods rely heavily on human operators performing routine monitoring tasks in hazardous and difficult-to-access environments, exposing personnel to safety risks while increasing operational costs and limiting inspection frequency. Recent advancements in artificial intelligence, computer vision, autonomous robotics, and industrial sensing technologies have enabled the development of intelligent inspection systems capable of improving safety, operational efficiency, and maintenance effectiveness. This paper proposes INSPECTRA-X, a novel autonomous industrial inspection framework designed to support real-time hazard detection, intelligent monitoring, and autonomous navigation in industrial environments (Figure 1). The proposed framework combines multiple sensing technologies, deep learning-based anomaly detection, Simultaneous Localization and Mapping (SLAM), and autonomous charging capabilities within a unified architecture. INSPECTRA-X is designed as a modular inspection platform capable of integrating thermal cameras, RGB cameras, gas detection sensors, LiDAR-based localization systems, microphone arrays, and obstacle avoidance mechanisms to operate in complex industrial facilities. The tracked mobility system is designed to traverse uneven terrain, narrow passages, and stairways commonly encountered in industrial environments. To validate the feasibility of the proposed framework (Figure 2), two independent artificial intelligence models were developed and experimentally evaluated. The first model focuses on gas leak detection and was trained to distinguish between leakage and non-leakage conditions using industrial image datasets. Experimental evaluation demonstrated an accuracy of approximately 90%, indicating the effectiveness of deep learning techniques for identifying potential gas leakage events (Figure 3). The second model focuses on thermal anomaly detection using thermal imagery to classify heat and no-heat conditions associated with industrial equipment monitoring. To improve computational efficiency while maintaining high detection performance, a transfer learning approach based on the MobileNet architecture was implemented. The thermal anomaly detection model achieved an accuracy of 94%, demonstrating strong capability in identifying abnormal thermal signatures that may indicate overheating components, equipment faults, or safety hazards.

In addition to the artificial intelligence modules, the proposed framework incorporates a SLAM-based localization and navigation subsystem that enables autonomous movement, obstacle avoidance, path planning, and hazard location mapping in GPS-denied industrial environments. The system is also designed to support autonomous docking and recharging operations through a battery monitoring mechanism that allows the robot to return to a charging station when battery levels fall below predefined thresholds. Hazard detections are transmitted through a real-time monitoring dashboard and wireless notification system, enabling rapid operator response and improved situational awareness.

Unlike many existing industrial inspection solutions that focus on a single sensing modality or inspection task, INSPECTRA-X integrates multiple hazard monitoring capabilities within a scalable and modular framework. Furthermore, the architecture allows additional inspection functionalities, such as acoustic anomaly detection, corrosion monitoring, liquid leak detection, and predictive maintenance, to be incorporated through specialized industrial sensors and future artificial intelligence modules.

It is important to note that this study focuses on the design and evaluation of the proposed framework and its artificial intelligence detection modules. The physical implementation of the robotic platform remains future work. Nevertheless, the experimental results obtained from the gas leak detection and thermal anomaly detection models demonstrate the feasibility of integrating artificial intelligence within an autonomous industrial inspection architecture. The proposed framework provides a practical foundation for next-generation industrial monitoring systems capable of reducing human exposure to hazardous environments, improving inspection efficiency, enhancing operational safety, and supporting Industry 4.0 initiatives.

Keywords

Autonomous Inspection Robot, Artificial Intelligence, SLAM, Industry 4.0.