AI-Powered Autonomous Industrial Monitoring: Integrating Robotics, Computer Vision, and Generative AI

The integration of Artificial Intelligence (AI), Robotics, and Computer Vision is revolutionizing industrial monitoring by automating traditionally manual and time-consuming tasks, increasing both efficiency and safety. This thesis explores the integration of these technologies into a proprietary AI framework, to enhance industrial visual analysis through Autonomous Mobile Robots (AMRs), focusing on two key tasks: Automated Gauge Reading and Autonomous Surveillance in Restricted Areas. The first task addresses the challenge of continuously monitoring pressure gauges in large-scale industrial environments, a process traditionally performed by human operators and prone to errors and inefficiencies. To overcome these limitations, this study employs Boston Dynamics' Spot, a quadruped robot functioning as a mobile Internet of Things (IoT) platform capable of executing pre-configured inspection missions. Equipped with high-resolution cameras and LiDAR sensors, Spot captures images of both analog and digital gauges that are then analyzed using state-of-the-art AI models, including Optical Character Recognition (OCR), Computer Vision techniques, and Multimodal Large Language Models (LLMs), to accurately extract measurements. The study compares different approaches for extracting and interpreting gauge readings, highlighting the strengths and limitations of each model. The second task focuses on autonomous surveillance, leveraging the DJI Matrice 3TD, a drone designed for industrial applications, integrated into an alarm system to conduct real-time inspections in restricted areas. By employing Computer Vision and Deep Learning models such as YOLO (You Only Look Once) for object detection, the system enables the drone to rapidly identify anomalies, unauthorized intrusions, and potential safety hazards with high accuracy. This section examines multiple analyses performed on the dataset used, aiming to optimize results based on key reference metrics. Using a research-driven approach, all models and methodologies are evaluated based on quantitative metrics and real-world applicability. The findings demonstrate significant improvements in efficiency, accuracy, and safety, contributing to the advancement of AI-driven industrial automation and monitoring. Future research will focus on enhancing model robustness, potentially integrating multimodal sensor data, and expanding capabilities to additional industrial applications.