Open Collections

Abstract

Advances in computer vision and automated data collection methods, such as the integration of physical and computational components in cyber-physical systems, have provided opportunities for rapid and automated monitoring of infrastructures. The recent shift towards edge computing in AI-based applications has enabled data processing closer to the source, reducing the need for costly data transmission and providing a real-time response with lower latency and cost. Additionally, this approach can extend operations to remote locations where reliable high-speed internet is inaccessible. The application of mobile robots in various fields, including structural health monitoring, has gained significant attention in recent years. However, research and development in unmanned vehicles for structural health monitoring has been limited and primarily focuses on manual outdoor operations with GPS availability. Indoor operations, such as construction site hazard inspections or indoor mapping, pose significant navigation and data collection challenges. An intelligent ground robot equipped with advanced hardware and software is needed to efficiently collect high-resolution data for structural health monitoring and navigation, process information in real-time, and save data for post-processing. The integration of robots has accelerated data collection and processing in structural health monitoring. However, current applications need more autonomous capabilities, leading to time-consuming, subjective, and expensive applications. To address the limitations of the earlier studies and fill the gaps found in the literature, this dissertation proposes an affordable solution for SHM using a cyber-physical system that incorporates both unmanned aerial vehicles and unmanned ground vehicles through a wireless robot operating system network. In addition, a novel transformer-based technique is proposed in this dissertation for high-resolution image segmentation to enable more accurate detection and quantification of structural elements and damages. The key feature and contribution of the proposed method are to achieve high-accuracy pixel-level segmentation in a faster way. An extension of image segmentation is developed to process point cloud data generated from 3D mapping and 3D scene reconstruction.