Date of Award
Doctor of Philosophy
Ahsan R. Choudhuri
Conventional, manual inspection methods are the most commonly used inspection approaches to this day; that cause downtime and can be erroneous due to their repetitive nature, heavy workload, and human error. The overarching goal of this work is to advance structural inspection with intelligent and autonomous techniques across infrastructures. In particular, this project will develop path-planning schemes for close navigation around the structures and intelligent algorithms for crack and corrosion detection.
The introduced novel navigation method uses advanced manufacturing techniques to generate aerial inspection trajectories in GPS-denied areas. The proposed method is validated using the `Gazebo' robotics simulator; the results confirm the usability for close-quarter inspection of any structural components with complex geometry.
The intelligent inspection algorithms are developed by leveraging Artificial intelligence's (AI) Deep Neural Networks (DNNs). Custom data sets are acquired and appropriately prepared for the specific model and anomaly classes, `crack' and `corrosion.' The models are further optimized to a lighter, lower latency version for real-time deployment at the edge. Developed custom models are tested for validation in industrial compounds, and they competently identify and localize the defects at the scene.
Lastly, an integrated multi-spectral inspection capability with a user interface (UI) is developed to advance and supplement the inspection method. It generates overlayed multi-spectral scopes fusing color and infrared sensor feeds, reads temperatures \& displays thermal profiles to the UI. Experimental studies are conducted to demonstrate the usability and advantage of the system in infrastructural defect detection. The proposed approaches are validated in laboratory and industrial setups.
Received from ProQuest
Mst Mousumi Rizia
Rizia, Mst Mousumi, "Real-Time Intelligent And Multi-Spectral Inspection Of Structural Components" (2020). Open Access Theses & Dissertations. 3721.