Date of Award

2022-12-01

Degree Name

Doctor of Philosophy

Department

Mechanical Engineering

Advisor(s)

Angel Flores-Abad

Second Advisor

Ahsan R. Choudhuri

Abstract

Robotic technologies for inspection purposes of large-scale structures have grown in interest.Such technologies are encouraged to reduce the risk in which human operators are involved and to reduce costs due to downtime of the equipment. In the Energy sector, high interest is placed on power plant components where their correct operation is paramount. This work is inspired by the synthetic vision systems for aerial vehicles that use three-dimensional space (3D) to provide pilots with clear and intuitive means of understanding their flying environment. This work can be separated into three main sections: Trajectory Generation from Computer-Aided Design (CAD) Models, Crack Detection using Convolutional Neural Networks (CNNs), and 3D Reconstruction. This work proposes a solution to complete flight missions in GPS-denied environments and a method of obtaining a reconstruction point cloud with segmented cracks. Human inspectors are the most common way to inspect high-interest structures in the energy sector. That means that such inspectors must be trained to inspect specific structures but also must put themselves at risk to correctly inspect such structures. Some of the common hazards include very high altitudes and some environments can be detrimental to the health of the inspectors due to particulate byproducts from the structures. The inspectors then keep a manual log of their findings and relay that information to maintenance crews to fix or replace the damaged components. This type of inspection practice is prone to miss defects, incorrectly logging, and incorrectly locating defects found by the inspectors. Another approach in the industry is to use robotic systems to inspect like the use of UAVs or drones. Most of the current UAV technologies depend on a stable GPS signal to operate and complete their flight mission tasks. This is a big disadvantage in the inspection industry because some inspections must take place inside big structures that cause heavy signal interference. Also, most of the technologies available require a human pilot in command to be proficient at operating the UAV to prevent collisions and to cover as much surface as possible. This leads to human error and can prove difficult to even the most experienced pilots to navigate the complex structures. The objective of this work is to provide a platform to perform inspection tasks on high-interest structures within the energy sector. We propose a CAD-based aerial trajectory generation and 3D mapping platform for close-quarter inspections. With the use of CAD models of structures that are readily available, we can generate an offline trajectory that employs a wall offset and is capable of reaching virtually all exposed surfaces of the structures of interest with a minimum surface offset distance. The system also employs the use of Artificial Intelligence to detect, segment, and localize desired defects within the inspections. This eliminates human error in classifying and documenting the defects while maintaining a record of the defects. This data could then be used to map the environment with the discovered defects to better assess the level of damage. Finally, our system employs photogrammetry and point cloud reconstruction algorithms to accurately reconstruct the inspected environments. This could also be used in instances where an initial CAD model of the inspection structure is not available. With this work, we hope to streamline the inspection procedures that employ robotic technologies to remove human inspectors from hazardous environments. By utilizing an autonomous UAV platform that does not employ GPS we hope to complete inspections in even the most demanding environments. The system would allow the inspectors to accurately view all the detected defects as soon as each flight mission is completed, thus allowing a more efficient maintenance plan for such plants.

Language

en

Provenance

Received from ProQuest

File Size

131 p.

File Format

application/pdf

Rights Holder

Angel Guillermo Ortega Castillo

Download

Share

COinS