2026-09-20T00:51:05Z rgu-repository.worktribe.com

oai:rgu-repository.worktribe.com:2801306 2026-09-04T14:47:27Z 084104101115105115 openaccess

Thesis Towards automated remote inspection of anomalies in offshore components. This dissertation marks a significant advancement in offshore structural inspections, focusing on the development, integration and evaluation of advanced deep-learning models. The research encompasses: a thorough literature review, identifying innovation opportunities in deep learning for industrial inspections; the development of a general classification model using cutting-edge architectures for precise classification of circumferential welds; the design and training of an anomaly detection model to enhance fault identification; the implementation of a human-in-the-loop system for improved model accuracy and reliability; and a comprehensive evaluation of these models' real-world applicability. The study not only showcases cutting-edge deep learning techniques for defect detection, but also highlights critical research gaps, providing a guide for future investigation. The novel incorporation of human expertise with machine learning via a human-in-the-loop approach is a significant innovation, bolstering decision-making and potentially lowering error rates. This research presents a comprehensive model that could serve as a benchmark in the field, valuable to both academics and industry professionals. It concludes by reflecting on the framework's successes and limitations, discussing its implications for offshore inspection practices, and suggesting future research directions and potential broader industry impacts. Toral Quijas, Luis Alberto 2024-07-31 E. Elyan and C. Moreno-Garcia oai:rgu-repository.worktribe.com:2801306 https://rgu-repository.worktribe.com/2801306/1/TORAL%20QUIJAS%202024%20Towards%20automated%20remote https://doi.org/10.48526/rgu-wt-2801306 Innovate UK Deep learning Machine learning Remote sensing Fault detection Offshore engineering https://rgu-repository.worktribe.com/output/2801306 2024 en openAccess Public