Tools
Tools
Patel, Deep Kantilal (2024) Construction Defect Detection using AI with Augmented Reality. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (929kB) | Preview |
Preview |
PDF (Configuration Manual)
Download (722kB) | Preview |
Abstract
This report explores how a systematic detection of building using deep learning has been developed, specifically targeting defects such as Blister, Cracks, Peels, Seepage, Mold. The data set initially exhibited class imbalance, with oversampling and under-sampling techniques used to ensure equal distribution of each defect type to improve the model's ability to generalize, data enhancement was implemented by techniques such as on rotation, flipping, zooming and light adjustment.
A convolutional neural network (CNN) was used as the primary model for fault detection. This model is integrated with OpenCV to facilitate pre-imaging and fault ranking, ensuring accurate detection and real-time classification of defects. The CNN model was trained on a compressed dataset and it showed high accuracy and robustness in classifying defects.
The results of this study highlight the effectiveness of deep learning models using traditional imaging techniques for automatic fault detection in building construction. The successful implementation of this system highlights its applicability to real-world production quality, providing scalable and effective solutions to improve product reliability and safety.
Actions (login required)
 |
View Item |