Detection of surface defects of aluminium extrudants using artificial intelligence

Poan, Chee Kent (2024) Detection of surface defects of aluminium extrudants using artificial intelligence. Final Year Project, UTAR.

Preview

PDF Download (4Mb) | Preview

Abstract

As the demand for aluminium profiles continues to rise, the occurrences of defects on the aluminium surface increase rapidly. While some factories still rely on manual defect detection, the small and unobvious defects often lead to high false detection rates due to the human eye's limitation. Although nowadays some manufacturing industries have implemented algorithms to automate the detection of defects, those algorithms face challenges on dealing with noises and lighting changes. This study aims to replace manual and inefficient automated defect detection with an approach that uses object detection. The objectives of this study include implement YOLOv8 model to identify and categorise the aluminium surface defect with the aid of data augmentation, transfer learning and addition of attention modules. The YOLOv8n model is trained to identify and localise the defect on the aluminium surface with the help of transfer learning. To solve the problem of limited datasets, data augmentation is used to expand the dataset to prevent overfitting. This study also compares the performance between YOLOv8n with and without attention modules. Attention modules included in this study are ECA and ResCBAM. However, the implementation of attention modules does not increase the performance of the model. The final model achieved a mAP@0.5 of 94.3% and 79.7% of mAP@0.5:0.95 compared to the original YOLOv8n model. This study shows the effectiveness and efficiency of YOLOv8n in detecting defects on aluminium surfaces. Besides, this study also proves the effectiveness of transfer learning and data augmentation in improving the overall performance of YOLOv8n in detecting various kinds of defect of aluminium surface.

Actions (login required)