Automated visual defect detection using deep learning
A manufacturing defect is a flaw that causes a product to deviate from its intended design, thereby losing its quality and no longer having its due value. There are two kinds of methods employed by the manufacturing industries to ensure that the manufactured products are well-conditioned and free of...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Year Project / Dissertation / Thesis |
| Published: |
2022
|
| Subjects: | |
| Online Access: | http://eprints.utar.edu.my/4707/1/fyp_CS_2022_LX.pdf http://eprints.utar.edu.my/4707/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | A manufacturing defect is a flaw that causes a product to deviate from its intended design, thereby losing its quality and no longer having its due value. There are two kinds of methods employed by the manufacturing industries to ensure that the manufactured products are well-conditioned and free of any defects, namely human quality inspection and artificial intelligence visual inspection. However, the former faces many limitations and problems, resulting in low quality control efficiency, while the latter is relatively reliable and effective. Artificial intelligence visual inspection is a technique which utilises computer vision and deep learning technology to mechanically “see” a product and determine whether it has defects, without any human involvement. The main goal of this project is to study and develop various automated defect detection models by utilizing state-of-the-art deep learning segmentation algorithms, including U-Net, Double U-Net, SETR, TransU-Net, TransDAU-Net, CAM and SEAM to perform semantic segmentation in fully supervised and weakly supervised learning manners. Model analysis and evaluation are performed to compare the performance of all the algorithms in a variety of aspects. In this project, a magnetic tile defect dataset and a production item surface defect dataset are employed to train and evaluate deep learning segmentation models. By applying the models, the detected defects will be segmented and classified, and the output results will be displayed to the user through a coloured segmentation mask for each defect. Manufactures will be able to automate the industrial inspection process by implementing the deep learning models proposed in this project. Quality control procedures of the industries will be sublimated to another level by the automation of product defect detection. |
|---|