Optimization of screw production using deep convolutional neural network (DCNN)
Date
2024-03
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Federal University of Technology, Owerri
Abstract
This research proposed a deep convolutional neural network (DCNN) based technique for the detection of micro defects on metal screw surfaces. Defects considered include surface damage, surface dirt, and stripped screws. Images of metal screws with different types of defects were collected using industrial cameras, which were then employed to train the designed deep CNN. To enable efficient detection, I first located screw surfaces in the pictures captured by the cameras, so that the images of screw surfaces could be extracted, which were then inputted into the CNN-based defect detector. Experiment results showed that the proposed technique could achieve a detection accuracy of 97%; the average detection time per picture is 1.2 seconds. Comparisons with traditional machine vision techniques, e.g., template matching-based techniques, demonstrate the superiority of the proposed deep CNN-based one. Furthermore, it could be seen that the accuracy of the proposed DCNN was much higher than the traditional LeNet-5 at the beginning of the network training and the accuracy of the training was to 100% with 550 iterations and about 100% accuracy was achieved with 800 iterations.
Description
Master Degree in Industrial Production Engineering
Keywords
Screw image, deep convolutional neural network, micro-defect detection, Internet of Things, Department of Mechanical Engineering
Citation
Ndukwe, C. O. (2024). Optimization of screw production using deep convolutional neural network (DCNN) (Unpublished Master's Thesis). Federal University of Technology, Owerri, Nigeria