Abstract:
Product defect detection is a significant step in production lines of industries
traditionally relying on human manual inspection which is wasteful of labor, prone to errors,
and non-productive, especially with increasing production levels. Such shortcomings can lead
to defective products reaching consumers, resulting in dissatisfaction, threat to safety, and loss
of reputation for manufacturers and cost loss through reworks or recalls. To address these
problems, this capstone project is going to develop an automated system of fault detection
through computer vision and deep learning procedures for the enhancement of industrial
product quality assurance.
We employ the use of advanced Convolutional Neural Networks (CNNs), i.e.,
leveraging existing models like HRNet under the Anomalib library, having the capability to
successfully detect various faults like scratches, open faults, stains, etc. This system should
capture images of products along the line of production and process them in real-time with
instant feedback to operators to enable timely correction. The system also possesses
capabilities of data logging and reporting, summing up data regarding detected defects, their
time stamps, and types, which could be represented through interactive dashboards for trend
analysis and continuous process improvement.
The findings are that a CNN-based approach offers significant advantages over
traditional manual inspection by offering a robust, scalable, and effective automated defect
identification. By automating this critical quality control process, our system mitigates human
error, significantly reduces labor costs, and contributes to improved product quality and
improved manufacturing efficiency, ultimately towards improved customer satisfaction and
manufacturer confidence. This project recognize the growth potential of AI and computer
vision in revolutionizing manufacturing quality control processes.
