DETECTION OF WAVINESS IN STEEL COIL PRODUCTION

DETECTION OF WAVINESS
IN STEEL COIL PRODUCTION

Surface Defect Detection System
Shanghai, China · Shanghai ShenKu Information Technology Co., Ltd. · Jul 2023 – Sep 2023 · Computer Vision Engineer Intern

This project focused on developing an automated visual inspection system for detecting surface defects on industrial steel coils. Currently, it is in the prototype stage. By combining deep learning and computer vision, the system achieved high-speed, high-accuracy detection, making it suitable for factory deployment.

Contributions:

Modelling: Assisted in training a ResNet-18 CNN for four-level defect grading (0–3) using transfer learning and data augmentation, achieving 95% test accuracy.
Preprocessing: Processed 10K+ images with OpenCV (rotation, cropping, denoising, normalization) to improve dataset quality.
Pipeline: Helped design a prototype data workflow from acquisition and inference to retraining and visualization.
Automation: Created Python scripts for LabelImg/LabelMe format conversion and validation, improving labelling efficiency by 40%.

Result:
The prototype demonstrated an accuracy rate of 80% or higher in field testing and provided a strong foundation for future production deployment.

Part of the sources: