Automated Categorization of Multiclass Welding Defects Using the X-ray Image Augmentation and Convolutional Neural...

Area of Science:

• Industrial Nondestructive Testing
• Machine Learning for Quality Control
• Image Processing for Defect Detection

Background:

• Manual X-ray inspection for weld defects is costly, time-consuming, and prone to human error.
• The need for automated, reliable, and efficient methods for identifying diverse welding flaws is critical in manufacturing.

Purpose of the Study:

• To develop an automated system for identifying and categorizing multiple types of weld defects from X-ray images.
• To leverage data augmentation and convolutional neural networks (CNNs) for enhanced defect detection accuracy.

Main Methods:

• Implemented a hybrid approach combining advanced data augmentation techniques (rotation, shearing, zooming, brightness, flips) with a CNN model.
• Trained and evaluated the model on an industrial dataset of 4479 X-ray images across six defect categories and normal samples.

Main Results:

• The automated system achieved an average accuracy of 92% in detecting and classifying various weld defects.
• The data augmentation strategy improved the generalization capability of the CNN model for multi-class defect identification.

Conclusions:

• The proposed automated approach demonstrates significant potential for reliable and efficient weld defect detection in industrial settings.
• This CNN-based method offers a viable alternative to traditional manual inspection, improving accuracy and reducing operational costs.