Classification of weld defect based on information fusion technology for radiographic testing system

Area of Science:

• Non-destructive testing
• Radiographic testing
• Materials science

Background:

• Accurate weld defect classification is crucial for radiographic testing systems.
• Existing methods face challenges with efficiency, accuracy, and limited training samples.

Purpose of the Study:

• To propose a novel weld defect classification method using information fusion and Dempster-Shafer evidence theory.
• To enhance the accuracy and efficiency of weld defect classification in radiographic images.
• To address the challenge of limited training samples in defect classification.

Main Methods:

• Defined 11 weld defect features based on sub-pixel level edges of radiographic images, introducing four novel features.
• Applied information fusion technology to combine diverse features for classification.
• Utilized a mass function and quartile-method for standard class calculation, addressing limited sample sizes.

Main Results:

• Achieved increased correct classification rates, particularly with limited training samples.
• Demonstrated the method's effectiveness in a steam turbine weld defect classification case study.
• Successfully addressed uncertainties inherent in weld defect classification.

Conclusions:

• The proposed information fusion method significantly improves weld defect classification accuracy.
• The technique is effective even when dealing with a scarcity of training data.
• This approach offers a robust solution for radiographic testing systems facing classification uncertainties.