FSW Weld Defect Prediction Feature

93% accuracy for defects

A lateral force average-based feature for tunnel defect prediction in friction stir welding, identified via machine learning and validated by physical mechanisms.

Guan, Wei · Yang, Chengle · Yuyuan, Cai · Dong, Tan · Shiqi, Zhang · Chen, Gaoqiang · Cui, Lei

Journal of Materials Processing Technology 2026

Specifications

Tunnel defect identification accuracy

{'zh': '93.0', 'en': '93.0'} %

Number of features analyzed

{'zh': '24', 'en': '24'}

Number of specimens

{'zh': '1278', 'en': '1278'}

Advantages

Predicts with a single force component

Information-gain analysis over 24 time- and frequency-domain features identifies the mean lateral force Fyavg as the most sensitive to tunnel defects, enabling high accuracy without complex feature engineering.

Under 7% error

Decision-tree models using four key features achieve 93.0% accuracy in identifying tunnel defects, corresponding to a misclassification rate below 7% suitable for online monitoring.

Applications

• Friction stir welding：Monitor lateral force during FSW to detect tunnel defects in real time and adjust parameters to avoid scrap.
• Weld quality monitoring：Use Fyavg as an online monitoring index to replace post-weld ultrasonic or X-ray inspection.
• Industrial robot welding：Integrate the feature into force sensor systems on welding robots for closed-loop quality control.