AI-powered predictive framework for crack detection in steel-copper laser welding.

Area of Science:

• Materials Science and Engineering
• Manufacturing Technology
• Computational Intelligence

Background:

• Laser welding of dissimilar materials like steel and copper often results in cracks, compromising joint integrity.
• Conventional inspection methods for weld defects are inefficient and prone to errors.
• There is a significant need for intelligent systems to predict defects in real-time during the welding process.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting defects in laser-welded steel-copper joints.
• To investigate the effectiveness of the Orange data mining tool in handling data imbalance for welding defect prediction.
• To optimize a neural network model for enhanced reliability and predictive performance in defect classification.

Main Methods:

• Utilized a dataset of 360 weld cross-sections generated via a definitive screening design (DSD).
• Evaluated multiple machine learning classifiers (Decision Trees, Random Forests, Gradient Boosting, SVM, Neural Networks) using Orange data mining software.
• Applied imbalance handling strategies and optimized a Multilayer Perceptron (MLP) neural network.

Main Results:

• The Multilayer Perceptron (MLP) neural network demonstrated superior performance.
• Achieved 94.9% accuracy, 86.3% sensitivity, and 96.8% specificity.
• Obtained an Area Under the Curve (AUC) of 0.961, indicating robust classification capabilities.

Conclusions:

• Neural networks are effective and scalable tools for classifying defects in steel-copper laser welds.
• The study highlights the successful application of the Orange tool for data imbalance in welding defect prediction.
• The findings support the implementation of intelligent process monitoring and predictive quality assurance in Industry 4.0 manufacturing.