Welding Defect Classification: A Low-Cost Approach Using XGBoost and Simulated Defect Data

Gas Metal Arc Welding (GMAW) is widely used in industries such as automotive, construction, and aerospace. Ensuring the integrity of welded joints requires real-time monitoring to detect defects like cracking, porosity, and burn-through. This study tackles the challenges of limited real-world data and resource constraints in deploying weld monitoring algorithms on edge devices. We introduce a method to generate realistic simulated weld signals with defects from good weld data samples and employ XGBoost, a lightweight machine learning classifier, for defect classification. Our approach optimises the XGBoost model through feature selection and model complexity adjustments, achieving high classification accuracy with less frequent data collection (lower sampling rates). Experimental results demonstrate the method's robustness and efficiency, highlighting its potential for large-scale industrial applications where the deployment of numerous IoT monitoring devices necessitates lightweight machine learning models for viable real-time processing and system scalability.