REAL-TIME DEFECT DETECTION IN ROBOTIC WELDING OF ALUMINIUM 5083 INTEGRATING PREDICTIVE ANALYTICS AND MACHINE LEARNING

Abstract Predictive analytics is applied to robotic welding cells to enhance the quality and reliability of aluminium 5083 welds, an alloy known for its high strength, corrosion resistance, and weldability. This research focuses on identifying and characterising welding defects such as porosity and lack of fusion by analysing the influence of key process parameters, including Wire Feed Speed (WFS), Torch Speed (TS), and Gas Flow Rate (GFR), which are varied during experimentation. Advanced robotic MIG welding systems are employed with real-time current data acquisition through an OPC-UA server. Statistical analysis of signal features, root mean square (RMS) value, standard deviation, skewness, kurtosis, and peak-to-peak value is performed to detect defects. Machine learning models, particularly Artificial Neural Networks (ANNs), are utilised to predict weld bead quality, with cross-validation ensuring model robustness and accuracy. Results indicate that the precise monitoring and analysis of process parameters can effectively identify defect patterns and improve understanding of weld quality. The integration of real-time monitoring and machine learning demonstrates significant potential for refining welding processes and advancing industrial production engineering.