Development of Situation Awareness Model in Robotic Spot-welding (RSW) System based on Sensor Data Visualization

Abstract

The advancement of prognostics and health management (PHM) using the industrial internet of things (IIoT) and artificial intelligence (AI) has enabled stable mass production with assured quality. However, the manufacturing industry often faces variable conditions like design and material changes. Before applying PHM systems, it is crucial to design a system that recognizes these changes. This study developed a robust situation awareness model for a robotic spot-welding (RSW) system, resilient to reduced training data, using sensor data visualization and convolutional neural networks (CNN). Four variable situations were established based on thickness and material changes: GI steel 0.6 t, GI steel 1.0 t, mild steel 0.8 t, and GA steel 0.8 t, with GI steel 0.8 t as the reference. Data were collected using process parameters (welding current and electrode force) for each situation. A fuzzy-based energy pattern image (FEPI) visualization technique was applied to visualize energy differences in the spot-welding process from four sensors (current, voltage, electrode force, displacement). Using these techniques, thickness and material variation awareness models were constructed. The classification accuracy of CNN models trained on time-series data was evaluated to verify the effectiveness under reduced training data conditions.