Multisensor fusion for on line monitoring of the quality of spot welding in automotive industry

Abstract Spot welding is used extensively in the fabrication of body in white structures and vehicle components. However, despite intensive work over the last 40 years, there has been little advancement in sensors systems for on line monitoring of the quality of spot welds. Most existing systems are based upon current, voltage and welding force. These parameters do not provide information on the condition of the electrode, where excessive wear can result in weakened weld nuggets. This paper presents an overview of a system that combines the traditional sensors with ultrasonic monitoring into a sensor cluster that monitors online the integrity of the weld parameters. The sensor inputs, combined with derived quantities such as power and peak dynamic resistance are fed into a neural network based system that predicts the quality and nugget size of each spot weld in real time. The prediction network, when operated at the borderline between acceptable and worn electrode tip conditions, can predict the nugget size to within ±0.15 mm.