Development of a Failure Detection Model for Welding Robots Using Acceleration Signals

Abstract

We have developed a machine learning model to detect welding failure states, using only signals from acceleration sensors attached to welding robots for assembling automobile bodies at automotive factories. By applying acceleration data obtained from such robots over a period of roughly one year, the proposed model for multiple mixed classifications generated over 1000 feature quantities, allowing it to classify three welding states: “normal”, “welding tip disengaged”, and “welding anomaly”.