[4F4-GS-10o] AI応用：マテリアル情報

[4F4-GS-10o-05] Establishment of new factor analysis method for robtic weld defect reduction

Knowledge and data fusion using Bayesian networks

〇Naoko Omachi¹, Takeshi Ikeda¹, Shota Masuda¹, Yoichi Motomura² (1. UACJ Corporation, 2. AIRC, National Institute of Advanced Industrial Science and Technology)

Keywords:Bayesian networks, Robot welding, Factor analysis

Robotic welding is widely used for mass production in such as automobile manufacturing industry on account of the operational stability and repeat accuracy, even though weld defects may occur due to poor followability to environment changes. Especially, it is essential to control weld heat input to prevent the burn through of the base material in aluminum welding.
In this study, we have studied the factor analysis for the burn through defect in robotic welding using a prediction model.The model is based on Bayesian networks which can easily incorporate data and knowledge. Among of a few models built with different dependencies to the knowledge, the model consists to the result of the verification experiment is selected. We expect further quality improvement by leveraging such models in preventive maintenance or adaptive control in the future.