Abstract: In order to avoid defects such as burn-through,incomplete fusion,wire blocking and slag inclusion appearing in welding process of bogie,the information acquisition system of bogie robot MAG welding process was constructed,which could collect real-time weld pool image,arc sound and voltage information. Key features closely related to quality were extracted and knowledge modeling and fusion analysis were carried out based on XGBoost model. Finally,the ability of information perception and quality analysis similar to welding masters were realized,and the comprehensive accuracy could reach more than 90%,thus improving the quality reliability and intelligent manufacturing level of bogie welding operation.