A GBDT Based Quality Prediction Method for the Resistance Spot Welding

The quality of welding joints is quite important for the safety and service life of automobile, and how to realize online monitoring of welding joint quality without destructive detection methods is a hot research topic in these days. Based on the manually labeled welding data obtained from an automobile factory, a gradient boosting decision tree based model was proposed to predict the quality of welding spots. The cross validation performance indicated that the GBDT based model would have both higher prediction accuracy and F1 score comparing with other 7 methods. These results implied that our proposed GBDT model had great potential in the application of quality prediction for the resistance spot welding.