Application Research on Prediction of Weld Ultrasonic Inspection Results Based on EasyEnsemble and XGBoost Algorithm

To reduce the missed inspection rate of unqualified welded seams of the hull, a model based on EasyEnsemble and XGBoost algorithm is proposed to predict the ultrasonic inspection results of welds. Based on historical data of weld ultrasonic inspection, parameters related to the welding quality were selected and these parameters were processed by feature engineering such as normalization and coding. Then effective features were extracted as the model input by principal component analysis (PCA). Considering the low recall of negative samples caused by extremely unbalanced sample data distribution, the EasyEnsemble algorithm was adopted to obtain a balanced training sample set and XGBoost algorithm was used as the base classification model of EasyEnsemble algorithm. The validity of the proposed model was proved by the experiment, the recall of negative samples was greatly improved and the missed inspection rate of unqualified welds was reduced.