IMPROVING PREDICTION QUALITY WITH XGBOOST MODEL FOR BENDING CAPACITY OF X65 DEFECTED PIPE

Abstract

There is a desire of a quality model predicting bending capacity of the defected pipe for pipeline integrity assessment. While the analytical model faces with the difficulty in modeling the local defect and corresponding local stress, the Finite Element method is a valuable alternative. A common approach for predicting interested variable is to scrape the result data and develop a data-driven model such as the classical linear regression, CART or XGBoost. Along with generating numerical database with FEM, this study illustrates the advance of XGBoost model compared with its counterparts in predicting the moment capacity of the defected X65 pipe.