A machine learning method for buckling design of internally pressurized torispherical heads considering geometric imperfection

Abstract

A machine learning (ML) method is proposed for buckling design of internally pressurized torispherical heads, and geometric imperfection (GI) generated in the fabrication process is considered in the prediction of buckling pressure. Firstly, principal component analysis (PCA) is used to extract the primary features. Then, four fabrication methods are transformed from categorical variables into numerical features which could be considered as GI feature in ML models. Finally, random forest (RF) and support vector machines (SVM) are applied to predict buckling pressure of torispherical heads with different input features. Results demonstrate that the ML models have better predictive performance than traditional method. For RF and SVM, the models with primary and GI features predict the most accurate buckling pressure and SVM with a design factor of 1.25 is recommended in buckling design of torispherical heads.