Neural Network-Based Evaluation of SCF Distribution at the Intersection of Tubular X-Joints

Abstract

Stress Concentration Factor (SCF) is used as a reference to quantify the peak stress in welded tubular joints for fatigue assessment. At present, complex equations are available for calculating SCFs at selected locations of the brace-chord intersection curve of tubular joints for individual load cases. This paper presents an alternative approach to calculate SCF distribution along the intersection curve of tubular X-joints using neural networks. An SCF database based on the results of 300 finite element (FE) models was used to train and test the neural networks. Neural networks trained by these FE results were found to provide close predictions of the SCF distributions under each independent loading as well as combined loadings. Based on the parametric study it can be concluded that properly trained and well calibrated neural networks can be reliable alternatives to complicated SCF equations for predicting SCF at selected critical locations or along the brace-chord intersection curve of tubular joints.

Keywords:

• neural network
• fatigue
• stress
• stress concentration factor
• tubular joint
• finite element method