Prediction of Stress Concentration Factor in Butt Welding Joints using Artificial Neural Networks

In welded constructions, there should be no defects in the welding seams, or defects should have in an acceptable range for obtaining more reliable welding operations. An undercut is one of the most important welding defects occurring on the workpieces produced by butt welding. Determining the correct value of the stress concentration factor (SCF) allows deciding whether it accepts welding defects, in which case. Many characteristics and ranges influence SCF, making it challenging to calculate a more precise SCF. In this study, six different artificial neural networks (ANN) models are developed for predicting SCF. These models differ in terms of the training dataset used (۷۰%-۹۰%) and the number of neurons (۵-۱۰-۲۰) in the hidden layer. Developed ANN models consist of three input variables the ratio of Undercut depth (h) and Undercut deep Radius (r), Reinforcement angle (Q۱), deep angle of welding seam (Q۲), and an output variable as SCF. The prediction performance of ۶ developed ANN models in different specifications is compared. The model with a ۹۰% training set and five neurons in the hidden layer performed the best with an accuracy of ۰.۹۸۳۴. According to the ANN model with these features, MAE, MAPE, and RMSE values are calculated as ۰.۰۰۹۴, ۲.۵۰%, and ۰.۰۱۲۹, respectively.

کلیدواژه ها:

Welding defects ، Stress Concentration Factor ، Artificial Neural Networks