A comparison study of BPN and SVM prediction models for inherent deformations of T-welded joints. Issue 4 (16th February 2021)

• Record Type: Journal Article
• Authors: Tian, Liang; Luo, Yu
• Is Part Of: Mechanics of advanced materials and structures. Volume 28:Issue 4(2021)
• Page Start: 365
• Page End: 383
• Publication Date: 2021-02-16
• Subjects: Inherent deformations -- T-joint fillet welds -- BPN model -- SVM model -- numerical simulation; Composite materials -- Mechanical properties -- Periodicals; Composite construction -- Periodicals; 620.118
• DOI: 10.1080/15376494.2019.1567881
• Languages: English
• ISSNs: 1537-6494

Abstract

Abstract: The inherent deformation method has a significant advantage in predicting the welding deformations for large and complex welded structures, but the premise of applying this calculation method is to determine the values of inherent deformations in a given welded joint in advance. Therefore, it is a problem worthy of study to provide an accurate and rapid evaluation for the inherent deformations in a specific type of welded joint. In this study, an uncoupled thermo-elastic-plastic finite element (TEP FE) model was first developed to simulate the T-joint fillet welding process, and the welding experiments were performed to verify the accuracy of the developed TEP FE model. Then the numerical experiments were carried out to provide the informative data for the establishment of the BPN and SVM models. The appropriate network of the BPN model and the optimal parameters of the SVM model were obtained by the trial-and-error method and genetic algorithm (GA), respectively. In these two models, the input variables are the welding speed, current, voltage, and plate thickness, and the transverse and longitudinal inherent deformations are considered to be the output variables. The research results demonstrate that both the BPN model and SVM model can give a better prediction for the inherent deformations with acceptable accuracy. However, compared with the BPN model, the SVM model shows a better generalization ability for the data outside the training set and is more suitable for predicting the inherent deformations within a certain range of welding parameters.