Accurate Prediction of Yield Strength and Welding Defects in FSW Joints of 2XXX and 6XXX Al Alloys Using Artificial Neural Network Based Correlation Analysis

  • A. K. Deepati Department of Mechanical Engineering Technology, CAIT, Jazan University, Jizan, Kingdom of Saudi Arabia
  • P. Anusha Department of Mechanical Engineering, Prasad V. Potluri Siddhartha Institute of Technology, Vijayawada, India
  • M. Naga Swapna Sri Department of Mechanical Engineering, Prasad V. Potluri Siddhartha Institute of Technology, Vijayawada, India
  • S. Guharaja Department of Mechanical Engineering, R.V.S. College of Engineering, Dindigul, India
  • S. Vijayakumar Mechanical Engineering Department, Saveetha School of Engineering, SIMATS, Tamil Nadu, India
DOI: https://doi.org/10.35219/awet.2025.09
Keywords: Friction Stir Welding, aluminium alloys, MATLAB, Artificial Neural Network, correlation coefficients and minimum validation error

Abstract

The accurate prediction of mechanical properties and defect formation in friction stir welding (FSW) of aluminium alloys is crucial for ensuring structural integrity and reliability in aerospace and automotive applications. In this study, an artificial neural network (ANN) was developed to predict yield strength (YS) and welding defects (WD) in FSW joints of AA2024, AA2219, and AA6061 alloys. A dataset of one hundred experimental cases obtained from peer-reviewed literature was employed, where process parameters rotational speed (RS), plate thickness (PT), shoulder radius (SR), axial pressure (AP), pin root radius (PR), pin tip radius (PS), and tool angle (TA) were served as inputs, and YS and WD were target outputs. The ANN was implemented in MATLAB R2016a and trained using backpropagation. Results showed strong predictive accuracy: for YS, correlation coefficients (R) of 0.96568, 0.99874, and 0.96278 were achieved for training, validation, and testing sets, with an overall R of 0.96764. The minimum validation error (MSE = 2.1901) occurred at epoch 11. For WD prediction, the overall R was 0.83229, with lowest validation error (MSE = 0.09041) at epoch 17. These findings highlight the potential of ANN-based models for real-time prediction and optimization of FSW quality. Future work will focus on expanding datasets, integrating hybrid AI techniques, and developing adaptive models for industrial-scale welding applications.

Downloads

Download data is not yet available.
Published
2025-12-15
How to Cite
Issue
Vol 36 (2025)
Section
Articles

Papers accepted for publication become the copyrighted property of the Annals of "Dunarea de Jos" University of Galati, Fascicle XII, Welding Equipment and Technology. No part of the publication may be reproduced or transmitted in any form, or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, except in limited quantities for the non-commercial purposes of scientific or educational advancement, without permission in writing from the Editorial Board.