Prediction of welding residual stresses using machine learning: Comparison between neural networks and neuro-fuzzy systems

Research output: Contribution to journalArticle

3 Citations (Scopus)
13 Downloads (Pure)

Abstract

Safe and reliable operation of power plants invariably relies on the structural integrity assessments of pressure vessels and piping systems. Welded joints are a potential source of failure, because of the combination of the variation in mechanical properties and the residual stresses associated with the thermomechanical cycles experienced by the material during welding. This paper presents comparative studies between methods based on artificial neural networks (ANN) and fuzzy neural networks (FNN) for predicting residual stresses induced by welding. The performance of neural network and neuro-fuzzy systems are compared based on statistical indicators, scatter plots and several case studies. Results show that the neuro-fuzzy systems optimised using a hybrid technique can perform slightly better than a neural network trained using Levenberg-Marquardt algorithm, primarily because of the inability of the ANN approach to provide conservative estimates of residual stress profiles. Specifically, the prediction accuracy of the neuro-fuzzy systems trained using the hybrid technique is better for the axial residual stress component, with root mean square error (RMSE), absolute fraction of variance (R2) and mean absolute percentage error (MAPE) error of 0.1264, 0.9102 and 22.9442 respectively using the test data. Furthermore, this study demonstrates the potential benefits of implementing neuro-fuzzy systems in predicting residual stresses for use in structural integrity assessment of power plant components.
Original languageEnglish
Pages (from-to)131-146
Number of pages15
JournalApplied Soft Computing
Volume70
Early online date22 May 2018
DOIs
Publication statusPublished - Sep 2018

Fingerprint

Fuzzy systems
Learning systems
Residual stresses
Welding
Neural networks
Structural integrity
Power plants
Fuzzy neural networks
Piping systems
Pressure vessels
Mean square error
Welds
Mechanical properties

Bibliographical note

NOTICE: this is the author’s version of a work that was accepted for publication in
Applied Soft Computing. Changes resulting from the publishing process, such as
peer review, editing, corrections, structural formatting, and other quality control
mechanisms may not be reflected in this document. Changes may have been made
to this work since it was submitted for publication. A definitive version was
subsequently published in Applied Soft Computing, VOL 70, (2018)
DOI: 10.1016/j.asoc.2018.05.017

© 2017, Elsevier. Licensed under the Creative Commons AttributionNonCommercial-NoDerivatives
4.0 International
http://creativecommons.org/licenses/by-nc-nd/4.0/

Keywords

  • Artificial neural networks
  • Adaptive neuro-fuzzy inference system (ANFIS
  • Welding
  • Residual stress

Cite this

@article{f81c48da19b04da09f9069088212d77b,
title = "Prediction of welding residual stresses using machine learning: Comparison between neural networks and neuro-fuzzy systems",
abstract = "Safe and reliable operation of power plants invariably relies on the structural integrity assessments of pressure vessels and piping systems. Welded joints are a potential source of failure, because of the combination of the variation in mechanical properties and the residual stresses associated with the thermomechanical cycles experienced by the material during welding. This paper presents comparative studies between methods based on artificial neural networks (ANN) and fuzzy neural networks (FNN) for predicting residual stresses induced by welding. The performance of neural network and neuro-fuzzy systems are compared based on statistical indicators, scatter plots and several case studies. Results show that the neuro-fuzzy systems optimised using a hybrid technique can perform slightly better than a neural network trained using Levenberg-Marquardt algorithm, primarily because of the inability of the ANN approach to provide conservative estimates of residual stress profiles. Specifically, the prediction accuracy of the neuro-fuzzy systems trained using the hybrid technique is better for the axial residual stress component, with root mean square error (RMSE), absolute fraction of variance (R2) and mean absolute percentage error (MAPE) error of 0.1264, 0.9102 and 22.9442 respectively using the test data. Furthermore, this study demonstrates the potential benefits of implementing neuro-fuzzy systems in predicting residual stresses for use in structural integrity assessment of power plant components.",
keywords = "Artificial neural networks, Adaptive neuro-fuzzy inference system (ANFIS, Welding, Residual stress",
author = "Jino Mathew and James Griffin and M. Alamaniotis and Stratis Kanarachos and Michael Fitzpatrick",
note = "NOTICE: this is the author’s version of a work that was accepted for publication in Applied Soft Computing. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in Applied Soft Computing, VOL 70, (2018) DOI: 10.1016/j.asoc.2018.05.017 {\circledC} 2017, Elsevier. Licensed under the Creative Commons AttributionNonCommercial-NoDerivatives 4.0 International http://creativecommons.org/licenses/by-nc-nd/4.0/",
year = "2018",
month = "9",
doi = "10.1016/j.asoc.2018.05.017",
language = "English",
volume = "70",
pages = "131--146",
journal = "Applied Soft Computing",
issn = "1568-4946",
publisher = "Elsevier",

}

TY - JOUR

T1 - Prediction of welding residual stresses using machine learning: Comparison between neural networks and neuro-fuzzy systems

AU - Mathew, Jino

AU - Griffin, James

AU - Alamaniotis, M.

AU - Kanarachos, Stratis

AU - Fitzpatrick, Michael

N1 - NOTICE: this is the author’s version of a work that was accepted for publication in Applied Soft Computing. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in Applied Soft Computing, VOL 70, (2018) DOI: 10.1016/j.asoc.2018.05.017 © 2017, Elsevier. Licensed under the Creative Commons AttributionNonCommercial-NoDerivatives 4.0 International http://creativecommons.org/licenses/by-nc-nd/4.0/

PY - 2018/9

Y1 - 2018/9

N2 - Safe and reliable operation of power plants invariably relies on the structural integrity assessments of pressure vessels and piping systems. Welded joints are a potential source of failure, because of the combination of the variation in mechanical properties and the residual stresses associated with the thermomechanical cycles experienced by the material during welding. This paper presents comparative studies between methods based on artificial neural networks (ANN) and fuzzy neural networks (FNN) for predicting residual stresses induced by welding. The performance of neural network and neuro-fuzzy systems are compared based on statistical indicators, scatter plots and several case studies. Results show that the neuro-fuzzy systems optimised using a hybrid technique can perform slightly better than a neural network trained using Levenberg-Marquardt algorithm, primarily because of the inability of the ANN approach to provide conservative estimates of residual stress profiles. Specifically, the prediction accuracy of the neuro-fuzzy systems trained using the hybrid technique is better for the axial residual stress component, with root mean square error (RMSE), absolute fraction of variance (R2) and mean absolute percentage error (MAPE) error of 0.1264, 0.9102 and 22.9442 respectively using the test data. Furthermore, this study demonstrates the potential benefits of implementing neuro-fuzzy systems in predicting residual stresses for use in structural integrity assessment of power plant components.

AB - Safe and reliable operation of power plants invariably relies on the structural integrity assessments of pressure vessels and piping systems. Welded joints are a potential source of failure, because of the combination of the variation in mechanical properties and the residual stresses associated with the thermomechanical cycles experienced by the material during welding. This paper presents comparative studies between methods based on artificial neural networks (ANN) and fuzzy neural networks (FNN) for predicting residual stresses induced by welding. The performance of neural network and neuro-fuzzy systems are compared based on statistical indicators, scatter plots and several case studies. Results show that the neuro-fuzzy systems optimised using a hybrid technique can perform slightly better than a neural network trained using Levenberg-Marquardt algorithm, primarily because of the inability of the ANN approach to provide conservative estimates of residual stress profiles. Specifically, the prediction accuracy of the neuro-fuzzy systems trained using the hybrid technique is better for the axial residual stress component, with root mean square error (RMSE), absolute fraction of variance (R2) and mean absolute percentage error (MAPE) error of 0.1264, 0.9102 and 22.9442 respectively using the test data. Furthermore, this study demonstrates the potential benefits of implementing neuro-fuzzy systems in predicting residual stresses for use in structural integrity assessment of power plant components.

KW - Artificial neural networks

KW - Adaptive neuro-fuzzy inference system (ANFIS

KW - Welding

KW - Residual stress

U2 - 10.1016/j.asoc.2018.05.017

DO - 10.1016/j.asoc.2018.05.017

M3 - Article

VL - 70

SP - 131

EP - 146

JO - Applied Soft Computing

JF - Applied Soft Computing

SN - 1568-4946

ER -