<p>Gas tungsten arc welding (GTAW) plays a crucial role in high-precision manufacturing, where optimizing process parameters is essential to minimize defects and ensure structural reliability. However, accurately predicting and optimizing welding parameters requires advanced modeling techniques, as conventional methods face significant constraints. Traditionally, the artificial neural network (ANN) training approaches often encounter challenges such as local minima entrapment and overfitting. To address these issues, this paper integrates metaheuristic optimization algorithms with ANN models, leveraging their global search capabilities and improved convergence properties. This study explores the integration of three metaheuristic optimization algorithms including weighted mean of vectors (INFO), gradient-based optimizer (GBO), and artificial rabbit optimization (ARO) to improve ANN performance. Using experimental data from 32 welding parameter combinations, our optimized models demonstrated significant improvements compared to conventional ANN models. For instance, ANN-INFO model achieved coefficient of determination (<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="170_2025_16128_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="21" /> </InlineMediaObject> <EquationSource Format="TEX">$R^2$</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mi>R</mi> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation>) test values of 0.9383, 0.9890, 0.9129, and 0.9583 for the weld characteristics of width, reinforcement, penetration, and dilution, respectively. Bald eagle search (BES) algorithm integration further optimized parameter selection, with the ANN-INFO-BES model achieving an optimal objective value of <InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="170_2025_16128_Article_IEq2.gif" Format="GIF" Height="13" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">$-$</EquationSource> <EquationSource Format="MATHML"><math> <mo>-</mo> </math></EquationSource> </InlineEquation>1.004, significantly higher performance than ANN-genetic algorithm (GA) of <InlineEquation ID="IEq3"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="170_2025_16128_Article_IEq2.gif" Format="GIF" Height="13" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">$-$</EquationSource> <EquationSource Format="MATHML"><math> <mo>-</mo> </math></EquationSource> </InlineEquation>0.447 in the previous study. This research work highlights the ANN models’ potential, combined with metaheuristic algorithms, for enhancing the predictive accuracy and optimization of GTAW processes. Our advanced hybrid models ensure optimal selection of welding parameters, thereby increasing industrial efficiency by reducing material waste and minimizing production time. With higher process reliability and repeatability, these models contribute to more sustainable and cost-effective manufacturing environments.</p>

An improved artificial neural network using weighted mean of vectors algorithm for precise GTAW weld quality prediction and parameter optimization

期刊论文
发表日期：

2025年7月22日

查看全文

Brahim Boucetta,
Faiza Boumediene,
Mohamed Abdessamed Ait Chikh,
Adel Afia

摘要

Gas tungsten arc welding (GTAW) plays a crucial role in high-precision manufacturing, where optimizing process parameters is essential to minimize defects and ensure structural reliability. However, accurately predicting and optimizing welding parameters requires advanced modeling techniques, as conventional methods face significant constraints. Traditionally, the artificial neural network (ANN) training approaches often encounter challenges such as local minima entrapment and overfitting. To address these issues, this paper integrates metaheuristic optimization algorithms with ANN models, leveraging their global search capabilities and improved convergence properties. This study explores the integration of three metaheuristic optimization algorithms including weighted mean of vectors (INFO), gradient-based optimizer (GBO), and artificial rabbit optimization (ARO) to improve ANN performance. Using experimental data from 32 welding parameter combinations, our optimized models demonstrated significant improvements compared to conventional ANN models. For instance, ANN-INFO model achieved coefficient of determination ( $R^2$ $R^{2}$ ) test values of 0.9383, 0.9890, 0.9129, and 0.9583 for the weld characteristics of width, reinforcement, penetration, and dilution, respectively. Bald eagle search (BES) algorithm integration further optimized parameter selection, with the ANN-INFO-BES model achieving an optimal objective value of $-$ $-$ 1.004, significantly higher performance than ANN-genetic algorithm (GA) of $-$ $-$ 0.447 in the previous study. This research work highlights the ANN models’ potential, combined with metaheuristic algorithms, for enhancing the predictive accuracy and optimization of GTAW processes. Our advanced hybrid models ensure optimal selection of welding parameters, thereby increasing industrial efficiency by reducing material waste and minimizing production time. With higher process reliability and repeatability, these models contribute to more sustainable and cost-effective manufacturing environments.