Scholarlink

<p>Wire-arc additive manufacturing (WAAM) is increasingly adopted for fabricating large-scale metallic components. However, reliable prediction and multi-objective optimization of weld-bead geometry remain critical challenges. This study systematically investigates the influence of voltage, wire feed speed (WFS), and torch travel speed (TS) on bead width (BW), bead height (BH), and microhardness (MH) using a structured Taguchi L18 experimental design. Statistical regression modeling and analysis of variance (ANOVA) were employed to identify significant process parameters and establish physically interpretable predictive relationships. To further assess potential nonlinear behavior, the XGBoost machine-learning algorithm was implemented as a surrogate model. Given the limited dataset size (<i>n</i> = 18), leave-one-out cross-validation (LOOCV) was adopted to obtain unbiased predictive performance estimates. Cross-validated results indicate that XGBoost provides performance comparable to linear regression within the investigated parameter window, achieving R²_LOOCV values of 0.879 (BW), 0.770 (BH), and 0.870 (MH). The comparatively lower prediction accuracy for BH is attributed to inherent melt pool instability and droplet transfer variability. Multi-objective grey relational analysis identified a locally optimal parameter combination of 15&#xa0;V, 5.61&#xa0;m/min WFS, and 10.9&#xa0;mm/s TS within the defined design space, enabling simultaneous minimization of BW and maximization of BH and MH. The developed framework provides a physically grounded and cross-validated predictive optimization methodology for WAAM parameter selection within constrained single-bead deposition conditions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Optimization and Predictive Modeling of Process Parameters in Wire-arc Additive Manufacturing Using Statistical and Machine Learning Approaches

  • 期刊论文

  • 发表日期：

    2026年5月11日

查看全文
  • Laukik P. Raut,
  • Sourabh Kumar Soni,
  • Prachi K. Tawele,
  • Ravindra V. Taiwade,
  • Vednath P. Kalbande,
  • Man Mohan,
  • Yogesh N. Nandanwar,
  • Amit Motwani,
  • Ashish Fande,
  • Byungmin Ahn

摘要

Wire-arc additive manufacturing (WAAM) is increasingly adopted for fabricating large-scale metallic components. However, reliable prediction and multi-objective optimization of weld-bead geometry remain critical challenges. This study systematically investigates the influence of voltage, wire feed speed (WFS), and torch travel speed (TS) on bead width (BW), bead height (BH), and microhardness (MH) using a structured Taguchi L18 experimental design. Statistical regression modeling and analysis of variance (ANOVA) were employed to identify significant process parameters and establish physically interpretable predictive relationships. To further assess potential nonlinear behavior, the XGBoost machine-learning algorithm was implemented as a surrogate model. Given the limited dataset size (n = 18), leave-one-out cross-validation (LOOCV) was adopted to obtain unbiased predictive performance estimates. Cross-validated results indicate that XGBoost provides performance comparable to linear regression within the investigated parameter window, achieving R²_LOOCV values of 0.879 (BW), 0.770 (BH), and 0.870 (MH). The comparatively lower prediction accuracy for BH is attributed to inherent melt pool instability and droplet transfer variability. Multi-objective grey relational analysis identified a locally optimal parameter combination of 15 V, 5.61 m/min WFS, and 10.9 mm/s TS within the defined design space, enabling simultaneous minimization of BW and maximization of BH and MH. The developed framework provides a physically grounded and cross-validated predictive optimization methodology for WAAM parameter selection within constrained single-bead deposition conditions.