Scholarlink

<p>Additive Manufacturing (AM) has revolutionized the fabrication of complex lattice structures; however, accurately predicting their mechanical properties remains a significant challenge due to the intricate interactions between geometric and process parameters. This study introduces a hybrid machine learning framework that integrates metaheuristic optimization algorithms—Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), Simulated Annealing (SA), and Cohort Intelligence (CI)—with Decision Tree (DT) and Support Vector Machine (SVM) regressors to predict the compressive strength (MPa) of FDM-printed Polylactic Acid (PLA) lattice structures. Using experimental data from 25 physical specimens, key design features—infill density, infill pattern, strut diameter, and strut type—served as input variables. Rigorous evaluation through nested cross-validation ensured robustness and generalizability. Among all models, the SVM-GA hybrid achieved the highest predictive accuracy, with an R² of 0.874 ± 0.02, MSE of 3.882 ± 0.30&#xa0;MPa², and MAE of 1.422 ± 0.08&#xa0;MPa. This marks a ΔR² of + 0.224 and a 63.2% reduction in MSE compared to the baseline SVM. Model interpretability was enhanced using SHAP (SHapley Additive exPlanations) analysis, which identified infill density (|SHAP| = 5.66&#xa0;MPa) and strut diameter (|SHAP| = 3.70&#xa0;MPa) as the most influential features, consistent with fundamental mechanical principles. This interpretable, data-driven framework provides a validated tool for AM design optimization, minimizing the need for costly trial-and-error experimentation. While current limitations include a constrained dataset size, future work will focus on expanding data diversity, incorporating additional process variables, and exploring real-time predictive capabilities. Overall, the study bridges machine learning and domain expertise, advancing predictive modeling in additive manufacturing.</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

Enhancing predictive modeling for additive manufacturing components using hybrid machine learning algorithms

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月27日

查看全文
  • Mustak Shaikh,
  • Parvez Patel,
  • Vijaykumar S. Jatti,
  • K. Balaji,
  • Vikas Gulia,
  • Shahid Tamboli,
  • Laxam P. Khose,
  • A. Saiyathibrahim,
  • R. Murali Krishnan

摘要

Additive Manufacturing (AM) has revolutionized the fabrication of complex lattice structures; however, accurately predicting their mechanical properties remains a significant challenge due to the intricate interactions between geometric and process parameters. This study introduces a hybrid machine learning framework that integrates metaheuristic optimization algorithms—Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), Simulated Annealing (SA), and Cohort Intelligence (CI)—with Decision Tree (DT) and Support Vector Machine (SVM) regressors to predict the compressive strength (MPa) of FDM-printed Polylactic Acid (PLA) lattice structures. Using experimental data from 25 physical specimens, key design features—infill density, infill pattern, strut diameter, and strut type—served as input variables. Rigorous evaluation through nested cross-validation ensured robustness and generalizability. Among all models, the SVM-GA hybrid achieved the highest predictive accuracy, with an R² of 0.874 ± 0.02, MSE of 3.882 ± 0.30 MPa², and MAE of 1.422 ± 0.08 MPa. This marks a ΔR² of + 0.224 and a 63.2% reduction in MSE compared to the baseline SVM. Model interpretability was enhanced using SHAP (SHapley Additive exPlanations) analysis, which identified infill density (|SHAP| = 5.66 MPa) and strut diameter (|SHAP| = 3.70 MPa) as the most influential features, consistent with fundamental mechanical principles. This interpretable, data-driven framework provides a validated tool for AM design optimization, minimizing the need for costly trial-and-error experimentation. While current limitations include a constrained dataset size, future work will focus on expanding data diversity, incorporating additional process variables, and exploring real-time predictive capabilities. Overall, the study bridges machine learning and domain expertise, advancing predictive modeling in additive manufacturing.