Scholarlink

<p>Aluminum matrix composites reinforced with ceramic particulates are widely used in structural and lightweight engineering due to their high specific strength, stiffness, and wear resistance. However, the mechanical behavior of these composites strongly depends on the reinforcement fractions and powder metallurgy processing parameters, making accurate property prediction a&#xa0;nonlinear and multivariable challenge. In this study, a&#xa0;comparative machine-learning framework is developed to predict the tensile strength and hardness of aluminum–Silicon Carbide (SiC) composites fabricated within defined processing conditions. Experimental data were generated for the SiC contents of 0–9 wt.%, compaction pressures of 400–800 MPa, sintering temperatures of 350–580 °C, and sintering times of 90–145 min. An Artificial Neural Network (ANN) was optimized using Genetic Algorithm (GA) and Bayesian Optimization (BO) for hyperparameter tuning. The GA-optimized ANN demonstrated superior predictive performance (<i>R</i><sup>2</sup> = 0.9710 and <i>RMSE</i> = 5.93 MPa) compared to the BO-optimized model (<i>R</i><sup>2</sup> = 0.9543 and <i>RMSE</i> = 7.40 MPa). Our residual analysis confirmed improved stability and reduced systematic bias for the GA. The SHapley Additive exPlanations (SHAP) technique revealed that compaction pressure, SiC fraction, and sintering temperature dominate mechanical response, consistent with densification and diffusion-driven strengthening mechanisms. The developed framework provides a&#xa0;robust predictive tool valid within the investigated processing domain.</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

Genetic algorithm-enhanced machine learning for predicting tensile strength and hardness of aluminum composite: A comparative optimization and SHAP analysis

  • 期刊论文

  • 发表日期：

    2026年6月2日

查看全文
  • Jagadish Lolugu,
  • Mahesh Kaza,
  • Venkateswara Reddy Kunduru,
  • K. Venkateswara Reddy

摘要

Aluminum matrix composites reinforced with ceramic particulates are widely used in structural and lightweight engineering due to their high specific strength, stiffness, and wear resistance. However, the mechanical behavior of these composites strongly depends on the reinforcement fractions and powder metallurgy processing parameters, making accurate property prediction a nonlinear and multivariable challenge. In this study, a comparative machine-learning framework is developed to predict the tensile strength and hardness of aluminum–Silicon Carbide (SiC) composites fabricated within defined processing conditions. Experimental data were generated for the SiC contents of 0–9 wt.%, compaction pressures of 400–800 MPa, sintering temperatures of 350–580 °C, and sintering times of 90–145 min. An Artificial Neural Network (ANN) was optimized using Genetic Algorithm (GA) and Bayesian Optimization (BO) for hyperparameter tuning. The GA-optimized ANN demonstrated superior predictive performance (R2 = 0.9710 and RMSE = 5.93 MPa) compared to the BO-optimized model (R2 = 0.9543 and RMSE = 7.40 MPa). Our residual analysis confirmed improved stability and reduced systematic bias for the GA. The SHapley Additive exPlanations (SHAP) technique revealed that compaction pressure, SiC fraction, and sintering temperature dominate mechanical response, consistent with densification and diffusion-driven strengthening mechanisms. The developed framework provides a robust predictive tool valid within the investigated processing domain.