Scholarlink

This investigation attempts to use two commonly used algorithms, namely, LMA and SCG optimization techniques to predict the compressive and flexural strength of concrete incorporating processed (PRA) and treated recycled aggregate (TRA). The input parameters used were: TRA, PRA, Slump, and age of concrete. The two approaches were tested on a neural network model using error measurements and convergence characteristics. The LM algorithm had the best validation performance with an MSE of 0.07145 at epoch 58, whereas the SCG algorithm had 2.4266 at epoch 905. The LM model has a higher convergence rate, with a final gradient of 1.47 × 10⁻6 compared to 6.2953 for SCG. R-values near 0.999 for training, validation, and test sets revealed good predictive accuracy for both models. LM algorithm error histograms showed a narrower error distribution centered on zero, suggesting more accurate predictions. Despite six validation failures, both models stabilized and generalized across datasets. The LM algorithm predicted concrete strength more accurately and efficiently, producing lower error in fewer epochs. These results imply that LM is better for this application than SCG, offering accurate results with faster convergence.

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

ANN-Based Models for Strength Prediction of Concrete Incorporating Processed Recycled Aggregate

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ashutosh Shishodiya,
  • Velaga Sarath Babu,
  • Yogesh Iyer Murthy

摘要

This investigation attempts to use two commonly used algorithms, namely, LMA and SCG optimization techniques to predict the compressive and flexural strength of concrete incorporating processed (PRA) and treated recycled aggregate (TRA). The input parameters used were: TRA, PRA, Slump, and age of concrete. The two approaches were tested on a neural network model using error measurements and convergence characteristics. The LM algorithm had the best validation performance with an MSE of 0.07145 at epoch 58, whereas the SCG algorithm had 2.4266 at epoch 905. The LM model has a higher convergence rate, with a final gradient of 1.47 × 10⁻6 compared to 6.2953 for SCG. R-values near 0.999 for training, validation, and test sets revealed good predictive accuracy for both models. LM algorithm error histograms showed a narrower error distribution centered on zero, suggesting more accurate predictions. Despite six validation failures, both models stabilized and generalized across datasets. The LM algorithm predicted concrete strength more accurately and efficiently, producing lower error in fewer epochs. These results imply that LM is better for this application than SCG, offering accurate results with faster convergence.