Scholarlink

Cement is one of the most commonly used materials on earth; however, it is associated with environmental concerns. While the focus of the world is on sustainable development goals, cement is being used with environmentally friendly materials, e.g., metakaolin. There are many available lab test methods for the compressive strength of mortar; however, they are not only expensive and time-consuming but also require thorough quality control. Also, there is a need to find the best consistency before 28 days. Empirical methods and basic regression techniques fail to incorporate the complex non-linear behavior and high number of input parameters of the mortar. The study presents a machine-learning-based simulation model for the prediction of compressive strength of metakaolin mortar using the Group Method of Data Handling (GMDH) and Extreme Learning Machine (ELM) models. For the purpose, a dataset totaling 276 is gathered from the available literature and validated using a Pearson correlation matrix and sensitivity analysis. The best performance of the model is achieved with 15 maximum layer neurons, 4 maximum layers, and 0.6 selection pressure for GMDH. Best performing ELM model is configured for 50 number of hidden neurons. The study concludes that both ELM and GMDH are a robust model for prediction of compressive strength of mortar (R2 = 0.9 for both the models in testing) and can be used as an alternative model once it is trained and tested on the in-situ data results, however ELM closely outperforms GMDH.

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

State-Of-The-Art ML-Based Prediction Models for Metakaolin-Based Mortar Using ELM and GMDH

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Manish Kumar,
  • Rishu Anand,
  • Krishna Deep,
  • Pursottam Rai

摘要

Cement is one of the most commonly used materials on earth; however, it is associated with environmental concerns. While the focus of the world is on sustainable development goals, cement is being used with environmentally friendly materials, e.g., metakaolin. There are many available lab test methods for the compressive strength of mortar; however, they are not only expensive and time-consuming but also require thorough quality control. Also, there is a need to find the best consistency before 28 days. Empirical methods and basic regression techniques fail to incorporate the complex non-linear behavior and high number of input parameters of the mortar. The study presents a machine-learning-based simulation model for the prediction of compressive strength of metakaolin mortar using the Group Method of Data Handling (GMDH) and Extreme Learning Machine (ELM) models. For the purpose, a dataset totaling 276 is gathered from the available literature and validated using a Pearson correlation matrix and sensitivity analysis. The best performance of the model is achieved with 15 maximum layer neurons, 4 maximum layers, and 0.6 selection pressure for GMDH. Best performing ELM model is configured for 50 number of hidden neurons. The study concludes that both ELM and GMDH are a robust model for prediction of compressive strength of mortar (R2 = 0.9 for both the models in testing) and can be used as an alternative model once it is trained and tested on the in-situ data results, however ELM closely outperforms GMDH.