Evaluation of compressive strength of concrete modified with admixtures using arithmetic optimization-based estimation algorithms

期刊论文
发表日期：

2024年6月25日

查看全文

Yaxuan Zhao

摘要

In the present study, various hybrid models are applied for developing estimation techniques to forecast the compressive strength ( $CS$ $CS$ ) of concretes, including metakaolin ( $MK$ $MK$ ) and fly ash ( $FA$ $FA$ ). These models can be valuable for the construction and concrete industry by providing a means to estimate $CS$ $CS$ under different conditions and ages of concrete. The models developed in this study can have practical applications in the concrete industry, helping engineers, architects, and construction professionals to make informed decisions regarding concrete mix designs, strength assessments, and aging effects. For this purpose, four various algorithms have been considered, named multilayer perceptron ( $MLP$ $MLP$ ), radial basis function ( $RBF$ $RBF$ ), support vector regression ( $SVR$ $SVR$ ), and least square $SVR$ $SVR$ ( $LSSVR$ $LSSVR$ ) to estimate $CS$ $CS$ by considering the most effective input parameters. These models are linked with the arithmetic optimization algorithm ( $AOA$ $AOA$ ) for raising the model's precision in estimating by determining the optimal values of hyperparameters ( $AOMLP$ $AOMLP$ , $AORBF$ $AORBF$ , $AOSVR$ $AOSVR$ , and $AOLSSVR$ $AOLSSVR$ ). The $AOMLP$ $AOMLP$ , $AOSVR$ $AOSVR$ , $AORBF$ $AORBF$ , and $AOLSSVR$ $AOLSSVR$ algorithms accurately estimate the compressive strength of concrete mixed with metakaolin and fly ash, with ${R}^{2}$ $R^{2}$ values exceeding 0.998 during both training and testing phases. Between systems, the $AOLSSVR$ $AOLSSVR$ model depict the highest values of ${R}^{2}$ $R^{2}$ in train and test phases at 0.9999 and 0.9998, respectively, as well as the lowest values in error-based metrics such as performance index ( $PI$ $PI$ ) at 0.0012 and 0.0021, lower than $AOSVR$ $AOSVR$ at 0.0022 and 0.0022, $AOMLP$ $AOMLP$ series and finally $AORBF$ $AORBF$ model at 0.0058 and 0.0063, in the training and testing stages.