ScholarWorks@경상국립대학교

초록

This paper presents a machine-learning pipeline that predicts the rheological properties of fresh-state concrete based on concrete mixing components using various machine-learning algorithms. The well-known idea of a correlation between rheological parameters and conventional fluidity values, several tries at matching rheological parameters with flow values have been suggested. Even though some successful studies were able to match two related values, each research showed a different relationship depending on the case. However, in this study, a reliable and sustainable rheology parameter prediction model is suggested. The prediction was based on the mixing components of concrete by building a pipeline that sequentially integrates models that predict the physical properties of specific concrete types using various machine learning algorithms. A pipeline was built to sequentially connect the two models evaluated as having a desirable prediction performance, and the rheology parameter was predicted by inputting the mixing component. To validate the developed model, the experimental data was compared with the predictions generated by the model. As a result, the flow prediction error rate was 4.96%, and the yield stress prediction error rate was 6.59%, which is a favorable prediction performance of a constructed pipeline. This study presents a new method that can accurately predict the physical properties of fresh state concrete based on concrete mixing factors. This method will increase efficiency and ensure quality control of fresh state concrete.

키워드