|
Title |
Concrete Deterioration Prediction Model Subjected to High Temperature Using a Machine Learning?Based Non?Destructive Method
|
|
Authors |
김원창(Won-Chang Kim) ; 이태규(Tae-Gyu Lee) |
|
DOI |
https://doi.org/10.4334/JKCI.2025.37.4.417 |
|
Keywords |
콘크리트; 고온; 머신러닝; 비파괴; 통계 concrete; high temperature; machine learning; non-destructive; statistic |
|
Abstract |
In this study, a statistical analysis and a machine learning-based model were developed and evaluated to predict concrete quality using data from a previous study on concrete strength prediction via ultrasonic pulse velocity (UPV) after exposure to high temperatures. A total of 22 studies on normal concrete were collected, and the data were categorized into four levels (W/C50, W/C40, W/C30, and W/C20) based on the W/C ratio. The results showed that both the mean and the standard deviation of compressive strength increased as the W/C ratio decreased. However, UPV and residual compressive strength did not exhibit significant variations across the categories. ANOVA tests performed on the four W/C ratio groups revealed statistically significant differences in compressive strength, except between W/C50 and W/C40. No significant differences were observed for UPV and residual compressive strength. In the simple linear regression analysis, although the p-value was below the significance threshold of 0.05, the R2 was very low, at about 0.47. Furthermore, in the multiple linear regression analysis, the residuals did not satisfy the assumption of homoscedasticity. When building and evaluating models using machine learning?based regression algorithms, both Random forest and XGBoost performed very well, with R2 values above 0.92 and RMSE values below 0.1.
|