Estimation of exposed temperature for fire-damaged concrete using support vector machine

A support vector machine (SVM) based on the statistical learning algorithm is used for estimation of the exposed temperature of fire-damaged concrete structures. The experimental plan specifically designed for SVM modeling was described in details and carried out. In total, 76 groups of experimental data on the deterioration of ultrasonic pulse velocity, splitting tensile strength and moisture absorption of concrete were obtained accordingly for various exposed temperatures. Afterwards, 80 sets of training and testing sample points were then randomly selected from these experimental data and used for the SVM analysis. Based on the output results from SVM analysis, the most effective parameter to raise the accuracy index of the correct estimation of exposed temperature was identified as the ultrasonic pulse velocity of concrete. The accuracy of estimation for the SVM analysis increases with the increases of number of effective parameters and the ratio of training data sets to total data sets being considered in the calculation of SVM modeling. This study shows that the SVM modeling has a strong potential for properly estimating the exposed temperatures.