Comparative analysis of combined non-destructive testing methods for the indirect estimation of the heating temperature of fire-damaged concrete using machine learning

This study compared four non-destructive testing methods (colorimetry, crack analysis, rebound hammer, and ultrasonic pulse velocity) for diagnosing the heating temperature of fire-damaged concrete and applied various combined non-destructive testing methods to expand the limited scope of existing combined methods. To this end, concrete specimens with different compressive strengths were fabricated, and heating experiments were conducted at 400–900 ℃. Random forest algorithm was used to compare the heating temperature prediction accuracies of the single and combined non-destructive testing methods. Colorimetry showed the highest accuracy among the single non-destructive testing methods, followed by ultrasonic pulse velocity and crack analysis. The rebound hammer showed the lowest accuracy and had a limitation in that it was difficult to measure the specimens exposed to high temperatures owing to destruction. The combined non-destructive testing method, which combines two non-destructive testing methods, showed higher accuracy in estimating the heating temperature than the single non-destructive testing methods. The combination of ultrasonic pulse velocity and colorimetry showed the highest accuracy in predicting the heating temperature. Compared to using a single non-destructing testing method, the use of combined non-destructive testing methods generally demonstrated higher prediction accuracy for heating temperature. When all four non-destructive testing methods were combined, the highest prediction accuracy was observed among all combinations.