Assessment of core strength of concrete by artificial neural networks

Abstract

The proposed work deals with the use of Ultrasonic pulse velocity technique as an alternative method to identify compressive strength of the core concrete. The use of non-destructive technique without causing damages to the structure is tedious with interpretation of results influenced by various factors. Hence, an empirical relationship is developed using artificial neural network model for creating a regression between pulse velocity and compressive strength of concrete core specimens. Tests were conducted on reinforced concrete cylinders at various orientation angles (0°, 45°, 90°). The tests were conducted based on the design of experiment using the Box-Behnken model. These results were trained using the Levenberg-Marquardt back propagation model with hidden layers. Results indicate that the prediction of core compressive strength for the grade mixes is nearer for the two-level factorial design with R2 = 0.897, and the sum of squared error is found to be 0.9968.