Predicting residual strength of non-linear ultrasonically evaluated damaged concrete using artificial neural network

Abstract

This paper deals with the combination of non-linear ultrasonic and artificial neural networks (ANNs) for the non-destructive evaluation of the damages in concrete under stressed state. Two networks, one using raw variables and another using dimensionless variables were trained and tested to predict concrete damages. Input data to the neural network is the time-domain signals of the received ultrasonic waves, obtained from the experimental studies carried out as reported in the earlier literature involving experimental data base of 75 ultrasonic measurements performed on concrete cubes with water–cement (w/c) ratios of 0.40, 0.50 and 0.60 respectively. Both networks were two-layer-perceptrons trained according to back-propagation algorithm. The results of this research highlight the potential of artificial neural networks for solving the problem of concrete damage evaluation using non-linear ultrasonic measurements. It was found that the proposed ANN models predict the strength of concrete laboratory cubes with low absolute errors. The performance of ANN model for predicting the residual strength of concrete using the raw data is better than the prediction using grouped dimensionless variables.