Time and frequency approaches to non destructive testing in concrete pillars using neural networks

In this paper, Multi Layer Perceptron neural networks have been trained to identify the position of defects in concrete structures analyzed using an ultrasound technique. A diagnostic model obtained by means of Finite Elements techniques has been used to model the ultrasound transmission through a concrete pillar of specified size affected by defects in different positions. The obtained signals have been processed both in the time and frequency domains, in order to reduce data dimensionality and to compute suitable features. Results show good accuracy in the identification of the position of the faults.