Abstract
The increased use of composite materials and their relatively high cost and limited availability make it essential to develop low cost, effective nondestructive testing and inspection techniques (NDT/NDI). One of the oldest and widely used NDT/NDI methods is the coin tap test. The objective of this research was to determine if the sound signals generated by tapping a composite sandwich panel could be classified by an artificial neural network (ANN) as originating from damaged or non-damaged areas on the panel and if possible, to make accurate damage level assessments. Tap sound signals were recorded from several test panels using an ordinary condenser microphone and related equipment. Two separate signal-preprocessing techniques were employed, one using Fourier transforms and one using Wavelet transforms. Wavelet transformation of the signals tended to produce the best results. Artificial neural network configurations were developed using the backpropagation-learning algorithm that correctly classified damaged vs. undamaged signals with 100% accuracy. The results further showed the potential of this process for accurately predicting the damage level present to within ±10%. Overall, the results showed the potential for using a combination of signal characteristic analysis with ANN's trained to recognize and classify the characteristics of simple tap test acoustic signals as an effective, low cost NDT/NDI technique.