Identification of Fiber-reinforced Plastic Failure Mechanisms from Acoustic Emission Data using Neural Networks

Abstract

The identification of the type of discontinuities or failure mechanisms within fiber-reinforced plastic (FRP) structures normally requires the use of local nondestructive testing (NDT) methods, which is time and labor intensive. The global NDT methods, e.g., the use of acoustic emission (AE) data, are viewed as a more powerful alternative for the identification of FRP failure mechanisms. Despite numerous investigations on the subject, no specific conclusions have been reached. In this study, the identification of the various failure mechanisms of FRP using AE data is investigated. The neural network technique is used to perform pattern recognition of AE data for the identification of FRP failure mechanism. An extensive experimental program, using coupon and full-scale specimens, is conducted to construct the AE database for training and testing the neural networks. Two network systems are developed based on two different training approaches: backpropagation and probabilistic method. In addition, two levels of neural networks - primary and secondary - are used to enhance the accuracy of the prediction. Various AE correlation plots are used as trial input data to feed the networks. It is demonstrated that the identification results from using the proposed network systems are very promising, with the overall performance of up to 97% accuracy.