Artificial neural network based delamination prediction in laminated composites

The present paper deals with an approach in predicting the presence of embedded delaminations (in terms of their size, shape and location) in fibre reinforced plastic composite laminates using natural frequencies as indicative parameters and artificial neural network as a learning tool. Here, a 3D finite element model has been used to model [0]20 graphite/epoxy plate having an embedded delamination at the interface of two [0]10 sub-laminates. Hundreds of finite element models have been run to generate natural frequencies up to 10 modes for various combinations of size, shape and location of embedded delamination in a graphite/epoxy plate and then these data have been used to train a back propagation neural network till the network learns to an acceptable level of accuracy. The trained network has been tested to predict the presence of a delamination along with its size, shape and location from the input natural frequencies. An optimum network architecture has been established which can effectively learn the pattern. It has been observed that, the network can learn effectively about the size, shape and location of the embedded delamination present in the laminate and can predict reasonably well when tested with unknown data set.