Characterization and detection of delamination in composite laminates using artificial neural networks

A procedure has been developed for characterizing and detecting delaminations in composite structures using a combination of accurate analysis techniques, a strain-based damage index, artificial neural networks, and genetic algorithms. An improved laminate theory developed using a refined layerwise deformation field description to characterize the presence of multiple discrete delaminations in composite laminates is used to compute the layerwise in-plane modal strains. A damage indicator, based on the layerwise in plane modal strains, is used to identify the presence of both discrete and overlapping delamination in composite plates. An artificial neural network (ANN) model is developed to signal the presence of damage. It is then applied in conjunction with a genetic algorithm to describe the geometry of a random delamination for which the damage index distribution is already known. Four different types of delaminations -through-the-width, seeded, multiple seeded, and seeded overlapping delaminations, have been considered in the current study. The ANN model is found to work well as a predictive tool for detecting delaminations.