Modal analysisusing fiber optic sensors and neural networks for prediction ofcomposite beam delamination

Abstract

Extrinsic Fabry-Perot interferometric (EFPI) fiberoptic sensors and a neural network provided a health-monitoringcapability for laminated glass/epoxy composite beams. The EFPIsensors experimentally determined the first five modalfrequencies of the cantilever beams. The feedforwardbackpropagation neural network used these modal frequencies to predict the size and location of delaminations in the compositebeams. Beam modal frequencies shift as a function ofdelamination size and location. Five beams with prescribeddelaminations, as well as a `healthy' beam with no delaminations,were excited by a surface-mounted piezoelectric actuator at frequencies up to 1 kHz. All beams had an eight-ply symmetricglass/epoxy composite design, were fabricated simultaneously,and had length and width dimensions of 26.04 and 2.33 cm,respectively. The beams with flaws had different delaminationsizes ranging from 1.27-6.35 cm long prescribed in themid-plane, i.e. between the fourth and fifth plies. The neural network was trained using classical-beam theory and tested usingthe experimental EFPI data. The delamination size and locationpredictions resulting from the neural network simulation had anaverage error of 5.9 and 4.7%, respectively. Also, analyticalclassical-beam theory, finite element methods, and ceramicpiezoelectric sensors validated the EFPI modal frequencymeasurements.