MACHINE LEARNING APPROACH FOR DAMAGE DETECTION IN COMPOSITE STRUCTURE USING DYNAMIC RESPONES

Abstract

Structural Health Monitoring is a process of damage identification, localization, classification and
prediction of remaining life of a structure. In this project, damage identification for the composite plates is
carried out using machine learning techniques. Six composite plates of unidirectional glass-epoxy in cross-ply
configuration is fabricated for the project. One plate is healthy, and five plates are damaged in the form of
delamination at various locations and of various sizes. The frequency domain and time domain features are
extracted from this dynamic response. These features then act as an input for data driven techniques for damage
identification purpose. The machine learning techniques mostly include classification methods and supervised
learning technique. A healthy and a damaged plate are used for training the classifier, whereas remaining four
plates are used for identification purpose. The damage identification of four laminated beams is compared using
decision tree and two ensemble methods namely Rotation Forest and Bagging with decision tree as the base
classifier. It isobserved that the classification accuracy of ensemble methods is much higher than the decision
tree classifier