Structural damage classification using support vector machines.

摘要

In this research, a methodology to classify crack and corrosion metallic damages using a time-frequency representation method and support vector machines is investigated. Piezoelectric ceramic actuators are utilized to generate guided wave signals on a set of aluminum beam coupons with different damage features, such as types, locations, and thicknesses. The short-time Fourier transform is applied to analyze the measured signals. For damage classification, the spectrograms obtained from finite element models are employed to train a two-class support vector machine learning classifier. The classifier is able to correctly classify different types of damages based upon the measured signals collected from the unknown damage sources. A multiple-class classifier is also generated to predict the damage extent of crack samples.