ABSTRACT
The aim of this work is to present a novel methodology for damage detection in a structure. In this method, new features based on Short Time Fourier Transform are used, dimensionality reduction was carried out by Principal Component Analysis and the classification was performed using Logistic regression. The efficacy of this method is demonstrated by detecting the presence of damage in a cantilever beam using the features based on transformation of displacement response.
Even though the difference between the displacement waveforms from damaged and the undamaged beam is not clearly perceptible, the difference is clearly visible in the principal components’ space. Data belonging to damaged and undamaged classes are linearly separable up to a certain level of noise after which linear separability is lost. A linear decision boundary was obtained corresponding to a particular noise level after the mean normalization and feature scaling of the data. Feasibility of dimensionality reduction is ensured by checking the loss percent in the reduction process. And generalization ability of the classifier has been assessed on some test sets.
