Structural damage detection in plates using a deep neural network–couple sparse coding classification ensemble method
Vahid Bokaeian, Faramarz Khoshnoudian, Milad Fallahian
Abstract
The present study aims at identifying damages in plate structures by applying a pattern recognition–based damage detection technique using the frequency response function. The large number of degrees of freedom is one of the crucial obstacles in the way of accurately identifying damages in plate structures. On the other hand, frequency response functions include many details that dramatically lower the computing speed and enlarge the memory needed for storing data, hampering the application of this method. Furthermore, this study performs principal component analysis as an authoritative feature extraction method with the purpose of reducing the dimensions of the measured frequency response function data and generating distinct feature patterns. Also, because there has been no individual optimal classifier applicable to all problems, an ensemble comprising two powerful classifiers containing couple sparse coding classification and deep neural networks is used to predict the structure damage. This study evaluates the accuracy of damage detection by the proposed method in square-shaped structural plates with the lengths of 1 m and 2 m under different damage scenarios, namely, single and multiple element.