Fundamental study on implementation of semi-supervised learning for damage judgement of RC member by hitting response based on k-nearest neighbor algorithm

Abstract

This study deals with non-destructive damage judgement of RC member by hitting response based on k-nearest neighbor algorithm. To advance the performance of judgement results, the semi-supervised learning aided by the k-means method is implemented which can increase the number of labeled training data for the k-nearest neighbor algorithm. Validity of the proposed method is verified through an shear loading experiment with RC beam specimen. Furthermore, the effects of parameters used in the semi-supervised learning on judgement results are discussed.