Laplacian Support Vector Machines as Data Classifier in Machine Learning Approaches of Structural Health Monitoring

作     者：Fazeli, Hassan Hassani, Nemat Safi, Mohammad 

作者机构：Shahid Beheshti Univ Civil Engn Coll Abbaspour Tech Campus Tehran Iran 

出 版 物：《JOURNAL OF EARTHQUAKE AND TSUNAMI》 (J. Earthqu. Tsunami)

年 卷 期：2025年第19卷第1期

核心收录：

学科分类：07[理学] 0708[理学-地球物理学] 0814[工学-土木工程] 

主　　题：Statistical pattern recognition support vector machines method semi-supervised learning algorithms system identification 

摘      要：With the progress of technologies in data collection systems, structural engineers are faced with a large amount of unlabeled data gathered from various states of different structures, directing studies toward using improved classifiers in machine learning approaches of structural health monitoring. In this research, Laplacian Support Vector Machines (LapSVMs) are used as a semi-supervised learning algorithm to classify mentioned data. Using vibrational data of structure, dynamic properties of the structure are extracted. Modal strain energy is used as damage sensitive features (DSF) to perform damage assessment in a statistical pattern recognition framework by a semi-supervised learning algorithm using LapSVMs to classify unlabeled and labeled data. Also, Support Vector Machines (SVMs) and Regularized Least Square Classifier (RLSC) are used as classifiers to compare results. To compare the effectiveness of the proposed algorithm, different states of structural response are determined by labeled and unlabeled data. These results show high accuracy of LapSVM methods compared to others in cases where the labeled dataset is small.