| Summary: | 碩士 === 國立交通大學 === 土木工程系所 === 108 === This study aimed to solve the issues which was the way of distinguishing entropy value and decreased the occurrence of undefined entropy value on the entropy method. Therefore, refined composite multiscale cross-sample entropy (RCMSCE) was utilized to enhance the reliability of entropy value and a new structural health monitoring strategy based on RCMSCE and artificial neural network. A neural network model is developed in accordance with a numerical model which is derived from the entropy value under the ambient vibrations. First, RCMSCE was implemented to extract the damaged feature and used ETABS to generate training samples which was changed stiffness terms to construct various damage pattern. A neural network model was trained and built by the entropy value with these damage patterns. After a seismic event, the proposed artificial intelligence-based structural health monitoring is employed to detect damage location and extent. In this study, a seven-story model was create to validate the performance of proposed method. Subsequently, a seven-story steel benchmark experiment with fifteen damage cases was conduct to compare the difference between numerical and experimental model. The confusion matrix was applied to evaluate the results. The performance evaluation of the proposed Structural Health Monitoring system showed the increase of accuracy to locate damage.
|
|---|
