하이퍼튜브 가이드웨이의 기하학적 이상상태 진단을 위한 특징 분석

Feature analysis was performed to predict and classify geometric anomalies of a Hypertube guideway using real-time sensor signals. In this study, geometric types of a guideway were classified into four classes and a dynamic model of the Hypertube vehicle was constructed. Based on simulation data obtained while driving on various guideway profiles, a total of 22 kinds of time-and frequency-domain features were extracted and effective features were selected among them. At this time, two feature selection approaches, the so-called multi-class (four types) and binaryclass (two types) feature selection approaches, were used individually. The performances of selected features from the two approaches were tested by applying them to guideway predictive models, which were based on an artificial neural network. As a result, it was demonstrated that the selected feature set using binary-class feature selection approach was superior to the other sets, achieving a predictive classification accuracy greater than 90%.