설비진단을 위한 초음파 신호의 특징분석 적용

Abstract

Ultrasound signal is widely used to detect fault by heterodyned signal. Typically an expert will scan around the object with the scanning module while listening through headphones and observing a display panel. But this diagnosis procedure is required by specialized expert and hardly detect early defect. In this paper, Feature selection based on GA (genetic algorithms) is selected from the features of ultrasound signal on frequency domain and time domain. Then, by using the Support Vector Machine one of the machine learning, the performance of classification is evaluated by extracted features and selected features. The results of classification is compared with feature extraction based on PCA (principal component analysis). Therefore, the feature selected for each defect can be used as a reference by feature analysis for ultrasound.