http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
진동 신호 기반 합성곱 신경망을 이용한 다양한 하중 조건의 유성기어박스 고장 진단
김수호(Sooho Kim),김현재(Hyunjae Kim),박정호(Jungho Park),윤병동(Byeng D. Youn) 대한기계학회 2018 대한기계학회 춘추학술대회 Vol.2018 No.12
Since a planetary gearbox have been frequently adapted for rotational system, the fault diagnosis of planetary gearbox has been highly required. In this purpose, the physics-based approaches have been suggested but they have required enough domain knowledge which is time-consuming to achieve. Hence, there have been a lot of attempts based on datadriven approach, especially employing machine learning method to overcome the requirements of domain knowledge. Even though these attempts have shown excellent performance, there is too high randomness on designing the architecture of machine learning. In the same time, the physical explanation of process in machine learning have been required, since it is related with the reliability of result. In this research, the End-to-end One-Dimensional Convolutional Neural Network (EODCNN) is proposed for fault diagnosis of planetary gearbox. In the process of designing architecture, the physical properties are considered to optimized the diagnosis performance and the effects of physical properties are compared. Furthermore, the process of trained model is investigated to discover the physical meaning which bring the reliability on the performance of diagnosis model.
Deep Neural Network based Disease Severity Regression for Diagnosis of Abdominal Aortic Aneurysm
Joo Hyeon Im(임주현),Sooho Kim(김수호),In Chan Lee(이인찬),Jin-Oh Hahn(한진오),Byeng D. Youn(윤병동) 대한기계학회 2020 대한기계학회 춘추학술대회 Vol.2020 No.12
This paper proposes an abdominal aortic aneurysm severity regression algorithm by combining a physical model and deep learning. There are three typical problems in providing personalized medical service: 1) insufficiency of data, 2) dispersion of disease severity, 3) lack of individuality. In this study, blood pressure waveform data was obtained by implementing AAA in arterial tree simulation model. The data was acquired when there was a disease and when it was normal, and individual diversity was given through parameter change. Also, the shape of the aortic aneurysm was made into four types. The data showed a similar tendency to the blood pressure waveform in the presence of an aneurysm in the literature. Severity regression was performed through a deep neural network using data from blood pressure waveforms with various severity levels as input. As a result, it was confirmed that the regression performed well through the deep neural network.
합성곱 신경망 기반의 위상 불변에 강건한 GIS PD 고장 진단 모델 개발
박종민(Jongmin Park),김선의(Sun Uwe Kim),김수호(Sooho Kim),정진교(Jin-gyo Jung),윤병동(Byeng D.Youn) 대한기계학회 2018 대한기계학회 춘추학술대회 Vol.2018 No.12
Gas Insulated Switchgear(GIS) is electrical equipment for stable transformation or transmission. Since GIS serves to transmit or disconnect high voltage currents, it is very important to maintain stable internal isolation condition. Previous researches have been conducted to detect Partial Discharge(PD) by Ultra High Frequency(UHF)sensors, focusing on partial discharge among the causes that deteriorate internal isolation condition. In this study, PD diagnostic method was developed using Phase Resolved Pulse Sequence(PRPS) image, which is an image of the UHF sensor signal of GIS used in the actual field. For actual data, there is a phase shift phenomenon of PRPS images due to measurement or synchronization errors, which makes it difficult to diagnose them. To solve this problem, a Convolutional Neural Network(CNN) based Deep learning architecture robust to phase shift is proposed and it showed high diagnosis accuracy compared to previous algorithm.