http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
시계열 데이터의 Denoising CNN 적용을 통한 물리 모델 기반 베어링의 잔여수명 추정
채선규(Sun Geu Chae),강민기,권준(Jun Kwon),배석주(Suk Joo Bae) 대한기계학회 2021 대한기계학회 춘추학술대회 Vol.2021 No.4
많은 산업 분야에서 시스템 또는 구성 요소 고장을 방지하는 것이 중요하며, 갑작스러운 고장을 방지하기 위해 기존의 정기적 보수를 통한 방식이 널리 사용되지만 효율성 및 신뢰성 요구를 충족하지 못한다. 이에 따라서 지능형 예지보전 (PHM) 기술이 중요해지고 있으며, 해당 분야의 가장 주요영역 중 하나는 잔여수명 (RUL) 추정이다. 전통적으로, RUL의 추정은 물리적 속성에 대한 충분한 사전 지식에 달려 있으나, 구성 요소에 대한 사전 지식을 얻는 것은 복잡하거나 많은 경우에 불가능하기 때문에, 확보된 데이터만을 사용하여 잔여 수명을 예측하는 데이터 중심 접근법이 많은 영역에서 제안되고 있다. 그러나 데이터 중심의 접근 방식을 사용하여 구성 요소의 수명을 예측하는 것은 과적합 문제와 모델의 학습 과정에서 과도한 리소스를 사용하기 때문에 한계가 존재한다. 학습과 현장에서의 적용 사이의 정확성과 학습 시간에 영향을 끼치는 이러한 문제를 극복하기 위하여 본 논문에서는 물리 모델에 기반한 노이즈 제거 단계와 잔여수명 추정 및 고장 진단 단계로 구성된 CNN 모델을 제안한다. 본 연구를 통하여 물리 모델 기반의 시뮬레이션 데이터를 통하여 고장 신호만을 출력할 수 있도록 학습된 CNN을 통하여 베어링의 잔여 수명 추정 문제에 접근함으로써 빠르고 정확한 추정이 가능함을 시사한다. Preventing system or component breakdown is critical in many industry areas. To prevent sudden failures, traditional approach of scheduled maintenance is used widely but fails to meet the demands of efficiency and reliability, thus the approach of intelligent prognostic and health management (PHM) technology is becoming important, and one of the most important areas of PHM is estimating remaining useful life of a component. Traditionally, estimating the RUL depends on sufficient prior knowledge of degradation process. However, acquiring the prior knowledge of the component is complicated or even not possible in many cases, thus data-driven approach, which predicts the remaining useful life by only using the acquired data, is proposed in many areas. But the accuracy of estimating the lifespan of the component using data-driven approach is still not satisfactory due to overfitting issues and using excessive resources during the process of training the model. To overcome these issues, which affects the accuracy and time gap between off-line training and on-line estimation, multi-stage convolutional neural networks comprised of two stages, denoising stage and estimating stage, is proposed in this paper. The result of this study suggests that, by approaching the problem of estimating the RUL of bearings using multi-stage CNN trained with simulated physics-based model, faster and accurate estimation is possible.
채선규(Sun Geu Chae),김규리(Gyu Ri Kim),배병용(Byeong-Yong Bae),배석주(Suk Joo Bae) 한국신뢰성학회 2021 신뢰성응용연구 Vol.21 No.4
Purpose: Failure in thermal power plant generators has high safety and financial risk. Diagnostics and Prognostics to detect abnormality in advance are crucial for failure prevention. Methods: In this research, fast independent component analysis was applied to select key features from sensor data, and abnormalities were detected when unnatural variation existed in multivariate control charts. Results: The proposed framework was applied to the dataset acquired from a thermal power plant, and exhibited promising results in detecting and predicting incipient failures. Conclusion: From the analytical results of an example, it was found that the proposed methodology has potential in failure diagnostics and prognostics to increase the availability of facilities through early detection of incipient failures.
PEMFC 고분자 막의 전기화학적 가속 열화에 미치는 평가조건들의 영향
오소형,유동근,배석주,채선규,박권필,Sohyeong Oh,Donggeun Yoo,Suk Joo Bae,Sun Geu Chae,Kwonpil Park 한국화학공학회 2023 Korean Chemical Engineering Research(HWAHAK KONGHA Vol.61 No.3
In order to improve the durability of the proton exchange membrane fuel cell (PEMFC), it is important to accurately evaluate the durability of the polymer membrane in a short time. The test conditions for chemically accelerated durability evaluation of membranes are high voltage, high temperature, low humidity, and high gas pressure. It can be said that the protocol is developed by changing these conditions. However, the relative influence of each test condition on the degradation of the membrane has not been studied. In chemical accelerated degradation experiment of the membrane, the influence of 4 factors (conditions) was examined through the factor experiment method. The degree of degradation of the membrane after accelerated degradation was determined by measuring the hydrogen permeability and effluent fluoride ion concentration, and it was possible to determine the degradation order of the polymer membrane under 8 conditions by the difference in fluoride ion concentration. It was shown that the influence of the membrane degradation factor was in the order of voltage > temperature > oxygen pressure > humidity. It was confirmed that the degradation of the electrode catalyst had an effect on the chemical degradation of the membrane.