http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
건물 에너지 소비 시스템에서 이상 탐지 성능향상을 위한 그룹 기반 변수 선택 방법
정다현(Dahyun Jung),전창재(Chang-Jae Chun) 대한전자공학회 2023 대한전자공학회 학술대회 Vol.2023 No.6
Since buildings occupy a large part of the worlds energy consumption and CO₂ emissions, it is necessary to improve the building energy consumption efficiency to solve energy demands and environmental problems. Therefore, many studies have been conducted in the field of energy consumption anomaly detection in order to reduce energy waste in buildings and operate buildings more efficiently. However, many conventional studies have used different features to detect anomalies at the discretion of the researchers. This is because there are a wide range of features that may affect energy consumption and there is no clear standard or methods to effectively select them.. In this research, the feature variables used for anomaly detection of building energy consumption data are classified into 6 groups and the performance was compared when a combination of different groups was used as a model input feature variable.
이지호(Ji Ho Lee),김지우(Jiwoo Kim),전창재(Chang-Jae Chun) 대한전자공학회 2023 대한전자공학회 학술대회 Vol.2023 No.6
The majority of manufacturing companies currently rely on manual labor by skilled personnel to identify product defects. However, this humanbased manual inspection process may lead to significant losses in company revenue due to inconsistent inspection criteria and worker fatigue. Recently, to overcome this inefficiency of the inspection process, artificial intelligence (AI)-based automatic inspection technology has been proposed for various fields. This AI-based detection method performs better than the previous human-based inspection. However, a large amount of time and costs are required to develop the AI models. In this paper, we propose a method for constructing an efficient defect detection model for PCB components using a convolutional neural network and transfer learning. VGG16 demonstrates the best performance across all 13 components among the various models. Additionally, it is confirmed that the essential data quality is ensured for model performance through preprocessing, and performance improvement was achieved through data augmentation to overcome limited data availability.