제목 |
결함 데이터가 필요치 않는 비지도 학습 기반 차량 센서 고장 진단 알고리즘 개발 |
분야 |
차량동역학 및 제어 |
언어 |
Korean |
저자 |
조경호(부산대학교), 유승한(한국기술교육대학교), 안창선(부산대학교) |
Key Words |
Fault detection(결함탐지), GAN(적대적생성모델), Anomaly d etection(이상치 탐지), Vehicle sensors(센 서), Autonomous driving(자율주행), ADAS(운전자 보조 시스템) |
초록 |
This paper presents a new approach to fault detection in vehicle sensors that is not dependent on specific vehicle dynamics models. The proposed technique utilizes easily accessible normal signal data, thus eliminating the need for expensive fault data collection methods. It focuses on a discriminator, which is developed by using Generative Adversarial Networks(GANs), that differentiates between genuine sensor signals and synthetic signals generated by a signal generator. This discriminator acts as a fault detection mechanism that helps identify faulty sensors. This study presents the GAN-based development process, neural network structures, data collection methods, and validation procedures on fault detection. Validation through simulation with CarSim demonstrates the feasibility of designing fault detection algorithms without having to collect fault data. |
원문(PDF) |
다운로드
Journal Site
|