DC Field | Value | Language |
---|---|---|
dc.contributor.author | 나경미 | ko |
dc.contributor.author | 정기욱 | ko |
dc.contributor.author | 이창훈 | ko |
dc.date.accessioned | 2022-08-05T00:00:12Z | - |
dc.date.available | 2022-08-05T00:00:12Z | - |
dc.date.created | 2022-08-02 | - |
dc.date.created | 2022-08-02 | - |
dc.date.issued | 2022-06-09 | - |
dc.identifier.citation | 2022 한국군사과학기술학회 종합학술대회 | - |
dc.identifier.uri | http://hdl.handle.net/10203/297821 | - |
dc.description.abstract | In a design process of a missile system, Monte-Carlo simulations are conducted for analysis of uncertainty that affects performance of the missile. Time-series data obtained from simulations not only can improve the understanding of the model, but also aid to estimate permanent model uncertainties. Therefore, this research focuses on the estimation of model uncertainty such as a location of center of pressure, fin bias that exist consistently during flight. Complex nonlinear relationships between the model uncertainty and states in flight can be approximated by 1-Dimensional convolutional neural networks(1D CNN) which are known as a proper model for adapting to time-series data. Features used for input in 1D CNN are extracted considering domain knowledge about missile dynamics and data which can be obtained from sensors. 1D CNN models estimate the uncertainty from the entire length of time-series data. As a result, the accurate model for the missile system is attained and the real-time application helps to continuously observe and manage the missile system. | - |
dc.language | Korean | - |
dc.publisher | 한국군사과학기술학회 | - |
dc.title | 모델 지식 및 신경망 기반 모델 불확실성 실시간 추정 기법 | - |
dc.title.alternative | Model Uncertainty Estimation Using Domain Knowledge and Neural Networks | - |
dc.type | Conference | - |
dc.type.rims | CONF | - |
dc.citation.publicationname | 2022 한국군사과학기술학회 종합학술대회 | - |
dc.identifier.conferencecountry | KO | - |
dc.identifier.conferencelocation | 제주컨벤션센터 | - |
dc.contributor.localauthor | 이창훈 | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.