Bae, Ji-Hoon
(Electronics and Telecommunications Research Institute, Daejeon, 34129, Republic of Korea)
,
Yeo, Doyeob
(Electronics and Telecommunications Research Institute, Daejeon, 34129, Republic of Korea)
,
Yoon, Doo-Byung
(Korea Atomic Energy Research Institute, Daejeon, 34057, Republic of Korea)
,
Oh, Se Won
(Electronics and Telecommunications Research Institute, Daejeon, 34129, Republic of Korea)
,
Kim, Gwan Joong
(Electronics and Telecommunications Research Institute, Daejeon, 34129, Republic of Korea)
,
Kim, Nae-Soo
(Electronics and Telecommunications Research Institute, Daejeon, 34129, Republic of Korea)
,
Pyo, Cheol-Sig
(Electronics and Telecommunications Research Institute, Daejeon, 34129, Republic of Korea)
In this paper, we propose a deep-learning-based pipe leak detection (PLD) technique using trajectory-based image features extracted from time-series acoustic data received from microphone sensor nodes. We developed root-mean-square-pattern and frequency-pattern images by reflecting the leakage signa...
In this paper, we propose a deep-learning-based pipe leak detection (PLD) technique using trajectory-based image features extracted from time-series acoustic data received from microphone sensor nodes. We developed root-mean-square-pattern and frequency-pattern images by reflecting the leakage signal characteristics in the time and frequency domains and used them for ensemble learning with the help of state-of-the-art residual networks. The experimental results obtained using the measured data of leakage signals in a laboratory-scale nuclear power plant environment are presented to validate the effectiveness of the proposed method for PLD. The results show that the proposed image features suitable for convolutional neural network-based deep-learning can provide reliable PLD performance in terms of classification accuracy despite the machine-driven complex noise environment.
In this paper, we propose a deep-learning-based pipe leak detection (PLD) technique using trajectory-based image features extracted from time-series acoustic data received from microphone sensor nodes. We developed root-mean-square-pattern and frequency-pattern images by reflecting the leakage signal characteristics in the time and frequency domains and used them for ensemble learning with the help of state-of-the-art residual networks. The experimental results obtained using the measured data of leakage signals in a laboratory-scale nuclear power plant environment are presented to validate the effectiveness of the proposed method for PLD. The results show that the proposed image features suitable for convolutional neural network-based deep-learning can provide reliable PLD performance in terms of classification accuracy despite the machine-driven complex noise environment.
이 논문을 인용한 문헌
관련 콘텐츠
*원문 PDF 파일 및 링크정보가 존재하지 않을 경우 KISTI DDS 시스템에서 제공하는 원문복사서비스를 사용할 수 있습니다.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.