Weak fault detection in bearing vibration signals remains a challenging task due to the low energy of fault-induced impulses and their susceptibility to noise and interference. To address this issue, this study proposes a Bearing fault diagnosis framework that integrates Multipoint Optimal Minimum Entropy Deconvolution Adjusted (MOMEDA), envelope analysis, and a Convolutional Neural Network (CNN). First, characteristic fault frequencies derived from bearing geometry are used to determine fault-specific periods, and MOMEDA is applied to selectively enhance periodic impulsive components corresponding to each fault type. The enhanced signals are then processed using the Hilbert Transform to extract the envelope, followed by Fourier Transform to obtain the envelope spectrum. Finally, the extracted frequency-domain features are used as inputs to the CNN-based deep learning model for fault classification. The proposed approach effectively enhances weak fault signatures and improves their representation in the frequency domain, enabling more reliable fault identification under noisy conditions.

1. 서 론
    1.1 연구 배경 및 필요성
    1.2 연구 범위
2. 관련문헌 연구
3. 베어링 고장진단 방법론
    3.1 MOMEDA
    3.2 힐베르트 변환
    3.3 푸리에 변환
    3.4 1D-CNN
    3.5 MOMEDA 기반 베어링 고장 진단
4. 실험
    4.1 데이터 소개
    4.2 실험 설계
    4.3 실험 수행
    4.4 실험 결과
5. 결 론
Acknowledgement
References

• Jiwan Jeong(Department of Industrial Systems Engineering, Kyonggi University) | 정지완 (경기대학교 산업시스템공학과)
• Kwanghui Shin(Department of Industrial Systems Engineering, Kyonggi University) | 신광희 (경기대학교 산업시스템공학과)
• Yongsoo Kim(Department of Industrial Systems Engineering, Kyonggi University) | 김용수 (경기대학교 산업시스템공학과) Corresponding author