본문 바로가기

Towards Reliable Intelligent Fault Diagnosis of Rolling Element Bearings: The Role of Signal Quality

문서 유형
연구출판물 · 학술대회 발표논문
등재
Scopus
피인용
Scopus 0
FWCI (Scopus)
0

저자

  1. Torabi, MohammadReza제1저자교신저자Isfahan University of Technology, Department of Mechanical Engineering, Isfahan, Iran
  2. Loghmani, AliIsfahan University of Technology, Department of Mechanical Engineering, Isfahan, Iran
  3. Ziaei-Rad, SaeedIsfahan University of Technology, Department of Mechanical Engineering, Isfahan, Iran
  4. 노유정Pusan National University, School of Mechanical Engineering, Busan, South Korea
  5. Zahabi, Sayed JalalIsfahan University of Technology, Department of Electrical and Computer Engineering, Isfahan, Iran

초록

Reliable and interpretable fault diagnosis of rolling element bearings is essential for condition monitoring and maintenance of rotating machinery. This study investigates the impact of vibration signal quality on the performance and reliability of deep learning based fault diagnosis, using the CWRU and PU benchmark datasets under varying fault sizes. Two neural network architectures, a Multilayer Perceptron (MLP) and a one dimensional Convolutional Neural Network (1D-CNN), are evaluated using multiple input representations, including raw signals, frequency spectra, envelope spectra, and cepstrum pre-whitened envelope spectra. The results demonstrate that signal quality, particularly the presence of classical fault related features, plays a decisive role in diagnostic performance and model generalization. When high quality signals are used for training, the models achieve significantly higher accuracy and exhibit behavior consistent with classical signal analysis principles. Among the evaluated representations, envelope based features outperform other inputs in most cases, especially when combined with the 1D-CNN architecture. The effects of signal length and data augmentation are also examined. Increasing input length improves frequency resolution, while Gaussian noise augmentation during training enhances robustness, yielding accuracy improvements of up to 8.65 percent. Overall, the findings highlight the importance of signal quality awareness, informed input design, and appropriate model selection for achieving reliable fault diagnosis in realistic industrial monitoring scenarios.

키워드

Deep LearningDomain ShiftReliable Fault DiagnosisRolling Element Bearing

식별자

Scopus EID
2-s2.0-105044976454