Deep transient feature learning for weak vibration signal detection. (July 2021)

• Record Type: Journal Article
• Authors: Li, Xiaomeng; Wang, Yi; Ruan, Hulin; Wang, Dong; Qin, Yi; Tang, Baoping
• Is Part Of: Measurement. Volume 179(2021)
• Publication Date: 2021-07
• Subjects: Deep transient feature learning -- Weak vibration signal -- Deep learning -- Repetitive vibration transients; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2021.109502
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: A deep filtering method is proposed for weak signal extracting. The deep transient feature extraction model is constructed based on simulated template signals. A mapping between time–frequency distribution (TFD) image of noisy signal and clean signal is obtained. The trained model can be generalized to practical industrial vibration signals without retaining. The experimental results indicate the proposed method outperforms conventional method. Abstract: Extracting repetitive vibration transients is a key step for machine fault diagnostics and prognostics. Most existing weak transient extracting techniques mainly rely on careful selection of optimal parameters. Besides, most of the currently available techniques cannot suppress in-band noise. Therefore, it is important to develop an intelligent method for extracting repetitive transients and suppressing in-band noise simultaneously. In this paper, a novel method based on deep transient feature learning is proposed to address this issue. The main idea of this paper is to construct a deep model based on simulated template signals. After the training process, time–frequency distribution (TFD) image of the noisy signal can be mapped to TFD image of pure repetitive transients. Afterwards, with the help of the phase spectrogram, clean repetitive transients can be reconstructed from the mapped TFD amplitudes. The experimental validation results indicate the proposed method is effective and reliable for repetitive transients extracting.