Machinery fault diagnosis based on a modified hybrid deep sparse autoencoder using a raw vibration time-series signal.

• Main Authors: Saufi, Syahril Ramadhan, Isham, Muhammad Firdaus, Ahmad, Zair Asrar, Abu Hasan, M. Danial
• Format: Article
• Published: Springer Science and Business Media Deutschland GmbH 2023
• Subjects: TJ Mechanical engineering and machinery
• Online Access: http://eprints.utm.my/106231/; http://dx.doi.org/10.1007/s12652-022-04436-1
• Institution: Universiti Teknologi Malaysia

Intelligent fault diagnosis (IFD) is an effective system to ensure the safe operation of mechanical components such as bearings, gears, and blades. The main challenge of IFD using traditional methods lies in finding the good features that reflect the machine conditions that need prior knowledge and essential expertise to identify the features. In order to solve this problem, this paper introduces a new IFD technique based on a deep sparse autoencoder (DSAE) using a raw vibration time-domain signal (1D time-series signal). A novel method called modified hybrid DSAE was developed to avoid the need for feature extraction and selection steps. First, a resilient backpropagation learning algorithm was employed on the three hidden layers of the DSAE network. Then, a combination of sigmoid and rectified linear unit (RELU) activation functions was proposed for the DSAE network. Finally, the hyperparameters of the DSAE network were optimized using the grey wolf optimizer technique (GWO). The proposed method was applied to analyze the 1D time-series signal of five machinery datasets, including three bearings, one gearbox, and one turbine blade. The analysis proved that the diagnosis performance achieved by the proposed model is highly reliable and applicable to fault diagnosis of machinery components. The model achieved 100 per cent diagnosis accuracy on four datasets (MFPT, CWRU, gearbox, turbine blade) and 95 per cent diagnosis accuracy on the MaFaulDa dataset. The results from the proposed model show superior diagnostic accuracy compared to other related studies.