A study on the application of MLP model trained using multiple signal features in aircraft bearing fault diagnosis

• Main Author: Xu, Linhan
• Other Authors: Soong Boon Hee
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2025
• Subjects: Engineering; FFT; WT; WPD; MLP; Fault diagnosis
• Online Access: https://hdl.handle.net/10356/182474
• Institution: Nanyang Technological University

At present, within intelligent bearing fault diagnosis techniques, the feature extraction phase rooted in signal processing is indispensable. This study investigates a fault diagnosis method for aircraft bearings based on a Multi-Layer Perceptron (MLP) model using voltage, current, and vibration signals. To reduce the time and computational cost associated with training models on raw data, preprocessing of input signals was performed. Features were extracted from aircraft bearing data using FFT, WT, and WPD methods to build the MLP-based fault diagnosis model. Experimental comparisons of feature extraction methods were conducted. Results show that FFT, which extracts features only in the frequency domain, performs poorly. WT and WPD methods, incorporating both time and frequency domain features, improved the results, but their accuracy was only about 80% when used individually. Combining features extracted by WT and WPD and inputting them into the MLP model significantly improved accuracy. Notably, integrating energy features of current signals with the time-frequency domain and energy features of signals achieved a fault diagnosis accuracy of 99.29%, substantially higher than the 92.02% accuracy obtained using only vibration signal features. Thus, combining multiple signal features significantly enhances fault diagnosis accuracy. Therefore, at the end of the study, this dissertation presents an MLP model that can be applied to aircraft bearing fault diagnosis based on multiple signal features training.