This research investigates different techniques of condition monitoring and fault diagnosis of rotating machines. These techniques are the classical Fast Fourier Transform (FFT), Discrete Wavelet Transform (DWT) coupled with different topologies of Neural Networks. A method for extracting features signal that is a combination of the horizontal and the vertical vibration time series is proposed. A technique for signal pre-processing for calculating the input feature is also adopted. The cumulants of magnitude of the vibrations provide a useful set of features for the detection of unbalance and rub faults. Pre-processing of the vibration signal is showed to amplify relevant spectral features improving the classification success. Results based on the data collected with a simple test rig that allow the simulation of rub and/or unbalance fault(s) are presented. For Neural Networks the results show that the performance of Self-Organizing Map (SOM) gives higher classification rate than the Feed-Forward Neural Networks (FFNN). A compound Neural Network with wavelet has classified the correct condition in over 99% of cases.
El-Mashad, Y. (2004). CONDITION MONITORING AND FAULT DIAGNOSIS OF ROTATING MACHINERY USING WAVELET AND NEURAL NETWORKS APPROACHES. ERJ. Engineering Research Journal, 27(1), 15-24. doi: 10.21608/erjm.2004.82601
MLA
Yehia El-Mashad. "CONDITION MONITORING AND FAULT DIAGNOSIS OF ROTATING MACHINERY USING WAVELET AND NEURAL NETWORKS APPROACHES". ERJ. Engineering Research Journal, 27, 1, 2004, 15-24. doi: 10.21608/erjm.2004.82601
HARVARD
El-Mashad, Y. (2004). 'CONDITION MONITORING AND FAULT DIAGNOSIS OF ROTATING MACHINERY USING WAVELET AND NEURAL NETWORKS APPROACHES', ERJ. Engineering Research Journal, 27(1), pp. 15-24. doi: 10.21608/erjm.2004.82601
VANCOUVER
El-Mashad, Y. CONDITION MONITORING AND FAULT DIAGNOSIS OF ROTATING MACHINERY USING WAVELET AND NEURAL NETWORKS APPROACHES. ERJ. Engineering Research Journal, 2004; 27(1): 15-24. doi: 10.21608/erjm.2004.82601