Machinery Fault Diagnosis using Continuous Wavelet Transform and Artificial Intelligence based classification

This work presents an investigation for a number of mechanical conditions (healthy, imbalance, misalignment, gear fault, bearing fault). The vibration conditions are simulated and based on real-time vibration data that are acquired from a Machinery Fault Simulator (MFS). The diagnosis process passes through two main stages after the data acquisition, where then the Continuous Wavelet Transform (CWT) method is applied to preprocess the obtained datasets (signals), and extract the features based on five statistical parameters namely: RMS, Kurtosis, Peak, Impulse Factor and Shape Factor. Then Artificial Intelligence (AI) based classification is applied using the Multilayer Feed-Forward Perceptron Neural Network (MLP) using different cases, where different number of neurons for the hidden layer and two different datasets of the input features with 250 features for each condition and also 2000 features for each condition once with normalized and also with non-normalized features to investigate the best cases for the performance classification in terms of neurons and features number, and also the impact of features normalization. The obtained results based on the classification performance using MLP-NN with the different cases are comparable, where the normalized features with less number of features and moderate neurons have shown better classification performance. Moreover, the results have shown the advantage of integrating CWT with the MLP-NN by providing significant classification rates for the different mechanical conditions.