Comparison of time-domain and time-scale data in bearing fault detection

Recently various machine learning techniques have been applied as a solution to the problem of timely and accurate detection of motor bearing faults. Conventional decision support systems consist of feature extraction and classification phases which require manual design and optimization of each block separately with increased computational cost. In our work, one dimensional Convolutional Neural Networks (1D CNN) with an inherent adaptive design are applied to efficiently combine the feature extraction and classification into a single learning body. Wavelet packet decomposition (WPD) is applied to represent the bearing vibration signals in time-scale domain efficiently. We then compare performance of raw time-domain vibration input and time-scale transformed input signals for classification of bearing faults using real experimental data.