Rotor Unbalance Severity Detection Using Maximum Overlap Discrete Wavelet Transform

The rotor unbalances a critical fault that increases stress at rotational parts like bearings and gears, resulting in higher power consumption and early machinery wear. The general behavior of the vibration spectrum under this fault changes with strength and rotational speed. To address this problem, the presented work proposes frequency domain data fusion of vibration signals obtained from sensors placed at three different locations. The fused signal retains maximum spectral information, which decomposes into a multi-scale matrix using energy-preserving maximum overlap discrete wavelet transform. To analyze the severity of unbalance, the decomposed scale matrix is encoded into a contour plot using the mean absolute deviation of individual scales as iso-reference lines. Finally, a two-stage classification is performed using a convolutional neural network. The proposed method is tested using a publicly available dataset from Fraunhofer Institute for Integrated Circuits. The results show an overall classification accuracy of 97.05% for unbalance severity which is significantly better than other studies using single-sensor data.