Compressive sensing of roller bearing fault using tunable Q-factor wavelet transform

Vibration signals generated by faulty bearings often constitute “big data”, and it is therefore difficult to sparsely decompose them for dimensional reduction. In addition, vibration signals are generally buried in noise, especially at the initial fault stages. This increases the difficulty of determining roller bearing status. Therefore, it is essential to reduce the data dimension and noise influence as much as possible. To overcome these problems, this paper proposes a compressed-signal-based fault detection method using a tunable Q-factor wavelet transform. With this technique, the sparsity of the decomposed vibration signals in the modified wavelet bases is increased, which helps to increase the compressibility and suppress noise of the vibration signals. Also, based on the decomposition of the vibration signals realized using the tunable Q-factor wavelet transform, a compressive sensing approach to roller bearing fault detection is developed. This technique can detect fault features from data that are under-sampled far below the Nyquist sampling rate. Compared with traditional fault detection methods, the proposed method increases the sparsity of vibration signal decompositions and relaxes the requirements on signal samples for roller bearing fault diagnosis. Successful applications in detecting roller bearing faults validate the effectiveness of the proposed method.