Adaptive adjacent difference group sparse representation for rolling bearing fault detection

Abstract Extracting fault feature from bearing vibration signals is vital and challenging because the fault feature is always weak due to noise interference in early stages of failure. This paper proposes a novel adaptive adjacent signal difference group sparse representation (Ada-ASDGSR) for bearing fault detection. Aiming to the time-domain feature of fault signal, the adjacent difference group sparse representation (ASDGSR) is developed firstly. Moreover, a weighted parameter-setup method is developed to set the regularization parameter of ASDGSR adaptively. Furthermore, a majorization–minimization (MM)-based optimization algorithm is derived for solving the proposed ASDGSR, which has the advantage of fewer iteration times and time-saving. Numerical simulation experiment and practical rolling bearing experiment are conducted to verify the performance of Ada-ADGSR in fault detection. The comparison results show that Ada-ADGSR is more effective for feature extraction in comparison with other state-of-the-art methods.