An adaptive noise-cancellation method for detecting generalized roughness bearing faults under dynamic load conditions

This paper proposes an adaptive noise-cancellation (ANC) method for detecting incipient bearing faults, i.e., generalized roughness, with a special focus on dynamic motor operations, including variable-load and variable-frequency conditions. The correlated frequency components in the motor stator current are treated as “noise” to the bearing fault detection and estimated using an adaptive Wiener filter. These estimated noise components are then subtracted from the original motor current and the remaining noise-cancelled current is normalized as the bearing fault index. The Wiener filter parameters are updated for each algorithm iteration, making the calculation of the wideband noise in the motor current (i.e. the bearing fault signatures), and thus adapt to changes of load and frequency, i.e. are independent of load and frequency variations. The feasibility of the proposed method has been primarily validated from experimental results of multiple bearings for different operating conditions. The operating conditions tested in this paper include constant load/frequency, variable load and variable frequency.