Informed sparsity-based blind filtering in the presence of second-order cyclostationary noise

This study investigates the potential to improve the fault detection capability of sparsity-based blind filtering. It optimizes a finite impulse response filter to maximize the sparsity of the squared envelope spectrum (SES) of vibration signals. However, the method is to be highly prone to fail optimization due to the immense number of non-fault-related second-order cyclostationary interferences. These interferences can skew the sparsity estimation of the SES and thus impair the sparsity-based blind filtering method. Even whitening or signal separation methods can fail to remove such cyclostationary components adequately. Hence, the technique is unlikely to function particularly on the signals measured on complex industrial machines. This paper extends the initial study introducing sparsity-based blind filtering in the literature. It proposes refining the filter optimization’s objective function by targeting narrow frequency bandwidths on the SES to avoid non-fault-related peaks. The narrow bandwidths are selected by exploiting the available engineering knowledge, such as an average shaft speed. The proof of concept is shown on simulated signals, and the performance tests are made on signals measured on an industrial rotating machine and a wind turbine gearbox. The study’s outcome demonstrates that targeting automatically selected narrow bandwidths bounded by shaft speed harmonics can significantly improve the detection capability of sparsity-based blind filtering.