MCSA-Based Fault Indicator and a Log Likelihood Algorithm to Estimate the Severity of Bearing Faults in Induction Machine

In this paper, a novel methodology for the quantitative diagnosis of bearings is proposed using measurements. The signature have a low dominance in the current spectrum as observed in the measurements. This demands a newer methodology to quantify the signature and correlate it to the severity of the bearing fault. The signal processing technique is developed which quantifies the sensitive frequency band in the spectrum using conventional parameters. A new parameter, named modified sparsity, is introduced which reduces the impact of higher noise levels in the frequency domain. This parameter evaluates the gain factor based on the strength of the peak occurrence and is used to correct the bias induced by the noise floor. Then, a control chart is applied to reduce the variability of the response, revealing the underlying trend. Finally, a data-based likelihood and log-likelihood approach is used to map the health indicator to a more physical parameter (displacement associated with one of the faulty bearings) so that the outcome of the health indicator is more meaningful to the user. The overall idea is to prevent interference from other sources towards the signatures while quantification and give more physical meaning to the severity based on the measurements.