Iterative characteristic ridge extraction for bearing fault detection under variable rotational speed conditions

Ridge extraction is an effective tacholess order tracking technique for the fault detection of bearings under time-varying speed conditions. Cost function ridge detection (CFRD) is the most widely used ridge detection method. However, improper bandwidth selection and unreasonable cost function construction significantly restrict the performance of the CFRD. To address the two shortcomings of the CFRD, an improved CFRD (ICFRD) method is firstly proposed in this paper. The ICFRD integrates an adaptive search bandwidth determination technique that varies the search region with signal signatures, as well as a novel cost function that comprehensively considers the trade-off between ridge amplitude and smoothness. An iterative characteristic ridge extraction (ICRE) strategy is then presented based on the ICFRD to extract multiple characteristic ridges in a time-frequency plane automatically. The average frequency ratios between the extracted characteristic ridges are calculated to estimate bearing fault characteristic orders and therefore detect bearing faults. The performance of the proposed method was tested using simulated signals and experimental vibration signals collected from a machinery test rig. Results show that the ICRE outperforms the conventional CFRD in terms of detecting bearing faults under variable speed conditions. The average relative errors between the extracted instantaneous frequencies and the theoretical ones of the ICRE are 0.85%, 2.11% and 0.63% for inner race fault, outer race fault, and healthy bearing vibration signal, respectively. These values are much smaller than the results of using the CFRD. (c) 2021ISA.PublishedbyElsevierLtd.Allrightsreserved.

Automated summary

This paper introduces an improved Cost Function Ridge Detection (ICFRD) method and an Iterative Characteristic Ridge Extraction (ICRE) strategy to estimate bearing fault characteristic orders by calculating the average frequency ratios between characteristic ridges, achieving the detection of bearing faults. The performance of the proposed method outperforms the conventional CFRD in detecting bearing faults under variable speed conditions, with much smaller average relative errors between extracted instantaneous frequencies and theoretical ones.