PhD Defense - Adrien MARSICK

Vibration-based damage assessment of rolling element bearings : cage cyclostationarity & trend analysis

This thesis explores the subject of estimating the severity of rolling element bearing faults from vibration signals.
This task, complicated for machines operating under stationary conditions, becomes a real challenge as soon as the systems are operating at variable speeds and loads.
This thesis applies to the monitoring of the shaft line of wind turbines of Engie Green, the industrial partner of this project.
This report explores two themes for making better use of vibration signals to help in maintenance decision-making.

Firstly, conventional signal processing tools are based on the assumption that fault signature exhibits cyclic statistical properties with respect the rotation of the shafts supporting the bearings.
Slippage during operation degrades the capabilities of existing tools. This work proposes to overcome this problem by reinterpreting the cyclostationary properties by changing the rotation of reference, from the shaft to the cage rotation. The causes for this shift are first studied. Armed with techniques for the estimation of the instantaneous rotation of the cage, a method for restoring the cyclostationary properties is proposed. On the basis of this restoration, an adaptation of synchronous averaging techniques to the case of bearings is studied.

Estimation of degradation is based on the ability to finely monitor changes in various indicators. In this respect, there is a serious lack of tools for processing series of vibration signals, most of which focus on the separate analysis of each signal. Rank statistics offer a robust, non-parametric framework for trend analysis.
Based on the Mann-Kendall test, two tools are proposed to address two issues in vibration monitoring: informative frequency band selection and spectral analysis.