Self-running bearing diagnosis based on scalar indicator using fast order frequency spectral coherence

Abstract

In vibration-based diagnosis of rolling element bearings, the complexity of the signals requires an expert to use advanced signal processing tools and to interpret the results based on his/her experience. Recently, a few autonomous methods have been proposed to alleviate the demand on the user’s expertise, yet they have been mainly focused on fault detection. This paper follows a similar direction but with wider objectives: it aims to develop an indicator that is able to detect, identify and classify typical faults on rolling elements, inner and outer-race. The indicator is based on the recently developed Fast Order-Frequency Spectral Coherence, a key tool of the theory of second-order cyclostationary processes: it condenses the whole information initially displayed in three dimensions into a scalar and provides an interpretation in terms of a probability of presence of a fault. In addition, the proposed indicator is able to return information for different levels of damages in both stationary and non-stationary operating conditions. It takes into consideration uncertainties in the bearing characteristic frequencies, which is crucial in bearing diagnosis. A new pre-processing step is provided to ensure an efficient and constant statistical threshold. The proposed indicator is intended to be used in an autonomous process without the need for visual analysis and human interpretation. The proposed indicator is compared with a recent indicator based on the Envelop Spectrum, in terms of classification and detection performance. Several applications using real and simulated data eventually illustrate the capability for self-running diagnosis.