First-principle and data-driven model- based approach in rotating machinery failure mode detection

Purpose: A major concern of modern diagnostics is the use of vibration or acoustic signals generated by a machine to reveal its operating conditions. This paper presents a method which allows to periodically obtain estimates of model eigenvalues represented by complex numbers. The method is intended to diagnose rotating machinery under transient conditions. Design/methodology/approach: The method uses a parametric data-driven model, the parameters of which are estimated using operational data. Findings: Experimental results were obtained with the use of a laboratory single-disc rotor system equipped with both sliding and hydrodynamic bearings. The test rig used allows measurements of data under normal, or reference, and malfunctioning operation, including oil instabilities, rub, looseness and unbalance, to be collected. Research limitations/implications: Numerical and experimental studies performed in order to validate the method are presented in the paper. Moreover, literature and industrial case studies are analyzed to better understand vibration modes of the rotor under abnormal operating conditions. Practical implications: A model of the test rig has been developed to verify the method proposed herein and to understand the results of the experiments. Hardware realization of the proposed method was implemented as a standalone operating module developed using the Texas Instruments TMS3200LF2407 Starter Kit. Originality/value: The parametric approach was proposed instead of nonparametric one towards diagnosing of rotating machinery.