Adaptive high-resolution order spectrum for complex signal analysis of rotating machinery: Principle and applications

Abstract

Order spectrum analysis is commonly used in rotating machinery vibration signal analysis under nonstationary conditions, because of its capability in detecting the rotating frequency and harmonics. Nevertheless, the conventional order spectrum suffers from some inherent drawbacks. For example, its application is limited by the inconvenience of rotary encoder or tachometer installation on running machines. Meanwhile, it is subjected to disturbances by speed-irrelative components, such as resonances and background noise, and angular resampling errors as well. More importantly, it is obtained by Fourier transforming the angular resampled signal, but the amplitude envelope after angular resampling is still nonstationary under time-varying conditions, leading to spectral blur of frequency orders and low resolution. To address the above issues, we propose an adaptive high-resolution order spectrum method, by exploiting the capability of multiple order probabilities approach (MOPA) in estimating rotating speed and that of adaptive iterative generalized demodulation (AIGD) in identifying true signal components of interest respectively. Firstly, the MOPA is utilized to extract the rotating speed from vibration signals, instead of using rotary encoders or tachometers. Then, the AIGD is adopted to identify true speed-related mono-components adaptively, thus eliminating interferences by speed-irrelative components. Finally, the order spectrum is constructed based on the average amplitude envelope and frequency order of each true component, instead of through angular resampling and Fourier transform. By doing so, the proposed order spectrum can adaptively identify rotating frequency orders in high resolution. The adaptability lies in that it does not need rotary encoders or tachometers, but adaptively estimates speed and identify rotating frequency harmonics, and therefore is more convenient in applications. The high resolution lies in that the energy highly concentrates at true frequency order, because it does not involve Fourier transform and angular resampling, thus being free of spectral blur, interferences by speed-irrelative components and resampling errors. The proposed method is illustrated and validated by analyses of numerical simulation and typical rotating machinery signals, including lab experiment and in-situ measurement of planetary gearboxes, and on-site measurement of a hydraulic turbine.