Mono-trend mode decomposition for robust feature extraction from vibration signals of rotating Machinery

Abstract

The analysis of vibration signals of rotating machinery plays an important role in equipment fault diagnosis and health monitoring. However, the decomposition of fault characteristic components often lacks robustness and accuracy under nonstationary operating conditions due to potential strong background noise and the complexity of existing components. In this paper, we address this issue by utilizing the relationships among components of vibration signals generated by rotating structures, described as the mono-trend modes (MTMs). Our proposed method, called the MTM decomposition (MTMD), enables robust feature extraction from vibration signals under nonstationary operating conditions. Initially, the instantaneous angular speed (IAS) of the input shaft is measured by tachometers or estimated by the framework of parameterized resampling time–frequency transform. The instantaneous frequencies (IFs) of the targeted components are then determined based on their orders relative to the IAS. Subsequently, the instantaneous amplitudes (IAs) of the targeted components are estimated by solving a joint optimization problem that includes a penalty term designed in consideration of the low variation characteristic of IAs. This penalty term enhances the efficient distribution of energy among close-spaced components. As a result, all targeted MTMs can be extracted simultaneously without any accumulated error. Various simulated experiments and practical experiments on a high-power planetary gearbox are conducted to analyze and validate the effectiveness of the proposed MTMD method in extracting robust features, especially in the presence of strong noise and close-spaced components.