Adaptive time-frequency fusion clustering extracting transform: A useful tool for characterizing time-varying and impulsive features with application to wind turbine fault diagnosis

Abstract

Time-varying and impulsive features play a crucial role in mechanical vibration analysis, as they often reflect the health status of mechanical equipment. However, most existing time-frequency (TF) analysis (TFA) techniques are constructed based on the frequency or time rearrangement schemes, limiting their performance in displaying time-varying and impulsive features simultaneously. Besides, noise is an unignorable factor, which conceals key feature information, hindering the accurate diagnosis of the equipment. To address these issues, this paper proposes a TFA technique called adaptive time-frequency fusion clustering extracting transform (ATFCET). Within the theoretical framework of the ATFCET, an adaptive TF fusion clustering (ATFC) strategy is first introduced to eliminate the noise and highlight the feature components in the TF spectrum. Then, a Rényi entropy-based optimal feature selection criterion is designed to obtain the optimal features to realize a reliable representation of time-varying and impulsive features. Finally, an impulse amplitude accumulation maximum slice (IAAMS)-based diagnostic scheme is constructed, it omits component extraction and reconstruction and directly locates fault impacts with the frequency slice of maximum accumulative amplitude of the impulsive features of ATFCET, thereby enabling fault diagnosis. To validate the effectiveness of ATFCET, bat echolocation and bearing vibration signals from a wind turbine are used as two practical examples. Through numerical and experiments with real-world data, the ATFCET is proven effective.