Adaptive Variational Mode Extraction-Based Fault Diagnosis Method for Wind Turbine Bearings Under Strong Nonstationary Regimes

Abstract

The nonstationary operation of wind turbines leads to evolving fault characteristics and spectral leakage in collected vibration signals. These complex operating conditions often mask weaker faults beneath more severe ones, causing existing methods to frequently overlook less prominent fault features. This article introduces a novel adaptive variational mode extraction (AVME) method to tackle this challenge. The key innovation of AVME is its data-driven approach to frequency band segmentation, which significantly reduces interference from both adjacent high-energy signals and nonperiodic signals. It utilizes K-means to segment the autopower spectrum into distinct bands and iteratively adjusts the variational mode extraction (VME) parameters to match the characteristics of each band. Furthermore, this article presents an AVME-based tacholess order tracking (TOT) technique for analyzing fault characteristic signals under nonstationary conditions. Validation of real-world wind turbine data confirms that the AVME method effectively handles complex operating conditions and accurately extracts fault characteristics from raw signals, demonstrating superior efficacy in fault feature extraction compared with spectrum kurtosis (SK), bandwidth-aware adaptive chirp mode decomposition (BA-ACMD), and variational mode decomposition (VMD).