Modeling Non-Linear and Non-Stationary Magnetic Signals: An Enhanced Signal Processing Strategy for Wind Time Series Analysis

Magnetic-flux-based condition monitoring techniques are becoming increasingly popular due to their advantages such as non-invasiveness, low-costs, and ease of sensor installation. However, developing diagnostics and prognostics-based condition monitoring systems become challenging as magnetic signals-based wind time series exhibit non-linear and non-stationary characteristics due to being subjected to diverse combination of dynamic system behavior. Present data-driven feature extraction techniques, as well as knowledge-based and parameter estimator models, fall short in effectively quantifying these characteristics due to increased signal complexity and computation cost. Additionally, methods like the short-time Fourier transform that assumes stationarity of signals and has a fixed window size and discrete wavelet transform that has pre-defined wavelet bases, have proven suboptimal for analyzing non-linear and non-stationary magnetic flux signals due time-frequency resolution tradeoffs. This research bridges the gap by investigating a new methodology of feature extraction to adeptly process the non-linear and non-stationary behavior in magnetic signals-based wind time series. This methodology is unique in nature since it utilizes magnetic signature-based fault condition indicators derived from the wind generator of a drive train model that accounts for the electromagnetic coupling. Additionally, the signal processing technique employed here considers non-linearity and non-stationarity that arises from wind characteristics, gearbox dynamics and grid conditions and its effects on the magnetic flux density of the wind generator. Notably, this methodology offers new insights on motivation, applications, and significance of magnetic-flux-based condition monitoring techniques and highlights its potential as a critical tool for non-invasive fault detection on multiple components of a wind turbine.