Stationary and non-stationary random vibration modelling and analysis for an operating wind turbine

• The output-only identification of an operating wind turbine dynamics is performed.
• Stationary/non-stationary methods employed and compared.
• TAR modeling used for the first time for wind turbines and enables TV identification.
• Cyclo-stationary behaviour due to blade rotation is found on the WT modes.
• Non-stationary behaviour due to wind strongly affects WT vibrations.

The problem of stationary and non-stationary random vibration modelling and analysis for an operating wind turbine is considered by using an acceleration vibration signal measured on a NegMicon NM52/900 fixed speed wind a turbine tower. Three stationary methods, namely Welch spectral estimation, autoregressive, and autoregressive moving average modelling, are considered, along with five non-stationary methods, namely non-parametric Wigner–Ville and spectral correlation, and parametric modelling by means of smoothness priors (SPs) time-dependent autoregressive modelling, functional series (FS) time-dependent autoregressive modelling, and adaptable functional series (AFS) time-dependent autoregressive modelling. The need for non-stationary modelling is mainly dictated by blade rotation and the wind dynamics. The results of the study confirm the cyclo-stationary and broader non-stationary nature of the random vibration, dictating the need for corresponding methods for accurate modelling and analysis. The capabilities and facets of the various methods in terms of modelling and analysing operating wind turbine random vibration are presented.