Data-driven technique for interpreting wind turbine condition monitoring signals

Increasing deployment of large wind turbines (WT) offshore and in remote areas requires reliable condition monitoring (CM) techniques to guarantee the high availability of these WTs and economic return. To meet this need, much effort has been expended to improve the capability of analysing the WT CM signals. However, a fully satisfactory technique has not been achieved today. One of the major reasons is that the developed techniques still cannot provide accurate interpretation of the WT CM signals, which are usually non-linear and non-stationary in nature because of the constantly varying loads and non-linear operations of the turbines. To deal with this issue, a new data-driven signal processing technique is developed in this study based on the concepts of intrinsic time-scale decomposition (ITD) and energy operator separation algorithm (EOSA). The advantages of the proposed technique over the traditional data-driven techniques have been demonstrated and validated experimentally. It has been shown that in comparison with the Hilbert-Huang transform the combination of ITD and EOSA provided more accurate and explicit presentations of the instantaneous information of the signals tested. Thus, it provides a much improved offline tool for accurately interpreting WT CM signals.