Remaining useful life prediction of wind turbine high speed shaft bearing by sparsity measure based long short term memory network

As a key component of wind turbine drive train, high-speed shaft (HSS) bearing often suffers from major failures. As a result, it is important to predict the remaining useful life (RUL) of wind turbine HSS bearing in advance. However, RUL prediction of HSS bearing often suffers from variable rotating speed of the wind turbine high speed shaft during real life application. Traditional features suitable under constant rotating speed condition are unable to track the degradation of HSS bearing in a consistent manner and thus in turn fail to facilitate efficient prediction of RUL. Considering aforementioned issue, in this research, RUL prediction method of wind turbine HSS bearing is proposed by utilizing the four representative sparsity measures namely kurtosis, gini index, negative entropy and reciprocal smoothness index. Degradation features have been fed into long short term memory (LSTM) network for calculating RUL and associated uncertainties. Effectiveness of the proposed research is validated by degradation data collected from real life 2 MW wind turbine HSS bearing. Experimental results show that the proposed approach enhances the RUL prediction capability of the wind turbine HSS bearing health along with associated uncertainty quantification in compare to traditional measures such as RMS and peak to peak value.