Fault diagnosis of wind turbine bearings based on improved dung beetle optimizer optimized LSTM

Abstract Rolling bearing fault diagnosis of wind turbines is an important part to ensure the safe operation of wind turbines and reduce the operation and maintenance cost, but since the rolling bearing fault signals of wind turbines exhibit strong nonlinearity and non-stationarity, coupled with the influence of the surrounding harsh environment, the shock characteristics associated with bearing faults are often masked by the elevated noise level, which leads to feature extraction becoming difficult. In order to overcome this limitation, an improved dung beetle optimizer (IDBO) is proposed in this paper to optimize Long Short-Term Memory (LSTM) for wind turbine bearing fault diagnosis. Firstly, the Levy flight strategy and T-distribution perturbation strategy are integrated on the basis of the traditional dung beetle algorithm to strengthen the convergence accuracy and local optimization ability; then the IDBO algorithm is combined with the Long Short-Term Memory network to construct the IDBO-LSTM fault diagnosis model, which has the advantage of smaller parameter compared with the traditional LSTM so that the model maintains the accuracy of the traditional temporal feature extraction methods while reduces the overall parametric and arithmetic volume of the module; finally, the open dataset of bearings from Case Western Reserve University is used as an example to verify that the fault diagnosis accuracy of the model can reach 98.3%. The experimental results show that the IDBO-LSTM algorithm can accurately recognize fault features even in the presence of cyclic interference and its accuracy is better than that of other models, which can effectively improve the fault diagnosis accuracy of rolling bearings in wind turbines.