声振信号分离提纯的风力发电机故障诊断方法研究

Research on Wind Turbine Fault Diagnosis Methodby Acoustic Vibration Signal Separation and Purification

由于风力发电机振动信号中掺杂大量声音信号,信号频带受到噪声干扰,致使风力发电机故障诊断效果不佳。为此,提出基于声振信号分离提纯的风力发电机在线故障诊断方法。分析导致风力发电机组声振信号干扰的因素。采用经验模态,将初始风力发电机在线故障信号分解成高频分量与低频分量两部分。采用小波阈值算法去除高频分量的噪声,并与低频分量作重构处理,得出风力发电机的实际故障振动信号。建立反向传播(BP)神经网络训练模型,使用粒子群优化算法得出该网络最佳的权值矩阵与偏置向量,并凭借网络训练算法诊断出故障数据的位置与类型。试验结果表明,所提方法在线故障诊断精度高,并能保证其训练效率与收敛速度,具有较高的实用价值。

The wind turbine vibration signal is mixed with many sound signals, and the signal band is disturbed by noise, resulting in poor fault diagnosis of wind turbine. To this end, an online fault diagnosis method of wind turbine based on the separation and purification of sound vibration signal is proposed. The factors leading to the interference of wind turbine acoustic vibration signal are analyzed. Empirical mode is used to decompose the initial wind turbine online fault signal into two parts: high-frequency component and low-frequency component. A wavelet threshold algorithm is used to remove the high-frequency component noise and do reconstruction processing with the low-frequency component to derive the actual fault vibration signal of the wind turbine. A back propagation (BP) neural network training model is established, and the optimal weight matrix and bias vector of the network are derived using particle swarm optimization algorithm, and the location and type of fault data are diagnosed by the network training algorithm. The experimental results show that the proposed method has high accuracy in online fault diagnosis and can ensure its training efficiency and convergence speed, which is of high practical value.