基于长短时记忆神经网络的风电机组滚动轴承故障诊断方法

A Method of Fault Diagnosis for Rolling Bearing of Wind Turbines Based on Long Short-term Memory Neural Network

摘要:

风能作为一种绿色能源在我国能源结构中发挥着越来越重要的作用;风电机组的滚动轴承作为传动系统的重要组成部分,是其主要故障部件之一;随着风电规模的不断增长,及时地发现风电机组滚动轴承的故障对风电场安全稳定运行具有重要意义;针对传统回归神经网络存在的梯度消失问题,提出了利用长短时记忆神经网络对风电机组滚动轴承进行故障诊断的模型;首先,利用小波包变换对风电机组滚动轴承振动信号进行处理,提取其特征向量,将其作为长短时神经网络的输入,从而诊断出风电机组滚动轴承的3种常见故障;通过算例分析,结果表明所提出的方法能够有效地对风电机组的滚动轴承进行故障诊断,并且在故障特征量差异不明显的情况下长短时记忆神经网络仍具有良好的故障诊断性能,说明了该方法的可行性和有效性。

Abstract:

Wind power as a green energy is playing an increasingly important role in China's energy structure. Rolling bearing as a main component of wind turbine drive system is one of the main defective parts. Timely detection of the faults of wind turbines’ rolling bearings is significant to the safe and stable operation of the wind farms. Aimed at the problem of gradient disappearing of the traditional recurrent neural network, the model is based on Long Short-Term Memory(LSTM) neural network to detect faults of wind turbines’ rolling bearings. First, to extract the feature vectors, wavelet packet transforming is used to process the vibration signals of wind turbines’ rolling bearings. Then the feature vectors are used as the input of LSTM neural networks, so as to diagnose the three common faults of wind turbines’ rolling bearings. Finally, through case studies, the validity of this method is verified. Even when the differences of faults’ features are not obvious, this method is still able to get good diagnostic results, which demonstrates the effectiveness of this method.