基于改进辅助分类生成对抗网络的风机主轴承故障诊断

基于振动信号的风电机组故障诊断方法是风电安全运维领域研究的重点之一。风电机组主轴承较少发生故障,给运用数据挖掘方法判断故障类型带来很大困难。针对该问题,文中提出了一种用于风电机组主轴承故障诊断的数据增强方法。通过对辅助分类生成对抗网络(ACGAN)的适应性进行改进,引入梯度惩罚,构建了改进ACGAN框架,以提高其学习稳定性;在判别器网络中引入池化层,以提升其在多分类场景下提取数据特征的能力。仿真结果表明,所提出的改进ACGAN框架能够实现对原始数据分布特征的有效学习,抗噪声干扰性强,相对于原框架训练过程更稳定,生成数据的质量更高;能够有效平衡风电机组主轴承故障振动数据,进一步提升了风电机组主轴承故障诊断的正确率。

Fault Diagnosis of Main Bearing of Wind Turbine Based on Improved Auxiliary Classifier Generative Adversarial Network

The fault diagnosis method of wind turbines based on vibration signals is one of the focuses in the field of safe operation and maintenance of wind power units. Faults in the main bearing of the wind turbine rarely happen, which makes it difficult to use data mining methods to determine the fault type. To solve this problem, a data enhancement method for fault diagnosis of wind turbine main bearings is proposed. By improving the adaptability of auxiliary classifier generative adversarial network (ACGAN) and introducing gradient penalty, an improved ACGAN framework is constructed to improve its learning stability. And a pooling layer is introduced into the discriminator network to enhance its ability to extract data features in multiple classification scenarios. The simulation results show that the improved ACGAN framework can effectively learn the distribution characteristics of the original data, has strong anti-noise interference, is more stable than the original framework in the training process, and has higher quality of generated data; it can effectively balance the fault vibration data of the main bearing of the wind turbine, and further improves the accuracy of fault diagnosis for the main bearing of the wind turbine.