基于深度自编码网络模型的风电机组齿轮箱故障检测

为了实现风机齿轮箱的故障检测分析,提出一种基于风电机组齿轮箱的数据采集与监视控制(SCADA)数据和振动信号的深度自编码网络模型。该模型作为一种典型的深度学习方法,通过逐层智能学习初始样本特征,可以获取数据蕴含的规则与分布特征形成更加抽象的高层表示。首先,利用限制性玻尔兹曼机对网络参数进行预训练和反向传播算法对参数进行调优,建立深度自编码网络模型。然后,通过对齿轮箱的状态变量进行编码和解码,计算重构误差并将其作为齿轮箱的状态检测量。为了有效检测重构误差的趋势变化,选用自适应阈值作为风机齿轮箱故障检测的决策准则。最后,利用对齿轮箱故障前、后记录的数据进行仿真分析,结果验证了深度自编码网络学习方法对齿轮箱故障检测的有效性。

Fault Detection of Wind Turbine Gearbox Based on Deep Autoencoder Network

In order to achieve the fault detection of wind turbine gearbox,a deep autoencoder network model from deep learning method based on supervisory control and data acquisition (SCADA) data and vibration signals of wind turbine gearbox is proposed in this paper.The deep autoencoder network,as one of the typical deep learning methods,can obtain the underlying rules and distribution characteristics of the data through learning features of original sample by layer-wise intelligent learning to form a more abstract and high-level representation.Firstly,restricted boltzmann machine was used to pre-train parameters and the back-propagation algorithm was used to optimize these parameters to build the deep autoencoder model in this paper.Then through encoding and decoding condition variables of gearbox,reconstruction error was computed as the gearbox condition monitoring variable.In order to monitor the trend change of reconstruction error effectively,the adaptive threshold was chosen as the decision criterion of gearbox fault.Finally,by utilizing the record data before and after fault to simulation,results showed the validity of deep autoencoder model on gearbox fault detection.