基于改进卷积深度信念网络的风电机组行星齿轮箱故障诊断方法

Fault diagnosis method of planetary gearbox based on improved convolutional deep belief network

中文关键词: 卷积深度信念网络  混沌粒子群算法  超参数  故障诊断

英文关键词: convolutional deep belief networks  chaos particle swarm optimization  hyper parameters  fault diagnosis

中文摘要:

      风力发电机组行星齿轮箱振动信号是一种非线性非平稳的复杂信号，传统的故障诊断方法面对此类信号时，能够很好地处理的范围有限。建立了卷积深度信念网络用于行星齿轮箱故障诊断，为了防止超参数选择有误造成识别的准确率不够，引入粒子群算法对网络的超参数进行优化，对粒子进行混沌初始化提高了粒子的全局搜索能力。首先将原始信号进行变分模态分解提取出冲击信息比较集中的本征模态函数作为网络的输入数据。然后利用训练集进行训练，将混沌粒子群算法根据适应度函数最小确定网络的超参数，利用逐层贪婪算法不断向前更新网络参数。最后将提取的故障特征经过分类器进行分类。经过验证，此方法具有较高的齿轮箱故障识别能力。

英文摘要:

      The vibration signal of planetary gearbox of wind turbine is a kind of non-linear and non-stationary complex signal.The traditional fault diagnosis method can deal with this kind of signal well in a limited range. The convolution depth belief network is established for planetary gearbox fault diagnosis. In order to prevent the wrong selection of hyper parameters from causing insufficient recognition accuracy, particle swarm optimization algorithm is introduced to optimize the hyper parameters of the network, and the chaos initialization of particles improves the global search ability of particles. Firstly, the original signal is decomposed by VMD to extract the eigenmode function which is relatively concentrated in the impact information as the input data of the network. Then, the training set is used to train, the chaos particle swarm optimization algorithmis used to determine the hyper parameters of the network according to the minimum fitness function, and the layer-by-layer greedy algorithm is used to continuously update the network parameters. Finally, the extracted fault features are classified by a classifier. This method is verified that it can diagnose the fault of planetary gearbox under different conditions.