深度置信网络迁移学习的行星齿轮箱故障诊断方法

实际工程中行星齿轮箱受工况、运行情况等因素的影响,获取的数据难以满足训练和测试数据独立同分布且训练数据充足的条件,直接影响故障诊断效果。为此,提出一种深度置信网络(Deep Belief Network,DBN)迁移学习的行星齿轮箱故障诊断方法。首先,将辅助标记数据的原始信号频谱作为DBN网络的输入,逐层更新网络的权重和偏置值对输入信号进行分级表达,以获得其分布式特征表达,得到基于辅助标记样本的DBN预模型。再利用少量的目标标记样本微调DBN预模型的网络权重和偏置值,实现DBN网络的权重和偏置值从源域到目标域的迁移以适应新的目标样本识别,最终提高目标域样本故障识别准确率。通过行星齿轮箱故障模拟实验验证了所提方法的可行性和有效性。

Planetary gearbox fault diagnosis method based on deep belief network transfer learning

Planetary gearbox is affected by many factors,such as,working condition and operational situation in practice.The obtained data are difficult to meet the condition of training and test data being independent and identically distributed and training data being sufficient.All these directly affect planetary gearbox fault diagnosis effect.Here,a planetary gearbox fault diagnosis method based on deep belief network(DBN)transfer learning was proposed.Firstly,the original signal spectrum of the auxiliary marker data was taken as the input of the DBN network.The weight and bias value of the network were updated layer by layer to hierarchically express the input signal and obtain its distributed feature expression,and a DBN pre-model based on the auxiliary marker samples was obtained.Then,a small number of target labeled samples were used to finely tune the network weight and bias value of the DBN pre-model,and realize the migration of the weight and bias value of the DBN network from the source domain to the target domain for the purpose of adapting to the new target sample recognition and finally improving the correctness rate of the target domain sample fault identification.Finally,the feasibility and effectiveness of the proposed method were verified with fault simulation tests of planetary gearbox.