基于迁移学习的异步电机故障诊断

Research on fault diagnosis algorithm of asynchronous induction motor based on transfer learning

针对异步电机故障诊断中,故障数据样本少导致传统深度神经网络模型泛化能力差的问题,提出一种异构迁移学习的异步电机故障诊断算法。首先,通过仿真平台模拟异步电机故障,以解决故障数据样本少的问题;其次,对正常和故障状态下的电流电压信号进行小波变换,作为深度学习网络的输入;然后,基于多核最大平均差异方法,获得仿真数据和实测数据的深度特征差异,对深度学习神经网络参数微调,使其深度学习特征具有跨域不变性。最终,在实验平台上验证文中所提算法,实验结果表明,该算法的故障诊断准确率高,依赖实测故障数据样本少。

In the fault diagnosis of asynchronous induction motor,a fault diagnosis algorithm for asynchronous induction motor based on heterogeneous migration learning is presented,to solve the problem of poor generalization ability of traditional deep neural network model,due to the small number of fault data samples.Firstly,the fault of asynchronous induction motor is simulated to solve the problem of fewer fault data samples.Secondly,the current and voltage signals in normal and failure state are transformed by wavelet transformation as input of deep learning network.Then,based on the multicore maximum average difference method,the difference of depth characteristics between simulated and measured data is obtained,and the parameters of the deep learning neural network are fine-tuned to make its deep learning characteristics cross-domain invariant.Finally,the proposed algorithm is validated on the experimental platform.The results show that the algorithm has high accuracy in fault diagnosis and fewer samples depending on the measured fault data.