Wang Ruirui, Feng Zhan, Huang Sisi, Fang Xia, Wang Jie
School of Mechanical Engineering Sichuan University, Chengdu 610041, China.
Micromachines (Basel). 2020 Jul 31;11(8):753. doi: 10.3390/mi11080753.
To solve the problem of vibration motor fault detection accuracy and inefficiency in smartphone components, this paper proposes a fault diagnosis method based on the wavelet packet and improves long and short-term memory network. First, the voltage signal of the vibration motor is decomposed by a wavelet packet to reconstruct the signal. Secondly, the reconstructed signal is input into the improved three-layer LSTM network as a feature vector. The memory characteristics of the LSTM network are used to fully learn the time-series fault feature information in the unsteady state signal, and then, the model is used to diagnose the motor fault. Finally, the feasibility of the proposed method is verified through experiments and can be applied to engineering practice. Compared with the existing motor fault diagnosis method, the improved WP-LSTM diagnosis method has a better diagnosis effect and improves fault diagnosis.
为解决智能手机部件中振动电机故障检测精度低和效率不高的问题，本文提出一种基于小波包和改进型长短时记忆网络的故障诊断方法。首先，通过小波包对振动电机的电压信号进行分解以重构信号。其次，将重构后的信号作为特征向量输入到改进的三层长短时记忆网络中。利用长短时记忆网络的记忆特性充分学习非稳态信号中的时序故障特征信息，然后，使用该模型诊断电机故障。最后，通过实验验证了所提方法的可行性，且该方法可应用于工程实践。与现有的电机故障诊断方法相比，改进后的小波包-长短时记忆诊断方法具有更好的诊断效果，提升了故障诊断能力。
