基于短序列时间卷积网络的电梯故障诊断方法

打开文本图片集

关键词： 电梯故障诊断； 振动信号； 短序列； 时间卷积网络； 特征提取； 样本序列长度； 神经常微分方程中图分类号：TN99.5⁃34；TP39.9；TU857 文献标识码： A 文章编号：004⁃373X（05）⁃037⁃

Abstract：Elevator malfunctions may lead to loss of resources and property， and even cause serious safety problems. The fault diagnosis technology can be used to detect and diagnose faults in a timely manner， preventing the continued operation of damaged equipment. Neural networks models have good performance in fault diagnosis of vibration signals， but there are still shortcomings in real⁃timely. A fault diagnosis method based on self⁃attention temporal convolutional network and neural ordinary differential equation （SA ⁃ TCNODE） is proposed to improve the accuracy of vibration signal fault diagnosis and input sequence length. Features can be extracted from short vibration signal sequences， and local fault features can be extracted quickly by adding gated convolution to the temporal convolutional network （TCN） and introducing neural ordinary differential equations （NODE） to build a deeper network. A self ⁃ attention （SA） mechanism is introduced to empower the model with global feature extraction capability， so as to improve the diagnostic accuracy. The experimental results show that SA⁃TCNODE can realize fault diagnosis accuracy of 97.2% with a sample sequence length of 150. In comparison with other methods， this algorithm can also detect faults in a short period of time after they occur， with good diagnostic accuracy and reliability. It can provide important knowledge sharing and reuse methods for elevator fault diagnosis and other similar application fields.

Keywords： elevator fault diagnosis; vibration signal; self ⁃ attention; temporal convolutional network; feature extraction; sample sequence length; neural ordinary differential equation

0 引 言

电梯是人们生活中最常用的特种设备，其运行状态直接关系到建筑的功能性和用户安全。（剩余15064字）