基于集成卷积神经网络的行星齿轮智能诊断方法
Intelligent Diagnosis Method of Planetary Gear Based on Integrated Convolutional Neural Network
黄克康 1武兵 2张志伟1
作者信息
- 1. 太原理工大学机械电子工程研究所,山西 太原 030024
- 2. 太原理工大学机械电子工程研究所,山西 太原 030024;太原理工大学新型传感器与智能控制教育部重点实验室,山西 太原 030024
- 折叠
摘要
针对行星齿轮箱故障振动特征需要预处理、识别困难以及诊断模型收敛速度较慢的问题,提出基于集成卷积神经网络的行星齿轮箱智能故障诊断方法.首先,采用一维卷积对齿轮的原始时域振动信号提取特征,之后通过采用两个弱分类器,根据弱分类学习错误率的性能更新样本权重,调整权重后根据训练集训练弱分类器.重复此过程,最后通过设置策略整合弱分类器,形成集成卷积神经网络;建立一个稳定用于行星齿轮箱的智能故障诊断的模型.实验结果表明:集成卷积神经网络能很好地对行星齿轮原始振动信号进行快速诊断.相对于传统卷积神经网络对齿轮原始时域振动故障信号的诊断具有更强的辨识能力和更快的收敛速度;所建立的智能诊断模型可以有效地诊断齿轮不同的故障状态.
Abstract
Aiming at the problems of planetary gearbox fault vibration characteristics that require preprocessing,identification difficulties,and slower convergence of the diagnostic model,an intelligent fault diagnosis method for planetary gearboxes based on integrated convolutional neural networks is proposed.First,one-dimensional convolution is used to extract features from the original time-domain vibration signal of the gear,and then two weak classifiers are used to update the sample weights according to the performance of the weak classification learning error rate,and the weak classifiers are trained according to the training set after adjusting the weights.Repeat this process,and finally integrate the weak classifiers by setting strategies to form an inte-grated convolutional neural network;establish a stable model for intelligent fault diagnosis of planetary gearboxes.Experimental results show that the integrated convolutional neural network can quickly diagnose the original vibration signals of planetary gears.Compared with the traditional convolutional neural network,it has stronger identification ability and faster convergence speed in the diagnosis of the original time domain vibration fault signal of the gear;the established intelligent diagnosis model can effectively diagnose the different fault states of the gear.
关键词
集成学习/卷积神经网络/原始振动信号/智能诊断Key words
Ensemble Learning/Convolutional Neural Network/Original Vibration Signal/Intelligent Diagnosis引用本文复制引用
基金项目
山西省科技重大专项项目(20181102027)
出版年
2024
NETL
NSTL
维普
万方数据