基于DAE-BiLSTM-CNN的滚动轴承故障诊断方法
Method of fault diagnosis for rolling bearings based on DAE-BiLSTM-CNN
王英杰 1朱景建 2龚智强 1何彦虎1
作者信息
- 1. 湖州职业技术学院智能制造与电梯学院,浙江 湖州 313000
- 2. 上海东海职业技术学院机电学院,上海 200241
- 折叠
摘要
滚动轴承作为机械设备中的核心组件,其运行状态直接影响系统的安全性与可靠性.由于轴承运转过程中的噪声干扰,传统故障诊断方法存在识别不准确、模型泛化有限等不足.为解决此问题,提出了一种基于DAE-BiLSTM-CNN 的滚动轴承故障诊断方法.通过去噪自动编码器(DAE)提高模型去除噪声干扰能力、采用双向长短时记忆网络(BiLSTM)提取轴承运转过程中的时序特征,再采用卷积神经网络(CNN)提取显著特征进行故障判别与分类.采用已公开数据对模型进行训练及超参数优化,并比较了提出的故障诊断模型与现有模型的准确性、精度、召回率及F1分数等性能评价指标.结果表明:相比于现有的故障诊断模型,所提方法具有更高的精度及召回率,验证了该故障诊断模型的准确性及可靠性,同时也说明该诊断方法对于实际工业应用中的滚动轴承故障诊断具备一定的理论参考价值.
Abstract
The rolling bearing is a core component of mechanical equipment,whose operation status directly affects the sys-tem's safety and reliability.The traditional method of fault diagnosis,which suffers inaccurate identification and limited model application,fails to address noise interference when the rolling bearing is in operation.In order to solve this problem,in this arti-cle,with the help of DAE-BiLSTM-CNN,efforts are made to propose a method of fault diagnosis for rolling bearings.The de-noising autoencoder(DAE)is used to improves the model's ability of noise anti-interference;the bidirectional long short-term memory network(BiLSTM)is used to extract the time series features when the rolling bearing is in operation;the convolutional neural network(CNN)is used to extract significant features for fault detection and classification.Some publicly-available data is used to train the model and optimize the hyperparameters.Besides,the model of fault diagnosis is compared with the current mod-el in terms of accuracy,recall rate,F1 score,and other performance indicators.The results show that compared with the current model,the DAE-BiLSTM-CNN model has a higher standard of accuracy and recall rate;it is verified that the model of fault diag-nosis is accurate and reliable.Also,the fault-diagnosis method provides theoretical significance for fault diagnosis of rolling bear-ings in industrial projects.
关键词
滚动轴承/故障诊断/去噪自动编码器/双向长短时记忆网络/卷积神经网络Key words
rolling bearing/fault diagnosis/denoising autoencoder/bidirectional long short-term memory network/convolu-tional neural network引用本文复制引用
出版年
2024
NETL
NSTL
万方数据