基于波束形成及CNN-LSTM的托辊故障距离估计模型  

Roller Fault Distance Estimation Model Based on Beamforming and CNN-LSTM

作　　者：张雄[1,2] 武文博 李嘉禄 董帆 万书亭[1,2] ZHANG Xiong;WU Wenbo;LI Jialu;DONG Fan;WAN Shuting(Key Laboratory of Health Maintenance and Failure Prevention of Electric Machinery Equipment in Hebei Province,North China Electric Power University,Baoding 071003,Hebei,China;Department of Mechanical Engineering,North China Electric Power University,Baoding 071003,Hebei,China)

机构地区：[1]华北电力大学河北省电力机械装备健康维护与失效预防重点实验室,河北保定071003 [2]华北电力大学机械工程系,河北保定071003

出　　处：《噪声与振动控制》2024年第5期114-119,共6页Noise and Vibration Control

基　　金：国家自然科学基金资助项目(52105098);河北省自然科学基金资助项目(E2021502038)。

摘　　要：托辊是带式输送机的关键组件,也是输送系统的薄弱环节,波束形成算法的定向性和深度学习算法的特征识别能力为托辊故障距离估计提供可能。针对输送机输送距离长、不易检测等问题,提出一种利用波束形成(Beamforming)及时空网络(Convolutional Neural Network-Long Short-Term Memory,CNN-LSTM)对托辊故障声源实现距离估计的方法(Beamforming-Convolutional neural network-Long short-term memory,BCL)。首先利用麦克风阵列采集不同距离的故障声源数据,经波束形成定位处理构建数据集;通过卷积神经网络(CNN)获得数据采样集的空间特性,再借助LSTM网络在序列上的建模功能,将由空间数据组成的序列信息输入LSTM网络,从而获得空间时序信息;最后,再将由LSTM网络产生的空间时序信息输入到Softmax分类器,实现故障距离估计。实验结果表明,BCL模型在有无噪声干扰的环境下都可以以高准确率实现托辊的故障距离估计且较其他模型拥有更好的识别能力。Idlers are the key components of belt-conveying operation as well as the weak links in conveying systems.The directionality of beamforming algorithm and the feature recognition ability of deep learning algorithm provide the possibility for estimating the fault distance of idlers.In view of the issues of long conveying distances and the difficulty of detection,a method based on beamforming and Convolutional Neural Network-Long Short term Memory(CNN-LSTM)is proposed to estimate the distance of the fault sound sources of the rollers.Firstly,the Microphone array is used to collect fault sound source data for different distances,and the data set is constructed through beamforming positioning processing.The spatial characteristics of the data sampling set are obtained through convolutional neural network(CNN).Then the sequence information composed of spatial data is input into the LSTM network with the help of the modeling function of LSTM network in the sequence,so as to obtain the spatial timing information.Finally,the spatial temporal information generated by the LSTM network is input into the Softmax classifier to achieve fault distance estimation.The experimental results show that the Beamforming CNN-LSTM(BCL)model can achieve high accuracy in estimating the fault distance of idlers in the environments with or without noise interference,and has better recognition ability than other models.

关 键 词：故障诊断 托辊 波束形成 BCL 距离估计 

分 类 号：TH17[机械工程—机械制造及自动化] TD562[矿业工程—矿山机电]