基于应力波卷积神经网络的齿轮故障诊断方法  

Fault Diagnosis of Gear Based on Stress Wave Feature and Convolutional Neural Network

作　　者：花海波 于洪亮[1] 闫锦[1] 廖建彬[1] HUA Haibo;YU Hongliang;YAN Jin;LIAO Jianbin(School of Marine Engineering,Jimei University,Xiamen 361021,China)

机构地区：[1]集美大学轮机工程学院,福建厦门361021

出　　处：《集美大学学报（自然科学版）》2023年第4期335-342,共8页Journal of Jimei University：Natural Science

基　　金：福建省自然科学基金项目(2022J01808);福建省科技计划项目(2021H0020,2020H0018)。

摘　　要：针对齿轮箱的故障诊断问题,提出一种应力波特征提取算法,并结合卷积神经网络进行齿轮箱故障诊断。对加速度传感器采集的振动信号进行高通滤波、等间隔重采样、峰值提取,以获取应力波特征信号;使用小波变换处理应力波信号,获得二维时频谱;将该谱图输入到卷积神经网络中,训练获得符合要求的故障诊断模型。基于凯斯西储大学的轴承数据集对该故障诊断算法进行初步有效性验证,以齿轮箱多故障状态下的振动加速度实验数据为基础进行故障特征分析。结果表明,在使用应力波特征提取算法后,卷积神经网络的故障诊断率以及诊断效率均有明显提升,证明了该算法的有效性。Aiming at the problem of gearbox fault diagnosis,a stress wave feature extraction algorithm is proposed and combined with convolutional neural network for gearbox fault diagnosis.Firstly,the vibration signal collected by the accelerometer was subjected to high-pass filtering,equal interval resampling,peak extraction,etc.to obtain the stress wave feature signal,and then the wavelet transform was used to process the stress wave signal to obtain a two-dimensional time frequency spectrum.Finally,the spectrum was used as an input parameter enter ing the convolutional neural network and trained to obtain a fault diagnostic model that meets the requirements.The preliminary effectiveness verification of the fault diagnosis algorithm was conducted based on the bearing dataset of case western reserve university,fault feature analysis was conducted based on experimental data of vibration acceleration signal under multiple fault states of the gearbox.tIt indicates that after using the stress wave feature extraction algorithm,the fault diagnosis rate and diagnosis efficiency of the convolutional neural network can be significantly improved,which proves he effectiveness of the algorithm.

关 键 词：齿轮箱 应力波 故障诊断 卷积神经网络 

分 类 号：TM311[电气工程—电机]