基于卷积神经网络的耐久测试台振动故障诊断方法  

Durable Test Bench Vibration Fault Diagnosis Method Based on Convolutional Neural Networks

作　　者：刘新刚 何永义 胡昱晟 张国旗 

机构地区：[1]上海大学机电工程与自动化学院,上海市智能制造及机器人重点实验室,上海200444

出　　处：《工业控制计算机》2024年第12期76-79,共4页Industrial Control Computer

摘　　要：为了保证高速旋转设备的稳定运行,准确快速地诊断出振动故障类型是其关键。目前常见的振动故障诊断方式主要是信号特征提取进行故障诊断,但这类方法有其局限性。提出一种基于信号处理和图像识别的振动故障诊断方法,首先对采集的振动信号进行降噪,去除噪声信号,其次通过小波变换得到其小波变换图谱,将得到的不同类型故障变换图谱进行神经网络训练,输出故障诊断模型文件。实验结果表明,该训练模型在振动故障诊断准确率和识别速度等多个参数上表现较好,识别准确率达到98%,相较于传统的故障诊断方法具有一定的优势,该研究可实现耐久测试台振动故障的快速诊断。Currently,common methods for vibration fault diagnosis involve signal feature extraction and the application of relevant signal processing algorithms.However,these methods have high hardware requirements and limitations associated with different algorithms.To address vibration fault diagnosis,this study proposes a signal processing and image recognition-based approach.Firstly,the collected vibration signals are denoised to eliminate noise.Then,the wavelet transform is employed to obtain the wavelet transform spectrogram.The obtained spectrograms of different fault types are used for neural network training,resulting in the generation of a fault diagnosis model file.Experimental results demonstrate that the trained model performs well in terms of vibration fault diagnosis accuracy and recognition speed,achieving an accuracy rate of 98%.Compared to traditional fault diagnosis algorithms,this research offers certain advantages.

关 键 词：机械故障诊断 信号处理 小波变换 卷积神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP277[自动化与计算机技术—控制科学与工程] TH17[机械工程—机械制造及自动化]