摘要
针对轧机齿轮箱结构复杂、故障信号识别困难、故障部位分类不清等难题,提出了一种基于多通道二维卷积神经网络(MC2DCNN)与长短期记忆神经网络(LSTM)特征融合的故障诊断方法。首先,设计了一种三通道混合编码的二维样本结构,以达到故障识别与分类目的,对齿轮箱典型故障进行了自适应分类;其次,该模型将齿轮箱的垂直、水平和轴向三个方向的振动信号融合构造输入样本,结合了二维卷积神经网络与长短时记忆神经网络的优势,设计了与之对应的二维卷积神经网络结构,其相较于传统的单通道信号包含了更多的状态信息;最后,分析了轧制过程数据和已有实验数据,对齿轮故障和齿轮箱全故障进行了特征识别和分类,验证了该模型的准确率。研究结果表明:模型对齿轮箱齿面磨损、齿根裂纹、断齿以及齿面点蚀等典型故障识别的平均准确率达到95.9%,最高准确率为98.6%;相较于单通道信号,多通道信号混合编码方式构造的分类样本极大地提升了神经网络分类的准确性,解调出了更丰富的故障信息。根据轧制过程中的运行数据和实验台数据,验证了该智能诊断方法较传统方法在分类和识别准确率上更具优势,为该方法的工程应用提供了理论基础。
Aiming at the complex structure of the rolling mill gearbox,the challenges in fault signal identification,and the lack of clarity in fault location classification,a fault diagnosis method based on the fusion of features from a multi-channel two-dimensional convolutional neural network(MC2DCNN) and a long short-term memory neural network(LSTM) was proposed.Firstly,a three-channel hybrid coding two-dimensional sample structure was designed to identify and classify faults,the adaptive classification function for typical gearbox faults was realized by it.Secondly,the vertical,horizontal,and axial vibration signals of the gearbox were fused by the model to construct the input sample.The advantages of the two-dimensional convolutional neural network and the short-term memory neural network were combined to design a two-dimensional convolutional neural network structure that encapsulated more state information than traditional single-channel signal processing.Finally,both the rolling process data and existing experimental data were utilized to identify and classify the characteristics of gear and gearbox faults,verifying the model's accuracy.The research results demonstrate that the model can accurately identify the typical faults of gearbox tooth surface wear,root crack,broken tooth,and pitting tooth surface,achieving an accuracy of 95.9%,with the highest accuracy rate reaching 98.6%.In comparison to single-channel signals,the classification samples constructed through multi-channel signal hybrid coding significantly enhance the classification accuracy of the neural network and demodulate more abundant fault information.A vast amount of rolling process and experimental bench data verify that the proposed intelligent diagnosis method outperforms traditional methods in terms of classification and recognition accuracy,and provides a theoretical basis for engineering applications.
作者
陈蓉
王磊
CHEN Rong;WANG Lei(Electromechanical Information Department,Huaiyin Commercial School,Huai an 223002,China;School of Mechanical Engineering,Southeast University,Nanjing 210096,China)
出处
《机电工程》
北大核心
2025年第2期287-297,共11页
Journal of Mechanical & Electrical Engineering
基金
国家自然科学基金资助项目(52075095)。
关键词
高精度轧机齿轮箱
智能故障诊断
多通道二维卷积神经网络
长短期记忆神经网络
数据分类
high precision rolling mill gearbox
intelligent fault diagnosis
multi-channel two-dimensional convolutional neural network(MC2DCNN)
long short-term memory neural network(LSTM)
data classification
