基于神经网络的电气机械设备传动齿轮故障信号自动化检测方法  

Automated detection method for transmission gear fault signals in electrical and mechanical equipment based on neural networks

作　　者：蔡晨 韩俊 陈超 张藏龙 方亮 高玉芹 CAI Chen;HAN Jun;CHEN Chao;ZHANG Canglong;FANG Liang;GAO Yuqin(Southeast University,Nanjing 210000,China;State Grid Anhui Electric Power Co.,Ltd.,Wuhu Power Supply Company,Wuhu 241000,China)

机构地区：[1]东南大学,江苏南京210000 [2]国网安徽省电力有限公司芜湖供电公司,安徽芜湖241000

出　　处：《电子设计工程》2025年第4期57-61,共5页Electronic Design Engineering

摘　　要：基于提高电气机械设备传动齿轮可靠性和安全性、降低故障率、提升生产效率的目的,提出一种基于神经网络的电气机械设备传动齿轮故障信号自动化检测方法。该方法通过振动传感器采集齿轮运行信号,利用集合经验模式分解算法提取故障信号分量,并通过Hilbert变换转化为包络谱以获取故障特征。进一步计算最大熵值筛选故障信号特征,并将其作为输入引入注意力机制的卷积神经网络(AMCNN)模型中进行训练,输出故障信号自动化检测结果。通过大量实验,得出该方法能够精准地捕捉故障信号特征,有效监测电气机械设备传动齿轮的点蚀、断齿及磨损故障,并实时追踪故障状态变化。在检测超过1200条故障信号时,模型交叉熵损失值为0,这充分证明了该方法的高检测精准性和强应用性。To improve the reliability and safety of transmission gears in electrical and mechanical equipment,reduce failure rates,and improve production efficiency,a neural network-based automated detection method for transmission gear fault signals in electrical and mechanical equipment is proposed.This method collects gear operation signals through vibration sensors,extracts fault signal components using ensemble empirical mode decomposition algorithm,and converts them into envelope spectra through Hilbert transform to obtain fault features.Further calculate the maximum entropy value to screen fault signal features,and use it as input to train the Convolutional Neural Network(AMCNN)model with Attention Mechanism,outputting automated fault signal detection results.The experiment shows that this method can accurately capture the characteristics of fault signals,effectively monitor the pitting,tooth breakage,and wear faults of transmission gears in electrical and mechanical equipment,and track the changes in fault status in real time.When detecting more than 1200 fault signals,the cross entropy loss value of the model is 0,proving its high detection accuracy and strong applicability.

关 键 词：卷积神经网络 注意力机制 电气机械设备 传动齿轮 故障信号 自动化检测 

分 类 号：TN957.51[电子电信—信号与信息处理]