摘要
该文基于音频特征的交流接触器电寿命预测方法,采用快速傅里叶变换法分析交流接触器线圈工作电压异常、触头磨损和壳体松动三种状态下接触器合闸时的音频特征,分别与正常的交流接触器音频特征进行对比。对比结果发现,非正常状态的交流接触器合闸时其音频特征与正常状态时明显不同。为进一步研究接触器合闸时的振动发声机理,基于振动方程和声压波动方程建立接触器合闸时的声场模型,仿真结果表明,该声场模型产生的音频频谱幅值随线圈工作电压的增大而变大、频谱分布随触头形貌改变而变化。分别运用BP神经网络和卷积神经网络(CNN)构建交流接触器合闸音频特征与电寿命的关联模型,对比了L-M算法、拟牛顿法、动量BP法和自适应梯度下降法四种BP网络学习算法,建立电寿命验证平台。实验结果表明,L-M算法性能最优,预测误差小于10%;CNN可以在线学习和提取音频特征,但其预测误差超过20%。
In this paper,the electrical life prediction method of AC contactor based on audio characteristics is studied.The audio characteristics of AC contactor under three states of abnormal coil voltage,contact wear and shell looseness are analyzed by using fast Fourier transform method.Compared with the normal AC contactor,the results show that the audio characteristics of AC contactor in abnormal state are significantly different from that in normal state.In order to further study the vibration and sound generation mechanism of contactor when closing,a sound field model is established based on vibration equation and sound pressure fluctuation equation.The simulation results show that the amplitude of the audio frequency spectrum generated by the acoustic field model increases with the increase of the working voltage of the coil,and the spectrum distribution changes with the change of the contact shape.BP neural network and convolutional neural network(CNN)are used to build the correlation model between AC contactor closing audio characteristics and electrical life.Four BP network learning algorithms,L-M algorithm,quasi Newton method,momentum BP method and adaptive gradient descent method,are compared.The verification results show that the performance of L-M algorithm is the best,and the prediction error is less than 10%;CNN can learn and extract audio features online,but its prediction error is more than 20%.
作者
游颖敏
王景芹
舒亮
倪侃
周新城
You Yingmin;Wang Jingqin;Shu Liang;Ni Kan;Zhou Xincheng(State Key Laboratory for Reliability and Intelligence of Electrical Equipment Hebei University of Technology,Tianjin 300130 China;Yueqing Institute of Industry Wenzhou University,Wenzhou 325035 China)
出处
《电工技术学报》
EI
CSCD
北大核心
2021年第9期1986-1998,共13页
Transactions of China Electrotechnical Society
基金
浙江省公益技术研究计划项目(LGC20E070001)
河北省自然科学基金创新群体项目(E2020202142)
浙江省教育厅项目(Y201737046)资助。
关键词
交流接触器
音频特征
梅尔频率倒谱系数
神经网络
电寿命预测
AC contactor
audio features
Mel frequency cepstrum coefficient
neural network
electric life prediction
