张传栋,何存富*,刘秀成,吴斌,张秀.基于BP神经网络的钢轴表面硬度磁巴克豪森噪声定量检测技术[J].实验力学,2020,35(1):1~8 |
基于BP神经网络的钢轴表面硬度磁巴克豪森噪声定量检测技术 |
Magnetic barkhausen noise technology for surface hardness evaluation in steel shaft based on BP neural network |
投稿时间:2019-06-04 修订日期:2019-07-19 |
DOI:10.7520/1001-4888-19-097 |
中文关键词: BP神经网络 钢轴 表面硬度 磁巴克豪森噪声 无损检测 |
英文关键词:BP neural network steel shaft surface hardness magnetic barkhausen noise nondestructive evaluations |
基金项目:国家自然科学基金(11527801)资助 |
|
摘要点击次数: 1170 |
全文下载次数: 297 |
中文摘要: |
利用磁巴克豪森噪声测试方法,实现了钢轴曲表面硬度的无损检测。首先,实验测试了传感器对曲表面磁巴克豪森噪声信号的重复检测性能,统计出不同励磁频率与幅值条件下的磁巴克豪森噪声信号包络线峰值和重复测试数据的变异系数图谱,并利用磁巴克豪森噪声的峰值与其变异系数的比值作为评价指标,优选了传感器工作参数;其次,利用变异系数评价传感器对磁巴克豪森噪声和切向磁场磁参量的重复测试性能,并筛选出变异系数较小的磁参量作为BP神经网络模型的输入;最后,研究了BP神经网络模型隐含层节点数对模型精度的影响,采用最佳优化模型对钢轴表面硬度进行预测,其预测平均误差仅为4.25%。 |
英文摘要: |
Non-destructive evaluation of surface hardness in steel shaft is achieved using magnetic Barkhausen noise (MBN) method. First, experiments are performed under the condition of different excitation frequencies and amplitudes to investigate the repeated detection performance of the sensor on measuring MBN signals. The charts, representing the average peak value of MBN envelope and the variation coefficient of the repeated test data are drawn. The ratio of the average peak value of MBN signals to the variation coefficient of the peak value of MBN signals is used as a criteria to help optimize the excitation parameters of the sensor. Second, the variation coefficient of measured magnetic parameters of MBN and tangential magnetic field is analyzed and magnetic parameters with smaller variation coefficient are selected as the input nodes of the BP neural network. Finally, the influence of the number of hidden layer nodes of BP neural network model on the accuracy of the model is investigated. The model with optimal structure is used to predict the surface hardness of the steel shaft and the average prediction error is only 4.25%. |
查看全文 下载PDF阅读器 |
关闭 |