摘要
磁性测厚仪的传感器信号是强非线性函数,其准确拟合是测厚仪高精度测量的基础。应用BP网络对其进行拟合是一种新的尝试。本文将基于贝叶斯正则化L-M算法的BP网络应用于测厚仪的研究中,编制了神经元训练程序。将训练好的神经元用于测厚仪的软件设计中,仿真实验和测厚仪的实际测试结果表明,相对于平方多项式和分段线性拟合算法,该算法可提高精度,具有良好的泛化能力,经过较少的训练步长,即可满足实用精度。
The sensor signal of magnetic coating pachometer is strong non-linear function. It is essential to fit the sensor's character well for high precision measurement application. BP neural network is a new attempt to fit the data. Bayesian regularization L-M algorithm is used to magnetic coating pachometer and a program is designed for training BP neural network. The pachometer's software based on the trained network is designed. Comparing to the algorithm of polynomial square or subsection linear, the simulation and practical test result indicated that the measurement precision is improved. It has well generalized characteristic and needs less training step to meet applied request.
出处
《电子测量与仪器学报》
CSCD
2006年第2期39-42,共4页
Journal of Electronic Measurement and Instrumentation
关键词
涂层
测厚
神经网络
曲线拟合
coatings, pachometer, neural network, curve fitting.