基于RBF神经网络的高精度在线激光测厚算法

摘要

To improve the accuracy of the single CCD dual light path laser triangulation dynamic online thickness measurement system,a method that fitting the relationship between the position of spots on CCD target surface and the thickness of the measured object based on RBF neural network was proposed.After theoretical analysis,the experiments with two methods were carried out and compared,the one is that fitting the relationship between the position of two spots on CCD target surface and the thickness of the measured object based on RBF neural network,the another is that the existing least squares fitting method of three polynomial models,and the two methods get a network and an approximate mathematical model respectively.Experiments are done in different locations through using ten standard thickness gauges,and it is found that the first method is better than the second method obviously.The results show that the method that fitting the relationship between two spots and the thickness of the measured object based on RBF neural network can improve the accuracy of existing system,and it has strong robustness and meets the requirements of time complexity.%为进一步提高单CCD双光路激光三角法动态在线测厚系统的测量精度,提出了一种基于RBF神经网络的CCD靶面上光斑位置与被测物厚度之间函数关系的拟合算法.通过理论分析之后,设计了基于RBF神经网络直接拟合CCD靶面上两光斑位置信息与被测物厚度之间关系和现有的最小二乘法拟合三次多项式模型方法进行实验对比,两种方法分别得到了一个网络和一个近似数学模型.通过使用十组标准厚度塞尺在不同位置进行验证实验,发现前者方法计算出的厚度值更加靠近塞尺的客观值.实验结果表明,用RBF神经网络拟合两个光斑坐标和被测物厚度之间的关系,成功地提高了现有系统精度,鲁棒性好,时间复杂度尚可.