基于神经网络的测头预行程误差预测研究

Neural Network-Based Prediction of Pre-Travel Error of Probe

机床在线检测(OMM,On-Machine Measurement)或者三坐标(CMM,Coordinate Measuring Machine)测量机检测中,主要考虑的测头误差有预行程误差、测头半径误差。预行程误差是其检测中极其重要的一项误差,占测头总误差的60%左右。以测头预行程误差预测为研究对象,先通过实验方法得出标准球上1297个测点的预行程误差,然后分析预行程误差的总体分布特征,运用Levenberg-Marquardt快速学习神经网络算法对其进行建模预测,经过对网络模型反复训练,以便得到较好的预测值,最后对预测误差进行统计学相关分析,分析结果表明所建立的神经网络预测模型是有效的。

In On-Machine Measurement （OMM） system or Coordinate Measuring Machine （CMM）, probe errors are mainly pre-travel error, probe radius error. Pre-travel error is an extremely important error during measuring which occupies about 60% of the total error. In this paper, we mainly regard the probe pre-travel error prediction as the research object, firstly, obtain pre-travel error on the 1297 standard ball-point through experiments, and then analyze the overall characteristics of a pre-stroke error, use Levenberg-Marquardtfast learning neural network algorithm to make modeled prediction, by repeated training on network models so that get better predictive value, and finally analyze the prediction error with statistics .The results show that the established nerve network prediction model is effective.