原位检测系统中触发式测头半径误差分析与建模∗

摘要

This paper analyses the major components and influencing factors of the radius error of touch trig-ger probes about on-machine verification system. An effective probe radius error modeling and compensa-tion, an implementation procedure and an experimental study are presented. The error compensation model has been built with BP artificial neural network optimized by Genetic algorithm. According to the normal vector information of measurement points, predicting the probe radius error values and establishing the map-ping relationship with the normal vector of measurement points, and correcting on-machine verification data off-line or on-line by the computer. The results of experiment have shown that the predictive values based the model and the actual probe radius errors have good consistency. 65% of the residual error in the test sample is less than 1μm, maximμm is 2. 9μm and the average value is 0. 94μm.%文章通过分析原位检测系统测量过程中触发式测头半径误差的主要组成及其影响因素,提出了一种基于遗传-神经网络的测头半径误差建模和补偿方法,阐述了其实现过程和实验测试方法。利用遗传算法优化BP神经网络建立测头半径误差补偿模型,根据实际测点的法矢信息预测建立测点法矢方向与测头半径误差的映射关系,并通过计算机实现原位检测数据的离线和在线修正。实验结果表明基于该模型的预测值和实际测头半径误差具有较好的一致性,测试样本中65%的残余误差小于1μm,最大为2．9μm,平均值为0．94μm。