In order to resolve the problem of reducing the dynamic errors for CMM,the dynamic error sources were firstly analyzed. A model of CMM’s dynamic errors was built up with BP neural network, which broke through the limit which was modeling only for each component of the system. Also the derivation of complicated mathematical relationships can be avoided. As the BP network has the problem of

local minimum value and its approximation rate is slow,the particle swarm optimizer algorithm for BP network was introduced. Initial weight of BP neural network model for CMM was optimized by PSO, and it can improve the ability of global optimization for BP network and processing speed. In order to get training data a standard ball was measured on CMM of type Global Class 9158, and error compensation model was established.Experiments were made to verify the model. Results show that the mean error of CMM can be reduced 2.3μm by using the model.

针对坐标测量机动态测量误差补偿问题,分析了测量机动态误差的来源，利用BP神经网络对坐标测量机动态误差模型进行建模，突破了以往只是针对其各组成系统进行建模的局限，避免了复杂数学关系的推导。引入粒子群优化算法对坐标测量机BP神经网络误差模型的初始权值进行了优化，提高了网络的全局优化计算能力和运算速度。应用Global Class 9158型测量机在不同输入参数条件下测量了标准球，获得了网络训练数据，建立了误差补偿模型，进行了测量补偿验证，结果证明该模型可使坐标测量机的误差均值减小2.3μm。

粒子群优化算法, 坐标测量机, 动态误差, BP神经网络