基于RBF神经网络的曲面加工误差补偿

Error Compensation of Surface Machining Based on RBF Neural Network

在自由曲面数控加工中,计算误差和机床误差都会带来加工工件的形状误差。形状误差在三维空间分布是无规律的,无法用普通的数学函数表达,导致很难实施误差补偿加工。为了建立误差补偿模型,本文提出了采用RBF(Radial Basis Functions)神经网络逼近误差的三维分布函数。测试结果表明,RBF网络模型具有较好的推广能力,它与传统的BP神经网络模型相比较,RBF网络具有更高的精度、更好的泛化能力和更快的收敛速度。通过修改后的数控NC指令驱动数控机床,使刀具中心偏离一个误差函数求出误差值,实现误差补偿。

In NC machining of free-form surface, computing error and machine tool error could result in the form error of workpiece. Because of indiscipline of error distribution, it is difficult to find the error distribution function by traditional method. In this paper, RBF （Radial Basis Functions） Neural Network was employed to approach the three-dimensional distribution function of the error and then establish the error compensation model. The test results show that RBF Neural model has good extending ability. It has higher precision, better generalization ability and quicker convergence speed than BP Neural Network mode. Finally, the error compensation was achieved by the error value that calculated by tool center deviations from a error function through driving the NC machine by the modified NC instruction.