基于RBF神经网络的数控车床热误差建模

对于数控车床而言，热误差是其最大的误差源，而其中最困难的是热误差建模．现有BP算法的神经网络模型存在学习收敛速度慢，容易陷入局部极小点的缺点．文中使用径向基函数理论建立了基于RBF神经网络的数控机床热误差数学模型．讨论了RBF网络参数的初始化及学习；给出了两种建模方式的RBF网络建模算例，将其建模性能指标与经典最小二乘法建模指标进行综合对比，可知RBF网络各项指标均优于经典最小二乘方法．最后验证了RBF网络建模的鲁棒性．结果表明：径向基神经网络模型与经典最小二乘线性模型相比，拟合性能更好，预测补偿能力强且建模时间短．

Thermal Error Modeling of CNC Turning Center Using Radial Basis Function Neural Network

The traditional BP neural network approaches have some drawbacks such as low convergence speed and local minimal point. A neural network based on radial basis function(RBF) was used to predict and compensate the thermal error of a CNC turning center. The initialization and learning approach of RBF neural network was discussed. RBF neural network examples by two modeling ways were demonstrated. The modeling performances of RBF approach and LMS approach were synthetically compared. The validation of the modeling robustness was given at last. The experiment result shows that RBF network model makes more accurate predictions and compensation with less modeling time than the LMS linear models.