基于BP神经网络的数控机床综合误差补偿方法

摘要

针对多轴数控机床热影响导致的加工精度衰减问题,结合神经网络自学习与数据拟合能力,提出基于优化BP神经网络的多轴数控机床综合误差补偿方法.针对BP神经网络神经元误差曲面下降缓慢影响收敛效率的问题,引入陡度因子和放大因子,并基于此对数控机床运动轴加工精度进行预测和补偿.将大型A/B双摆角龙门数控铣床各关键发热源的温度检测数据和运动轴误差检测数据作为精度预测模型的输入量和输出量,采用改进后的BP神经网络进行训练,获得温度变化与位移误差量之间的非线性映射关系,并据此修改被加工工件的刀位数据文件,实现数控机床加工精度的提高.模拟算例和实验结果表明,该方法降低了传统BP神经网络的预测误差和运算时间,对机床平均误差补偿率达到50％以上.开发的数控机床误差补偿系统无须对现有机床进行大规模硬件改造,应用简便易于推广.%Machining precision is attenuated by thermal factors.Making use of neural network's strong self-learning and data fitting ability,a novel method based on improved BP-neural network algorithm is proposed for comprehensive error compensation of multi-axis machine tools.An improved algorithm introducing steepness factor and amplification factor is presented to improve the convergence efficiency affected by slow decline of neurons error surface,and to predict and compensate machining precision of motion axes.The measured data of the temperature of each key heat source and displacement errors of motion axis of the large A/B double-pendulum angle gantry milling machine are respectively regarded as input and output.A machining precision error prediction model of machine tool is established and trained by the improved BP-neural network.The nonlinear relationship between temperature and displacement errors is obtained.Cutter location data file of workpiece is modified accordingly and machining precision is improved.Simulation and experiments indicate that this method reduces the prediction errors and computing period compared with the conventional BP-neural network.It is unnecessary to greatly reform the existing machine tools for application of the comprehensive error compensation system.