【作者】 刘志强；

【导师】 李建刚；

【作者基本信息】 哈尔滨工业大学 ， 电子信息（专业学位）， 2023， 硕士

【摘要】 数控机床是高端制造业的基础,被广泛应用于叶片、齿轮等高精密零件的加工,其控制精度直接决定着零件的加工质量,提高数控机床的控制精度能够显著提高零件的加工质量。目前常用的轮廓控制方法例如交叉耦合和迭代学习控制只能对已经产生的轮廓误差进行补偿,在五轴非重复性加工的任务中,难以实现较好的补偿效果。针对五轴数控机床非重复性加工场景,本文提出了一种基于神经网络预测的五轴数控系统轮廓控制方法,对数控机床加工过程中的轮廓误差进行补偿。具体的研究内容如下:根据数控系统是一个近似线性系统并存在非线性扰动的特点,选取了适用于神经网络预测的线性和非线性特征,并搭建了用于五轴仿真的数控系统仿真模型。提出了一种五轴随机刀具路径的生成方法,通过非均匀有理B样条曲线对生成的随机单轴控制点进行拟合,从而得到随机的五轴刀具路径。采用时间卷积神经网络对数控机床单轴的跟踪误差进行预测,并对不同参数下神经网络的预测效果进行了对比,选取了一组最佳的实验参数。针对五轴刀具位置轮廓误差和刀具方向轮廓误差存在耦合的问题,提出了一种针对五轴的轮廓误差补偿方案,分别对刀具位置和刀具方向轮廓误差进行补偿,得到补偿之后的刀具位置和刀具方向。针对轮廓误差补偿过程中单次补偿无法补偿到位的问题,提出了一种根据误差预测结果来进行迭代补偿的方法,先通过神经网络建立数控机床的误差预测模型,然后根据误差预测模型的预测结果来进行迭代补偿。最后,在五轴数控机床上对本文提出的轮廓误差补偿方法进行验证,实验结果验证了本文提出的轮廓控制方法能够很好地对轮廓误差进行补偿,五轴加工过程中的轮廓误差能够得到明显的降低,并且不会出现发散的问题。更多还原

【Abstract】 Computer Numerical Control(CNC)machine tools are the foundation of high-end manufacturing and are widely used in the processing of high-precision parts such as blades and gears.The control accuracy of CNC machine tools can directly determines the processing quality of parts.Improving the control accuracy of CNC machine tools can significantly improve the processing quality of parts.Currently commonly used contour control methods such as cross-coupling and iterative learning control can only compensate the contour errors that have occurred,and it is difficult to achieve a better compensation effect in five-axis non-repetitive machining tasks.Aiming at the non-repetitive machining scene of five-axis CNC machine tool,this thesis proposes a contour control method of five-axis CNC system based on neural network prediction to compensate the contour error during the machining process of CNC machine tool.The specific research content is as follows: Firstly,according to the characteristics that the numerical control system is an approximate linear system and there are nonlinear disturbances,the linear and nonlinear features suitable for neural network prediction are selected and a simulation model for five axis NC system is built.Secondly,a five axis random tool path generation method is proposed.The random single axis control points generated are fitted by non-uniform rational B-spline curves,and the random five axis tool path can be obtained.Temporal convolutional neural network is used to predict the single-axis tracking error of the CNC machine tool,and the prediction effect of the neural network under different parameters is compared,and a set of optimal experimental parameters is selected.Then,aiming at the coupling problem of five-axis tool position contour error and tool direction contour error,a contour error compensation scheme for five-axis is proposed,which compensates the tool position and tool direction contour errors respectively,and obtains the tool position and tool direction after compensation.Aiming at the problem that a single compensation cannot be fully compensated in the process of contour error compensation,an iterative compensation method based on the error prediction results is proposed.The error prediction model of the CNC machine tool is first established through the neural network,and then perform iterative compensation according to the prediction results of the error prediction model.Finally,the contour error compensation method proposed in this paper is verified on a five-axis CNC machine tool.The experimental results prove that the contour control method proposed in this paper can compensate the contour error well.The contour error in the five-axis machining process can be significantly reduced and there will be no divergence problems.更多还原