Session: 03-20-03: Manufacturing: General

Paper Number: 145249

145249 - Improving Contouring Accuracy of Cnc Systems by Improved Rnn Prediction and Pre-Compensation 

Computer numerical control (CNC) system is the core enabling device of intelligent manufacturing, and high-precision motion control is crucial for achieving high-speed and high-precision processing in CNC systems. Existing research mostly focuses on improving the tracking accuracy of single-axis motion in CNC systems. However, the machining motion of machine tools is essentially composed of contour motion formed by the combination of multiple feed axes. Therefore, direct multi-axis contouring control can more effectively improve the motion control accuracy of CNC systems.

Contour error is a control evaluation index in geometric space, defined as the shortest distance between the actual motion position and the theoretical trajectory. Typically, it cannot be directly measured by sensors. Therefore, contouring control requires precise modeling and calculation of contour error. This study proposes a high-precision contour error prediction method based on recurrent neural network (RNN). Initially, trajectory tracking is performed on randomly generated Non-Uniform Rational B-Splines (NURBS) curves to obtain training datasets. Then, an RNN is employed to establish an input-output mapping model for single-axis servo feed systems to obtain the expected motion positions of each feed axis. Finally, the Newton numerical calculation method is utilized to compute the shortest distance between the predicted actual motion position and the theoretical trajectory, achieving high-precision prediction of contour errors.

Existing contouring control methods can be divided into two categories: feedback-error-based contour controller design and pre-compensation strategies for correcting errors in the theoretical trajectory. The pre-compensation method can eliminate potential contour error before machining motion without the need for adjustments to the controller, making it suitable for closed-loop commercial CNC systems. Traditional pre-compensation methods involve mirroring the predicted contour errors onto the theoretical trajectory in a feedforward manner. However, this slight path adjustment may introduce unnecessary low-frequency vibrations into the machining trajectory, leading to a decrease in surface quality. This study further proposes a pre-compensation method considering vibration suppression based on accurate prediction of contour error. Finite Impulse Response (FIR) filter is used to smooth the motion trajectory after feedforward compensation. It can reduce the low-frequency vibrations in the compensated trajectory while improving contouring accuracy.

Validation experiments were conducted on a multi-axis CNC experimental platform. The proposed method achieved an average tracking error prediction deviation of less than 3μm for each axis of the test trajectory. The root mean square value of the compensated trajectory contour error was reduced by half, and the maximum value was effectively suppressed. Compared to traditional compensation methods, the proposed pre-compensation method effectively suppressed the vibration of the machine tool caused by the compensated trajectory，while sacrificing some contouring accuracy.

Presenting Author: Jiamu Song Tsinghua University

Presenting Author Biography: Jiamu Song is a doctoral candidate in the Department of Mechanical Engineering at Tsinghua University, China. His work primarily focuses on intelligent numerical control technology and precision control technology

Authors:

Jiamu Song Tsinghua University
Bingran Li Tsinghua University
Peiqing Ye Tsinghua University