Proceedings of the
9th International Conference of Asian Society for Precision Engineering and Nanotechnology (ASPEN2022)
15 – 18 November 2022, Singapore

Linear Regression-based Parameter Identification of Machine Tool Feed Drive

Jong-Min Lim1 and Byung-Kwon Min1,a

1School of Mechanical Engineering, Yonsei University, 50, Yonsei-ro, Seodaemun-gu, Seoul, Republic of Korea


Feed drive model is essential for simulating the behavior of machine tools. For establishing the model, it is required to identify equivalent mass and friction coefficients of a feed drive. The dominant friction behavior is the Stribeck effect, which is linear at high speed and exponentially nonlinear at low speed. The algorithms for identifying these parameters can be categorized as online estimation and global optimization. The online identification algorithms have merits in high memory efficiency and low computational load. However, the nonlinearity in the model causes an inaccuracy of identification and difficulty of implementation. On the other hand, the global optimization methods can be used with high accuracy regardless of model linearity, while it requires a considerable amount of process time and high computational load. Thus, the algorithm has a limitation in the application of machine tools to achieve sufficient accuracy. For overcoming these problems, this study proposed a parameter identification method by using linear regression analysis-based approach. Measured system data were classified into positive and negative velocity sections based on the feed direction. The feed drive model was modified by multiplying a common term according to each model parameter. Each term of the equation can be divided into conservative and nonconservative with respect to velocity after integrating the modified model about each velocity section. From this, a linear equation with the slope of each model parameter can be derived. The parameter is identified by calculating the corresponding slope using linear regression analysis. Stribeck velocity, which cannot be represented as a slope of a linear equation, was calculated using a gradient descent-based method to make the corresponding linearity of other friction parameters maximum. The proposed identification method can significantly reduce the computational load and required time compared to global optimization methods. In the experiment, the feed drive model parameters of the testbed were identified. The computation time and the identification accuracy were compared to the genetic algorithm to validate the proposed method.

Keywords: Curve fitting, Conservative, Sliding friction, Stribeck curve

PDF Download