A kernel-based approach for identifying the dynamics of feed drives and robot joints

Date

2023

Authors

Ahmadi, K.
Mehrabi, M.
McPherson, J.D.

Journal Title

Journal ISSN

Volume Title

Publisher

Control Engineering Practice

Abstract

Linear regression models are widely used in controlling machine tool feed drives and robot joints. However, identifying the linear model’s parameters requires removing unmodelled dynamics such as machining forces or friction nonlinearity from the identification tests. This can be impractical, challenging, and sometimes impossible. This paper presents a new approach that identifies the linear regression parameters and unmodelled dynamics simultaneously. The presented approach utilizes kernel-based system identification to learn and extract the contribution of the unmodelled dynamics to the measured data. As a result, unbiased linear parameters and a function describing unmodelled dynamics are identified from the same test. We present two experimental case studies to show the effectiveness of this approach in important industrial applications. The first case simultaneously identifies the linear rigid body dynamics of a machine tool feed drive and the unmodelled machining forces from the controller signals. The second case identifies the linear rigid body dynamics of a robot joint and its unmodelled nonlinear friction from the controller signals. We compare the results with current standard methods, showing that the presented approach leads to unbiased linear parameters and physically consistent descriptions of unmodelled dynamics without requiring extensive specialized tests, adding external sensors, or designing elaborate filters.

Description

Keywords

System identification, Least Squares Support Vector Machine, Feed drive dynamics, Robot joint dynamics

Citation

Ahmadi, K., Mehrabi, M., & McPherson, J.D. (2023). A kernel-based approach for identifying the dynamics of feed drives and robot joints. Control Engineering Practice. Preprint.