Modelling and Identification of a Six Axes Industrial Robot

This paper deals with the modelling and identification of a six axes industrial Stäubli RX90 robot. A non-linear finite element method is used to generate the dynamic equations of motion in a form suitable for both simulation and identification. The latter requires that the equations of motion are linear in the inertia parameters. Joint friction is described by a friction model that describes the friction behaviour in the full velocity range necessary for identification. Experimental parameter identification by means of linear least squares techniques showed to be very suited for identification of the unknown parameters, provided that the problem is properly scaled and that the influence of disturbances is sufficiently analysed and managed. An analysis of the least squares problem by means of a singular value decomposition is preferred as it not only solves the problem of rank deficiency, but it also can correctly deal with measurement noise and unmodelled dynamics. INTRODUCTION Robotic manipulators for laser welding must provide accurate path tracking performance of 0.1 mm and less at relatively high tracking speed exceeding 100 mm/s. To study the applicability of industrial roboticmanipulators as in Fig.1 for laser weld∗Address all correspondence to this author. ing tasks, a framework for realistic dynamic simulations of the robot motion is being developed [1]. Realistic dynamic simulations require that the model parameters are known with sufficient accuracy. Determination of the unknown values from Computer Aided Design (CAD) models may not yield a reliable and accurate representation as the CAD models often are insufficiently detailed; they may not include the actual models of the servo motors, the servo wiring and/or the bearings. Furthermore, obtaining the rotational inertias of the transmissions requires a dynamic analysis which is generally not a feature that is included in the CAD system. The most efficient way to obtain accurate values of the unknown model parameters may be experimental parameter identification using the assembled robot. The problem of obtaining the dynamic model parameters by means of experimental identification has been addressed by many authors. A general overview of experimental robot identification using linear least squares techniques can be found in the textbooks of [2] and [3]. This paper addresses the topic of scaling the the least squares problem in order to obtain an accurate estimate of the parameters. Furthermore, by means of a singular value analysis, the number of accurately identifiable parameters with respect to the level of noise and unmodelled dynamics will be determined. One of the difficulties in experimental identification is that it is not possible to uniquely identify all model parameters. This leads to the fact that the resulting parameter values loose their 1 Copyright c © 2005 by ASME Joint 1 Joint 2 Joint 3