Abstract
The identification of the dynamic parameters constituting a robotic manipulator
dynamic model has the purpose of accurately estimating the values of such parameters
starting from experimental measurements of the robot motion and is the only practical
method that allows obtaining reliable values for a mechanical system with a minimal
complexity. The importance of dynamic parameter identification becomes obvious
especially in inverse dynamics control as well as dynamic simulation applications. In this
thesis, dynamic parameter identification is addressed, from both a theoretical and an
experimental point of view, in open chain robotic manipulators.
On the one hand, the dynamic model of a robotic manipulator is developed
starting from the dynamics equations according to the Gibbs-Appell formalism.
Therefore, the robot is assumed to be constituted by rigid links, being the dynamic
behaviour of the actuators independently modeled. Several friccion models linear with
respect to the coefficients are also considered in order to model friction phenomena at the
joints. Later, the equations constituting this dynamic model are rewritten linearly with
respect to the dynamic parameters to be identified and in matrix form, in order to allow
the application of numerical algorithms both for the analysis and reduction and for the
solution of the constituted equation system.
On the other hand, some fundamental aspects of dynamic parameter identification
are addressed both theoretically and experimentally. Thus, the generation of optimized
trajectories is addressed, making use of parameterization by means of finite Fourier
series, which allows taking profit of their periodicity. Also, a procedure is proposed for
the solution of the equation system which ensures the physical feasibility of the identified
dynamic parameters. This procedure is based on the application of technics for non-linear
optimization with non-linear constraints.
Finally, two identification methods are presented: the first one according to a
direct identification scheme, while the second is a combination of the direct and the
indirect identification schemes.
Both the robotic manipulator dynamic model and the different identification
methods proposed in this thesis are validated by means of their experimental application
to dynamic parameter identification in a PUMA 560 industrial robotic manipulator
provided with an open control architecture. The obtained results allow to extract some
conclusions with respect to their practical implementation.