Name: | Description: | Size: | Format: | |
---|---|---|---|---|
220.46 KB | Adobe PDF |
Advisor(s)
Abstract(s)
Grinding and polishing are typical application paradigms in which efficient control is needed for approximately and partially known multivariable, nonlinear, strongly coupled mechanical systems (robots) under strong dynamic interaction with an unmodeled environment. A novel adaptive approach to this problem using uniform structures and procedures as well as a passive compliant component as an essential part of the control was recently invented. The method seems to overcome the limitations of the classic approaches as limited speed of motion and supposed separability in the operational space supposing free directions for force/torque components and for free components of translation in their orthogonal sub-spaces. Like Soft Computing, instead developing the formally exact analytical model of the robot, its environment and the dynamic interaction between them the proposed method uses uniform structures but these are derived from the Euler-Lagrange equations considered in a general and formal level of abstraction. In contrast to the general approach fit to a quite wide class of problems, these structures are rather fit to a far narrower task of modeling and control of mechanical devices. This results in a drastic reduction in the number of tunable parameters, fast tuning for those parameters for which no a priori linguistic rules are available and uses simple fuzzy rules for tuning other parameters for which at least qualitative a priori known tuning are known. The proposed technique also is free from "scaling problems" so characteristic to the classic ones. The method is proved and illustrated via simulation in the case of a 3 DOF SCARA arm used for polishing a convex surface as an application paradigm.
Description
Keywords
Adaptive control Soft computing Uniform structures Reduced number of free parameters Machine learning