With higher demand in industry for accurate, reliable manipulators comes an implicit need for robot users to know well the condition of their machines. Rising expectations of performance require improved methods of performance assessment to improve reliability.

The present work is a study of one aspect of robot reliability:​ the integrity of the mechanical structure of the manipulator. This work investigates how to include looseness faults in a manipulator model, and how to assess the integrity of the structure with a reasonable number of sensors. Alternative modeling strategies were assessed in the context of looseness diagnostics, leading to the choice of a model based on joint and end effector motion.

A method is developed for locating a single looseness fault in a serial-link manipulator from random joint inputs. The method looks for changes in manipulator end effector motions when looseness is suspected, by producing normal random motions in the manipulator joints. The direction of the non-normal component of end effector motion is the direction of looseness. Individual joints are then moved to new poses, starting from the proximal end, until the link of concern changes orientation with respect to the end effector, causing a change in the direction of the looseness motion.

Parametric and nonparametric methods were investigated for identifying the direction of looseness motion at the end effector. Results came both from simulation and from experiment using an industrial manipulator, a custom-designed looseness rig, and a planar manipulator with variable kinematics. Both structural looseness and joint backlash were successfully identified in the planar manipulator.

The conclusion discusses industrial implementation of the method, and recommends further work and strategies for robot diagnostics.