SUBJECT: M.S. Thesis Presentation
BY: Qifeng Liu
TIME: Friday, November 12, 2021, 12:30 a.m.
PLACE: Institute of System Dynamics, n/a
TITLE: Geometric Error Identification for 6DoF Robotic Manipulator Calibration to Improve Absolute Positioning Accuracy
COMMITTEE: Dr. Andrei Fedorov, Co-Chair (ME)
Dr. Oliver Sawodny, Co-Chair (Stuttgart)
Dr. Aldo Ferri (ME)
Dr. Cristina Tarin (Stuttgart)


This thesis tackles the manipulator inaccuracy issue from two aspects to achieve the desired accuracy enhancement: tighter mechanical tolerances and a closer matching kinematics model with the actual robot. For these purposes, according to the geometry of a pneumatically driven six degrees of freedom manipulator, a 6D parametric kinematics model is first formulated. The proposed model is highly flexible in terms of introducing, anywhere in each linkage of the manipulator, any number of virtual mechanical tolerance points that lump effects of dimension and orientation deviations caused by geometric errors. Therefore, whichever concerned mechanical tolerances can be added to the model and studied through Fuzzy arithmetic to analyze their influence on the tool center point position.

Meanwhile, geometric errors are also the primary source of discrepancies between the nominal model and real hardware. The model can include translational and rotational error parameters that need identification to quantify the effects from the geometric errors at the locations of the virtual tolerance points. For an effective identification, dependent error parameters can be systematically eliminated with QR decomposition. Once the model reduction is complete, the nonlinear least-squares optimization technique using the Gauss-Newton line search method is applied to identify the remaining independent error parameters. The identification process is eventually verified on the experimental manipulator. In a nutshell, the thesis presents, on the one hand, a tolerance analysis tool that offers insights for potential targeted manufacturing improvements to decrease the dominant tolerances, and on the other hand, a capable parameter identification process that rectifies the nominal kinematics model to agree with the hardware behaviors.