| Due to little research on the dynamics of parallel kinematic mechanisms/machines (PKMs) or lack thereof has created an increasing awareness of the dynamic needs. Therefore, the thrust of the work presented in this dissertation is to establish a dynamic parameter identification approach for parallel kinematic machines. This approach is to acquire parameter values necessary for modeling the dynamics of parallel kinematic machines, e.g., mass, inertia, and friction.; To determine the best approach, several identification methods were compared and evaluated in terms of accuracy, convergence, ability to handle noisy data, and ease of implementation. The main comparison studies were conducted between the least squares (LS) method and the unscented Kalman filter (UKF) method. To illustrate the superiority of the unscented Kalman filter (method of preference), both the LS and UKF methods were applied to the inverted double pendulum case. Using results found in the literature, the extended Kalman filter capabilities were compared to these two methods, further substantiating the superiority of UKF method for nonlinear systems. In addition, force-based and energy-based modeling methods are compared to determine if there is any benefit other than reduced modeling effort in deriving the theoretical model. A significant improvement using energy-based method were not generally realized. To experimentally validate the capability and the implementation of the UKF method for dynamic parameter identification, the University of Florida Space, Automation and Manufacturing Mechanisms Laboratory's PKM dynamic parameters were identified.; The simulated and experimentally validated results show that the unscented Kalman filter performs well in identifying the system parameters. This merit corresponds to the significant reduction in modeling effort required to generate the basic algorithm used for the system identification. The overall impact to technology is provision of a parameter identification method that yields improved rigid body dynamics models of parallel kinematic mechanisms. This will lead to the ability to introduce advanced controllers that exploit the model dynamics to improve system performance. In addition, the presented method sets the foundation for formulating an online tuning algorithm. |
