| Friction severely limits performance in precise positioning systems. This thesis investigates one possible approach to counter friction related performance losses, friction cancellation. In this method a friction observer is used to estimate friction in real time. The friction estimate is then is used to provide a torque at the input to the system that is equal in magnitude and opposite in phase to machine friction. The majority of friction observers are model-based, i.e., phenomenological or empirical modeling is used to characterize friction behavior. Models are often constructed with specific operating conditions in mind, such as normal load and driving frequency, and do not perform well when conditions change. In addition, many adaptive models assume slowly time varying parameters and require persistent excitation—requirements that often cannot be guaranteed. This thesis investigates non-model-based estimation, an alternative approach to friction estimation. In this method, known rigid body dynamics of a machine and motion measurements are used to extract unknown external forces acting on the system. Three non-model-based approaches to friction estimation are considered. These include the classical Kalman filter, and two more recent methods, the predictive filter and local function estimation. This thesis also fuses model-based and non-model-based methods, leading to the development of two combined friction estimation methods. To compare the performance of model-based, non-model-based, and combined methods, exhaustive comparative studies are performed experimentally by applying friction cancellation during position tracking in a DC motor driven inertia. Comparative studies and robustness studies are also performed through numerical simulation of friction cancellation in a gear driven inertia with flexible gear teeth. The results show that non-model-based and combined friction cancellation have superior performance and excellent robustness compared to model-based methods. Closed loop stability of the DC motor positioning system is proven for the three non-model-based friction cancellation methods considered. |
