Vision-Based Control of Flexible Robot Systems
This thesis covers the controlling of flexible robot systems by using a camera as a measurement device. To accomplish the purpose of the study, the estimation process of dynamic state variables of flexible link robot has been examined based on camera measurements. For the purpose of testing two application examples for flexible link have been applied, an algorithm for the dynamic state variables estimation is proposed. Flexible robots can have very complex dynamic behavior during their operations, which can lead to induced vibrations. Since the vibrations and its derivative are not all measurable, therefore the estimation of state variables plays a significant role in the state feedback control of flexible link robots. A vision sensor (i.e. camera) realizing a contact-less measurement sensor can be used to measure the deflection of flexible robot arm. Using a vision sensor, however, would generate new effects such as limited accuracy and time delay, which are the main inherent problems of the application of vision sensors within the context. These effects and related compensation approaches are studied in this thesis. An indirect method for link deflection (i.e. system states) sensing is presented. It uses a vision system consisting of a CCD camera and an image processing unit. The main purpose of this thesis is to develop an estimation approach combining suitable measurement devices which are easy to realize with improved reliability. It includes designing two state estimators; the first one for the traditional sensor type (negligible noise and time delay) and the second one is for the camera measurement which account for the dynamic error due to the time delay. The estimation approach is applied first using a single link flexible robot; the dynamic model of the flexible link is derived using a finite element method. Based on the suggested estimation approach, the first observer estimates the vibrations using strain gauge (fast and complete dynamics), and the second observer estimates the vibrations using vision data (slow dynamical parts). In order to achieve an optimal estimation, a proper combination process of the two estimated dynamical parts of the system dynamics is described. The simulation results for the estimations based on vision measurements show that the slow dynamical states can be estimated and the observer can compensate the time delay dynamic errors. It is also observed that an optimal estimation can be attained by combining slow dynamical estimated states with those of fast observer-based on strain gauge measurement. Based on suggested estimation approach a vision-based control for elastic shipmounted crane is designed to regulate the motion of the payload. For the observers and the controller design, a linear dynamic model of elastic-ship mounted crane incorporating a finite element technique for modeling flexible link is employed. In order to estimate the dynamic states variables and the unknown disturbance two state observers are designed. The first one estimates the state variables using camera measurement (augmented Kalman filter). The second one used potentiometers measurement (PI-Observer). To realize a multi-model approach of elastic-ship mounted crane, a variable gain controller and variable gain observers are designed. The variable gain controller is used to generate the required damping to control the system based on the estimated states and the roll angle. Simulation results show that the variable gain observers can adequately estimate the states and the unknown disturbance acting on the payload. It is further observed that the variable gain controller can effectively reduce the payload pendulations. Experiments are conducted using the camera to measure the link deflection of scaled elastic ship-mounted crane system. The results shown that the variable gain controller based on the combined states observers mitigated the vibrations of the system and the swinging of the payload. The presented material above is embedded into an interrelated thesis. A concise introduction to the vision-based control and state estimation problems is attached in the first chapter. An extensive survey of available visual servoing algorithms that include the rigid robot system and the flexible robot system is also presented. The conclusions of the work and suggestions for the future research are provided at the last chapter of this thesis.