Development of an Automated Robotic Edge Deburring System
Theses
Overview
Overview
abstract
This thesis describes the development of a system for automated robotic edge deburring. The main emphasis is on accurate sensing and control of the chamfer depth. The depth is controlled through a custom built active end effector mounted to a PUMA-762 robot. The end effector's design objectives are obtained from an analysis of the combined robot arm, end effector, and deburring process dynamics. The completed unit had a bandwith of 65 Hz, and an accuracy of 0.01 mm. The depth is first controlled indirectly, by minimizing the variance of the normal deburring force. Several non-adaptive control algorithms based on a time series process model are investigated. Following computer simulations, experiments are performed on 1018 steel, straight edged parts. The extended horizon design is found to achieve the lowest force variance (0.11 N²), and the smoothest chamfer, with a roughness of 9.5 μm (ISO Ra). Simulations are performed to access the potential benefits of parameter adaptive force control for robotic deburring's time varying process dynamics. An adaptive version of the Smith predictor is found to be the more robust, and have a faster response time than a non-adaptive version. A model reference adaptive control algorithm is also investigated. A version sensor is then developed to more directly measure the chamfer depth. The sensor has a measurement rate of 105 Hz, with an accuracy of ±13 μm over a 1 m range. An automated inspection system is developed using the same hardware with modified software. The information from the vision and force sensors is combined using a sensor fusion scheme. This results in a more accurate and reliable measurement than possible using each sensor alone. Based on the force control results, adaptive generalized predictive control is chosen to control the depth. Separate process models are used for the contact and non-contact dynamics. To adapt to changes in dynamics the contact model parameters are estimated on-line. The control algorithm is then modified to include learning control. This new algorithm is shown, both theoretically and experimentall, to improve the regulation performance of the original algorithm, without affecting its stability, for processes with a partially repeatable disturbance. Depth controlled experiments are performed on straight edged and planar parts made of 1018 steel, and 304 stainless steel. Feedrates of 25 and 50mm/s, and depth setpoints of 0.3 and 0.4 mm are used. In comparison to non-adaptive control, the regulation is improved with adaptive control, and further improved with the adaptive learning algorithm. The part material, depth setpoint, and feedrate are found to have little effect on the deburring performance. The system's feedrate are found to have little effect on the deburring performance. The system's accuracy is found to be dependent on the part geometry. For straight edged parts the accuracy is ±0.03 mm, for the most complex planar part tested it is ±0.06 mm.
