Computer vision has been on the rise over the last decades. The applications of computer vision are mostly found within the domain of robotics and Unmanned Aerial Vehicle's. The use of computer visions in precision industry is new and offers both opportunities and challenges. In
...
Computer vision has been on the rise over the last decades. The applications of computer vision are mostly found within the domain of robotics and Unmanned Aerial Vehicle's. The use of computer visions in precision industry is new and offers both opportunities and challenges. In high performance precision systems the demands on sensors is set high in terms of precision, delay and sample rate. In previous work a computer vision sensor was developed for the control of a microscopy stage. This development opened a path for computer vision sensors in precision systems. However, the limitations of computer vision sensors are challenging in the domain of control. A computer vision sensor needs to process an image before it can estimate the position of an object. This processing requires time which introduces a time delay into the control system. Due to the nature of the computer vision algorithm and computing devices, the delay in the sensor is time varying. Furthermore, since a controller typically requires a tenfold sample rate of the control bandwidth the limited frame rate of computer vision sensors imposes additional challenges.
In previous work, the delay and limited sample rate of the computer vision sensor limited the control bandwidth of a microscopy to 0.87Hz in translational range. Since high performance precision systems require higher bandwidths a new computer vision sensor is developed in this thesis. An existing microscopy stage is used to test the performance of a computer vision sensor. Ground vibrations in a laboratory environment have a peak value at 10Hz which defines the target bandwidth. In order to reach this target bandwidth a new computer vision sensor is proposed. The sensor uses a cost-effective camera and a computer to run the computer vision algorithm.
By decreasing the delay from 60ms to 10ms a control bandwidth of 7Hz, using PID, is obtained in this work. Since the target bandwidth is 10Hz time delay compensation is considered to improve the performance. By implementing a Filtered Smith Predictor the target bandwidth of 10Hz is reached. The maximum control bandwidth of the controller, 15Hz, was derived in order to inspect the limit of computer vision sensors in precision systems.