When designing system that require a high positioning accuracy, a stiff and light design is required. However, to truly exploit a good design, proper control strategies need to be applied. Therefore, during my master´s I minored in motion control theory. I took several courses where advanced control strategies where studied and put to practice in real-life motion systems. Here, two examples of such systems are presented.
The SISO is a mass-spring/damper-mass rotating system. The aim of the project is trajectory planning and feedforward tuning to reduce the positioning error of the non-collocated mass.
Frequency response function (FRF) measurement of the plant is performed in order to identify the transfer function of the system. To further reduce the error when controlling the non-collocated plant, a feedforward tuning is performed. For this purpose, a third order set-point trajectory is designed and applied to the system in order to quantify the effects of velocity, acceleration and dry friction components in the error. From the tracking error profile, a feedforward tuning trajectory is performed
Additionally, a power spectral density analysis is performed to gain further insight in the error sources governing the system.
Here the error profile after applying a "challenging" trajectory to the system can be appreciated. In blue the error before feed-forward tunning is shown. After that, the feedforward tunning is performed an at each step of the tuning the feedforward compensates for one of the above mentioned components of the error. Finally in black the resulting error after tuning can be appreciated.
ILC is an efficient control strategy that “learns” from previous tasks on a system. This strategy has been successfully implemented in diverse systems such as wafer stage in photolithography machines. It is based on the idea that a predictable error source can be accounted for when designing a controller.
A printer is a suitable system to perform iterative learning control, as the operations are highly repetitive and consequently the positioning error has a distinguishable pattern. Taking advantage of this phenomena the controller can learn from previous tasks and improve the positioning accuracy. The learning algorithm is run several times into the system, and the controller is updated at each iteration
Here the ILC algorithm is applied to the system. It can be appreciated how the system converges after several iterations (7 iterations).