Recursive Learning of Feedforward Parameters in High-Tech Motion Systems: An Experimental Case Study

This article provides an experimental case study of a state-of-the-art recursive feedforward parameter learning framework on a high-tech industrial metrology and inspection machine. The aim of the learning framework is to recursively adapt the feedforward parameters to compensate for time-varying and position-dependent system behavior, e.g., caused by wear, position and temperature dependent actuator characteristics, changes in shape and stiffness due to thermal expansion, and sample time jitter. The strength of the approach is demonstrated through experiments on a high-tech motion system which show a peak error reduction of circa 45% compared to the industrial controller with offline calibrated feedforward parameters.