Hydraulic Press Commissioning Cost Reductions via Machine Learning Solutions

Abstract

In industrial processes, PI controllers remain as the dominant control technique due to their applicability and performance reliability. However, there could be applications where the PI controller is not enough to fulfill certain specifications, such as in the force control loop of hydraulic presses, in which specific pressure profiles need to be ensured in order not to damage theworkpiece. An Iterative Learning Control scheme is presented as a Machine Learning control alternative to the PI controller, in order to track the pressure profiles required for any operational case. Iterative Learning Control is based on the notion that a system that realizes the same process repeatedly, e.g. hydraulic presses, can improve its performance by learning from previous iterations. The improvements are revealed in high-fidelity simulations of a hydraulic press model, in which the tracking performance of the PI controller is considerably improved in terms of overshoot and the settling time of pressure signal.