Learning Adaptive Controller for Hydraulic Machinery Automation

Abstract

The automation of hydraulic machinery has the potential to improve productivity and reduce human labor in many industries. However, the complex dynamics of hydraulic actuators, variability from machine to machine, and system degradation over time make it challenging to design controllers for hydraulic machine automation. Consequently, existing approaches rely on manual tuning and data collection. In this letter, we propose an approach to train an adaptive controller for this problem. The controller can be trained purely in simulation, and at the time of deployment, it can adapt to the dynamics of the real system within minutes. After the adaptation, precise motion control can be achieved. We validated the approach by testing a single controller trained with the proposed method on two hydraulic machines that are distinctly different in size, application, and age. The results show comparable control performance of our general approach compared to previous methods, which rely on machine-specific data and training.