Towards Data-Driven Real-Time Hybrid Simulation: Adaptive Modeling of Control Plants

Abstract

We present a method for control in real-time hybrid simulation (RTHS) that relies exclusively on data processing. Our approach bypasses conventional control techniques, which presume availability of a mathematical model for the description of the control plant (e.g., the transfer system and the experimental substructure) and applies a simple plug 'n play framework for tuning of an adaptive inverse controller for use in a feedforward manner, avoiding thus any feedback loops. Our methodology involves (i) a forward adaptation part, in which a noise-free estimate of the control plant's dynamics is derived; (ii) an inverse adaptation part that performs estimation of the inverse controller; and (iii) the integration of a standard polynomial extrapolation algorithm for the compensation of the delay. One particular advantage of the method is that it requires tuning of a limited set of hyper-parameters (essentially three) for proper adaptation. The efficacy of our framework is assessed via implementation on a virtual RTHS (vRTHS) benchmark problem that was recently made available to the community. The attained results indicate that data-driven RTHS may form a competitive alternative to conventional control.