Evaluation of data-driven NARX model based compensation for multi-axial real-time hybrid simulation benchmark study

Actuator control takes a pivotal role in achieving stability and accuracy, particularly in the context of multi-axial real-time hybrid simulation (maRTHS). In maRTHS, multiple hydraulic actuators are necessitated to apply precise motions to experimental substructures thus necessitating the application of multiple-input multiple-output (MIMO)control strategies. This study evaluates the data-driven nonlinear autoregressive with external input (NARX) based compensation for the servo-hydraulic dynamics within the maRTHS benchmark model. Different from previous study, nonlinear terms are incorporated into the NARX model. Online least square and ridge regression techniques are utilized to estimate the model coefficients to achieve optimal compensation. The influence of various model order and window length is assessed for the NARX model-based compensation. The findings of this research demonstrate that NARX-based compensation has significant potential not only in facilitating precise actuator control for maRTHS but also in enabling robust control in the presence of unknown uncertainties inherent to the servo-hydraulic system.