HILS system with time series by machine learning

Abstract

HILS is an analysis method that controls a hardware based on values calculated by the analysis model. And measurement results of the hardware are reflected in the analysis model. In co-simulation, the system of interest is divided into multiple subsystems for analysis. The accuracy of co-simulation is degraded when the time step is large. In the case of a HILS system with a large computational load, such as a HILS system using multibody analysis, a large step time is required to solve the system in real time. In this study, we attempted to improve the accuracy of co-simulation by using time series prediction technique to estimate values one step ahead. First, we performed co-simulation using time series prediction for a 2-DOF vibration system, and examined how co-simulation accuracy could be improved. Time series prediction was then applied to a HILS system using a test experimental setup simulating a tire-suspension system. The effect of time series prediction on the HILS system was evaluated by comparing the results for fine and coarse time increments.