Estimating model parameters of liquid rocket engine simulator using data assimilation

Abstract

To generate training data for fault detection and diagnosis based on machine learning, a System-Level Simulation (SLS) is developed to model the global behavior of a reusable liquid rocket engine employing the expander-bleed cycle. Model parameters are estimated automatically by an Ensemble Kalman Filter. Parameters are estimated starting from the turbopump component model, followed by the whole rocket engine model, to increase accuracy. Sufficient convergence is obtained within one day by parallelizing the SLS computation for the ensemble members by two. Compared to the static-firing test results, reasonable agreement is obtained within error of 2.3% in the steady-state engine condition. Even for the startup and shutdown sequences, the present SLS reasonably agrees with the test results except for the temperature. Comparing the model parameters obtained for different static-firing test results, the effect of the overhaul inspection on the turbopump is captured as a difference in the model parameters.