Implementation of an Ensemble Kalman Filter for the LUMEN Oxidizer Turbopump Simulator

Kurzfassung

This paper documents the implementation and testing of a filtering framework based on the Ensemble Kalman Filter (EnKF) for the oxidizer turbopump (OTP) EcosimPro/ESPSS simulator model of the liquid upper stage demonstrator engine (LUMEN), a fully operational engine developed by the German Aerospace Centre (DLR). In the context of LUMEN, novel concepts based on artificial inteligence (AI) are investigated among other aspects. AI approaches are data driven, thus, require large amounts of data, which are usually not available in aerospace applications. Therefore, simulations are a fast and, theoretically, simple way to gather large amounts of data. However, they need to be consistent with data obtained through real tests, where Data Assimilation (DA) could offer a potential solution. DA tries to reduce the inevitable approximation-caused errors of simulations by combining measurements with simulation results, which are both error-prone. A common concept in this context are filters based on the linear Kalman Filter (KF). This work shows the selection of a KF-based filter that allows precise estimation of several model parameters and output variables of the OTP simulator. Based on multiple factors, the EnKF is chosen as the most promising filter variant. The EnKF is implemented in Python including a framework for the OTP simulator that should allow simple adaptions for future tasks. Afterwards, the EnKF is tested to ensure correct functionality of the implemented filter, where the required observational data is created by a second simulation environment. This simulated real-world system is then used in the last step to show that the chosen and implemented EnKF is capable of estimating ten out of ten output variables and eight out of ten model parameters correctly. Additionally, insights are gathered during this testing that should be helpful for the successful utilization of the implemented filter in future applications.