State estimation concept for a nonlinear melting/solidification problem of a latent heat thermal energy storage. (October 2021)

• Record Type: Journal Article
• Authors: Pernsteiner, Dominik; Schirrer, Alexander; Kasper, Lukas; Hofmann, René; Jakubek, Stefan
• Is Part Of: Computers & chemical engineering. Volume 153(2021)
• Publication Date: 2021-10
• Subjects: Extended Kalman filter -- Phase change material -- Nonlinear distributed-parameter system -- Singular value decomposition -- Balanced truncation; Chemical engineering -- Data processing -- Periodicals; 660.0285
• DOI: 10.1016/j.compchemeng.2021.107444
• Languages: English
• ISSNs: 0098-1354

Abstract

Highlights: An efficient observer for nonlinear high-order melting/solidification problems is proposed. The basis is the extended Kalman filter with balanced truncation and state extension. The efficiency, robustness, and accuracy of the observer approach are demonstrated. A mode-shape-based placement criterion for temperature sensors is presented. Abstract: Latent heat thermal energy storages (LHTES) utilize a material's phase transition to store energy at an almost constant temperature. To fully exploit their high energy density, reliable state estimation is essential, which requires a suitable model-based observer. In previous works, high-precision and real-time-capable models have been developed to solve the arising coupled Navier-Stokes and energy equations. In the present work, these high-order nonlinear models are applied to predict (simulate) the states of the LHTES one time step ahead. Then, an extended Kalman filter uses a reduced-order observer model derived from the prediction model by linearization and balanced truncation to compute a state update based on measurements. This approach increases both, computational efficiency and performance, since the observer can only update the state of the prediction model compliant to its dominant behavior. Different types of measurements can be accurately combined in the observer, resulting in fast convergence despite model errors.