Author(s): Arthur Favrel; Quentin Dollon

Linked Author(s):

Keywords: No Keywords

Abstract: In this paper, we propose a Bayesian approach to infer the parameters of 1D hydro-acoustic models of hydroturbine draft tubes. Our method combines sparse and noisy pressure data from monitoring systems with non-linear Kalman filters (KFs), effectively merging data and physics to estimate the underlying dynamics of the hydraulic system. We demonstrate the proof of concept using synthetic data from a SIMSEN model of a 140 MW Francis turbine unit, which includes a lumped model of the draft tube flow under part-load conditions. SIMSEN simulations with predefined model parameters are performed at specific stationary operating conditions, in resonance and non-resonance conditions. The resulting pressure signals serve as groundtruth for the inference step. During inference, the draft tube model parameters are considered unknown and then estimated using a Kalman filter. Our results show that this approach successfully retrieves the targeted parameters with low uncertainty, paving the way for real-time calibration of hydropower units physics-based digital twins using monitoring data.

DOI: https://doi.org/10.1088/1755-1315/1561/1/012039

Year: 2025