Answers are grounded in this paper and cite their sources.
Hydraulic turbines are increasingly operated under non-optimal operating conditions to meet modern power regulation demands. These conditions significantly reduce their lifespan due to an increase in the forces and moments acting on the machine. This article presents a novel methodology based on physics-informed neural networks to predict the loads on the runner, rotor centre of gravity, generator, exciter and bearings. The intensity of the loads at different operating points and locations within the turbine was also investigated. The performance of the new methodology was tested using measurements of displacements at the bearings and bending moments along the shaft carried out on a Kaplan turbine prototype operated at four distinct operating points. The study demonstrates that the model can learn the dynamics of complex rotor-dynamic systems and solve inverse problems effectively. The load predictions show that the induced damage is not only operating-point dependent but also strongly influenced by the load’s type, direction and location. It suggests that isolated vibration measurements tracking the degradation of specific components cannot be generalized. Thus, it is essential to calculate the loads acting directly on the component of interest. Finally, the model shows a fast execution time, taking around 5.8 s to compute and return the results. This makes the current model suitable to be integrated in digital twins or in real-time monitoring tools.
No takes yet. Share an insight, caveat, or question.
Roig et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: