Author(s): Martin Gagnon; Quang Hung Pham; Vincent Mai; Arthur Favrel

Linked Author(s):

Keywords: Virtual sensors; Strain measurement; Francis turbine; Neural networks

Abstract: Strain on the blades of a Francis turbine cannot be measured continuously. A dedicated measurement campaign is required. This paper describes a method of using data gathered by a turbine’s continuous monitoring system to estimate strain on its blades indirectly. The methodology uses neural networks (NN) to estimate strain signal envelope during turbine startup, making further optimization of the turbine startup scheme possible even after the strain gauges are removed. The methodology also allows uncertainty quantification and credibility assessment across a wide range of head and downstream water levels given historical data from the unit monitoring system.

DOI: https://doi.org/10.1088/1755-1315/1483/1/012013

Year: 2023