Improved condition monitoring of hydraulic turbines based on artificial intelligence techniques
- Autores: Weiqiang Zhao
- Directores de la Tesis: Eduard Egusquiza (dir. tes.), Alexandre Presas Batlló (codir. tes.)
- Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2021
- Idioma: español
- Materias:
- Texto completo no disponible (Saber más ...)
- Resumen
As a type of renewable energy that can provide rapid response to the requirement of the power grid, hydropower plays a fairly important position in the energy market. In recent years, with the enormous entrance of new renewable energies (NREs) such as wind energy and solar energy, the stability of the power grid has been challenged: the intermittent power supply from the NREs requires the hydraulic turbines to work more in off-design conditions and regulate the output much more frequently than they did before. In this new scenario, several problems have appeared in hydraulic turbine units. In order to reduce the maintaining periods and critical damages on the unit, condition monitoring techniques have been proved to be a useful tool for operators. However, these techniques shall be improved and updated in order to consider this new situation for hydropower.
At present, hydraulic turbines have been monitored by different types of sensors. However, new data analysis technologies such as artificial intelligence haven’t been implemented in the systematical analyses of the prototypes. These techniques could improve the actual condition monitoring systems and could help to improve the diagnosis capacity for some critical problems, where classical analysis may fail.
In this study, existing monitoring and field test data from various types of turbines (Pump turbine, Francis turbine and Pelton turbine) has been used and several artificial intelligence (AI) techniques and data-driven methods have been applied in order to improve the existing condition monitoring techniques.
Firstly, for the pump turbine analyzed, artificial neural network (ANN) have been used to generate vibration hill charts based on the indicators used for condition monitoring. This has helped to analyze abnormal behaviors of the machine and to propose a better condition monitoring based on the generated maps. This can provide effective guidance for the operation plan of the unit.
Secondly, the limits of operation of a large Francis turbine due to overload instability have been analyzed. AI techniques have been applied on existing data to analyze the feasibility to detect the overload instability several seconds before it occurs. It is shown that by implementing these techniques in the existing condition monitoring system, the operating range of the unit could be safely increased.
Finally, for a failure that occurred in a Pelton turbine (broken bucket), artificial neural networks combined with dimension reduction techniques have been used to build up a model that can accurately predict the damage, which is helpful for the scheduled maintenance.
This is an Article-Based Thesis, so it is based on three Journal Papers that have been published during the thesis duration. These three Journal papers are about the improved hydro turbine condition monitoring and fault diagnosis based on AI techniques, and they are attached and commented though the whole document of this thesis.
