With the increasing entrance of wind and solar power for the generation of electricity, more flexibility is demanded to hydropower plants. More flexibility means that hydro turbines have to increase the operating range between minimum and maximum power. In Francis turbines the maximum power is limited by the appearance of a strong hydraulic excitation called overload instability. When the turbine operates at loads higher than design, the cavitating vortex rope that is generated in the draft tube may become unstable, producing huge pressure fluctuations, vibrations and power swing. Turbines are not allowed to operate under these conditions in order to avoid the destruction of the unit. The overload instability emerges abruptly, even when the machine is operating in a smooth condition. No visible transition can be detected by the monitoring system, so turbine operators have no margin to react. To avoid this phenomenon, operators limit the maximum power much before reaching this condition. By doing that, the maximum power is limited as well as the regulation capacity of the unit. In this paper, the feasibility of detecting the onset of this phenomenon is analyzed. Data-driven methods and artificial intelligence techniques, including principal component analysis, self-organizing map and artificial neural networks, are applied to the data available from experimental tests in a Francis turbine. The signals of vibration, pressure fluctuations and other parameters are combined and studied. The possibilities of a premature detection of the instability before it occurs are discussed. The method could be implemented in the monitoring system of the unit so that the operating range could be safely increased.