The ever-increasing penetration of intermittent energy sources introduces new demanding operating regimes for the bulk power generation in the power grid. During a worstcase power system disturbance scenario, the generator needs to operate beyond its limits to maintain stable operation. Therefore, a new online low-order thermal model of a hydrogenerator has been recently proposed, where the periodic extension of the long-forgotten capability diagram of the machine was in-depth investigated. An increased performance can be obtained if the total thermal capacity of the generator is exploited by applying optimal control theory in the Automatic Voltage Regulator (AVR). This paper proposes a Non-linear Model Predictive Controller (NMPC) combined with an Unscented Kalman Filter (UKF) with a modeling framework geared for use in a supervisory control structure for the conventional control system. The method provides maximum utilization of the machine's thermal capacity by providing the controller with an Enhanced Capability Diagram (ECD). Case studies on a single-machine system and a 16-bus multi-machine system were investigated. The results show that a satisfying controller action during different long-term voltage instability scenarios is realized.