Despite the progressing decarbonization of our power sources through the increase of renewable energy, conventional power stations like combined cycle power plants (CCPP) still contribute to the power mix by providing highly efficient backup power. To maintain an efficient operation of these plants, the early identification of occurring degradation is essential. In this context, this work deals with a novel approach of automated fault detection carried out by a neural network based on simulated process data. The initial comprehensive literature research on failure modes in combined cycle power plants and their thermodynamical impact serves as a solid foundation for their realistic simulation. Due to the necessity to generate large amounts of data to constitute every failure mode under various plant operation conditions, an automated workflow of data generation, preparation and validation is introduced. The process of constructing a neural network and enhancing its performance by optimizing the underlying data structure and the networks’ hyperparameters are shown. Finally, a statistical evaluation of different network models and their achieved results is conducted. The networks’ ability to detect both, the occurrence of single and multiple failure modes at a time, is evaluated. It can be shown that the developed neural network is capable of detecting the failure modes with high precision, even when noise is applied to the simulated process data to mimic the scatter of real plant measurements.