During the last decades there has been a rise of awareness regarding the necessity to increase energy systems efficiency and reduce carbon emissions. These goals could be partially achieved through a greater use of gas turbine - solid oxide fuel cell hybrid systems to generate both electric power and heat. However, this kind of systems are known to be delicate, especially due to the fragility of the cell, which could be permanently damaged if its temperature and pressure levels exceed their operative limits. This could be caused by degradation of a component in the system (e.g. the turbomachinery), but also by some sensor fault which leads to a wrong control action. To be considered commercially competitive, these systems must guarantee high reliability and their maintenance costs must be minimized. Thus, it is necessary to integrate these plants with an automated diagnosis system capable to detect degradation levels of the many components (e.g. turbomachinery and fuel cell stack) in order to plan properly the maintenance operations, and also to recognize a sensor fault. This task can be very challenging due to the high complexity of the system and the interactions between its components. Another difficulty is related to the lack of sensors, which is common on commercial power plants, and makes harder the identification of faults in the system.

This paper aims to develop and test Bayesian belief network based diagnosis methods, which can be used to predict the most likely degradation levels of turbine, compressor and fuel cell in a hybrid system on the basis of different sensors measurements. The capability of the diagnosis systems to understand if an abnormal measurement is caused by a component degradation or by a sensor fault is also investigated. The data used both to train and to test the networks is generated from a deterministic model and later modified to consider noise or bias in the sensors.

The application of Bayesian belief networks to fuel cell - gas turbine hybrid systems is novel, thus the results obtained from this analysis could be a significant starting point to understand their potential. The diagnosis systems developed for this work provide essential information regarding levels of degradation and presence of faults in gas turbine, fuel cell and sensors in a fuel cell – gas turbine hybrid system. The Bayesian belief networks proved to have a good level of accuracy for all the scenarios considered, regarding both steady state and transient operations. This analysis also suggests that in the future a Bayesian belief network could be integrated with the control system to achieve safer and more efficient operations of these plants.

This content is only available via PDF.
You do not currently have access to this content.