The Application of an Artificial Neural Network as a Baseline Model for Condition Monitoring of Innovative Humidified Micro Gas Turbine Cycles

Abstract The MGT market is still considered to be niche and there are R&D&I challenges that need to be addressed to further promote this technology in distributed generation applications. Innovative MGT cycles based on a cycle humidification concept, can be considered to obtain higher system performance. However, given the fact that MGTs are installed close to the consumption points, where they are operated by non-technical prosumers with very limited access to maintenance services, they should also offer high availability and reliability to avoid unexpected outages and secure the supply. Therefore, intelligent monitoring systems are needed that can support non-expert end-users to detect degradation and plan maintenance before a breakdown occurs. In this study, we investigated and developed advanced methods based on artificial neural networks (ANNs) for condition monitoring of a humidified MGT cycle under real-life operational conditions. To create a high-performing model, extensive data preprocessing has been conducted to remove data outliers and select optimum model features, which provide best results. Additionally, the model hyperparameters such as learning rate, momentum and number of hidden nodes have been altered to achieve the most accurate predictions. The results of this study have provided a baseline ANN model capable of conducting condition monitoring of a micro-Humid Air Turbine (mHAT) system, which will be applied to additional studies in the future.