ABSTRACT
This study investigates the potential pre-warning of critical component failures in power plants using machine learning algorithms, more specifically regarding feedwater pumps and diesel engines. There are several sensors (e.g. temperature, pressure and smoke) that check the operational environmental condition through an interconnected system for transporting data or materials (Hossain et al., 2015). In this study, four machine learning models (ANN,RNN, LR, NB) was trained with a significantly large dataset and performance metrics like recall, precision was measured. The authors concluded that such an approach needs to be tailored according to specific power plant situations. “We showed that although ANNs offered precise estimates, RNNs were time-critical, LR gave constant prediction results while NB predictions varied more than others.” * Model selection: select the suitable model in order to correctly predict failures and promote plant reliability. Scope: This research will be helpful in providing the ground study to machine learning for power sector operational excellence, conjure predictive maintenance scheduling and avoiding catastrophic failures from causing big ticket damages.
