Industrial Steam Turbine Generator Predictive Maintenance Based on Measurement Availability

Large steam turbine generators’ reliable operation is very important due to catastrophic consequences for factory safety and power grid availability in case of unexpected faults. Complexity of the large system and its synergistic nature with production provide a large number of highly nonlinear measurements from integrated monitoring systems. This creates a big data volume with difficulty in storage and real-time processing, resulting in loss of information or general amplitude threshold monitoring. In this work, journal bearing imbalance fault is investigated using the installed sensor data of a large steam turbine as received by the thermal power plant monitoring system. Based on a short review of relevant literature state-of-the-art, a vanilla Long Short-Term Memory Recurrent Neural Network as a time-series predictor is proposed to mitigate trips during start-up and investigate correlation between measurements and bearing remaining useful life.