Towards a digital twin: operational data evaluation techniques for online monitoring and condition assessment of large rotating equipment

With the energy transition, operational regimes of large rotating equipment are subject to significant changes. Turbogenerators that have been operating steady-state are facing more volatile regimes with increased load changes and reactive power control. In this scenario, the reliability of such assets is especially important for safe grid operation. To ensure this, techniques from statistical learning and digital twin implementations can assist domain experts supervising operation and maintenance. In this paper, two approaches are employed. First, an algorithm is introduced to detect long-term trends early before conventional protection rules would be triggered. Therefore, an approach is suggested using random sample consensus (RANSAC) regression models and binning by stator current. This procedure is demonstrated for a use case with bearing vibration. Secondly, the importance of quantifying the impact of changing operational behaviour with equivalent operating hours is emphasized. In this context, the contribution from start-stop cycles and load swings is growing.