Wind energy plays a crucial role in the energy transition. However, it is often seen as an unreliable source of energy, with many production peaks and lows. Some of the drivers of uncertainty in energy production are the unexpected wind turbine (WT) failures and associated unscheduled maintenance. To support an effective health management and maintenance planning of WTs, we propose an integrated data-driven framework for Remaining Useful Life (RUL) prognostics and inspection planning of WTs. We propose a Long-short term memory (LSTM) neural network with Monte Carlo dropout to estimate the distribution of the RUL of WTs, i.e. we develop probabilistic prognostics. Different from existing studies focused on prognostics for single components, we consider the simultaneous health-monitoring of multiple components of the WTs, thus seeing the turbine as an integrated system. The obtained prognostics are further included into a stochastic planning model which determines optimal moments for inspections. For this, we pose the problem of WT inspections as a renewal reward process. We illustrate our framework for four offshore WTs which are continuously monitored by Supervisory Control and Data Acquisition (SCADA) systems. The results show that LSTMs are able to estimate well the RUL of the WTs, even in the early phase of their usage. We also show that the prognostics are informative for maintenance planning and are conducive to conservative inspections.
Read full abstract