Condition-Based Maintenance for Wind Farms with Partial and Inaccurate Prognostics Information

Advances in monitoring techniques have led to the increasing use of prognostic information to reduce maintenance costs in wind power systems. However, when the prognostic information is not accurate and comprehensive enough, it is unreliable to determine maintenance actions for wind turbines based only on prognostics, as in existing studies. To bridge this gap, a maintenance optimization strategy is developed for wind farms in terms of condition monitoring with auxiliary maintenance information. In this work, we determine predictive maintenance thresholds while also optimizing the planned inspection intervals at the wind farm level. A new maintenance basis denoted as the posterior remaining useful life (RUL) is proposed to calibrate prognostics with the help of rough estimates of the turbine component lifetime from inspections. When prognostics is confident, maintenance actions can be guided by the predicted RUL alone, while conversely, standard predictive thresholds can also be optimized to complement time-based maintenance to prevent turbine component failures within two adjacent inspections, thus providing the potential for maintenance planning with partial and inaccurate prognostics information. We formulate the maintenance costs and policies into a large-scale optimization model to capture the economic dependencies between components within a turbine, as well as components across turbines in the farm. A numerical example is provided to demonstrate and validate the proposed method. Insights on the influence of prognostic error on maintenance costs are driven by comparative studies