Predictive maintenance framework applied to wind turbines sensor data

Maintenance costs have an important impact in wind energy profits. The main goal of this project is to implement a predictive maintenance framework as an added routine for controlling wind turbines status. Based on an statistical approach, this framework has a relevant impact in increase production and saving operational costs. Predictive maintenance methodology is based on an early warning system to detect failures (work orders emitted from the control system) before they occur. Two approaches are developed. Latent environmental and internal factors might be detected with power curve study, where individual power curve in a normal performance status is compared against individual empirical power curves in different states. On the other hand, specific part failures are more suitable to be found using normal models from sensors and then comparing with individual signals. Both ways end up with an on go failure detection system. Condition monitored data is extracted from turbines through SCADA, a system that provides sensor signals from different parts of the system. Failure data is processed from a the register of work order per machine