Wind turbines (WTs) are complex systems, subject to a hostile environment that promotes accelerated aging of components, which increases maintenance cost and operational expenditure (OPEX). So, the development of methods of fault diagnosis and remaining useful life (RUL) estimation will make it possible to substitute preventive or corrective maintenance strategies by a conditional maintenance strategy to reduce the downtime for maintenance, as well as the number and the cost of interventions. Although many physical models have been developed for the design control of WTs, fault diagnosis and prognosis methods based on physical models such as observer-based methods, parity space, or analytical redundancy are not widely used compared to data-driven approaches because they require system observability, a condition often unverified because of the cost of the required instrumentation and the difficulty of deploying some sensors on the system. Data-driven methods are the most used in this field, but the cost of maintenance and the high number of downtimes for maintenance show the limitations of these approaches, related to the facts that the databases available for learning are incomplete and do not cover all