Validation of data-driven fault diagnosis strategies for a wind turbine test rig

The fault diagnosis of safety critical systems, such as a wind turbine plant, represents a difficult issue, especially for offshore installations, which motivates the investigations addressed in this work. In fact, these systems should be able to maintain specified operable and committable conditions, and at the same time, should avoid expensive unplanned maintenance works. Therefore, this paper considers this problem and develops a data–driven fault diagnosis approach that is verified on a wind turbine high–fidelity test-rig. In particular, the proposed design derives nonlinear filters that provide the estimation of the fault by using the input–output data acquired from the monitored process. Moreover, the proposed approach represents an effective method for managing data affected by measurement errors, disturbance and model–reality mismatch. In more detail, the developed strategies exploit fuzzy systems and neural networks, which are able to derive the nonlinear dynamic functions between input–output measurements and faults. Moreover, these dynamic nonlinear structures represented by fuzzy prototypes include autoregressive with exogenous input structures, with the ability to approximate any nonlinear dynamic functions with arbitrary degree of accuracy. The capabilities of the developed fault estimators thus exploited for monitoring and fault diagnosis purpose are verified using a wind turbine test–rig, which allows also to analyse their robustness and reliability features. In fact, this test–bed relies on a hardware–in–the–loop technique that is able to take into account uncertainty and disturbance, thus emulating a very realistic environment.