Robust fault diagnosis of wind turbines based on MANFIS and zonotopic observers

Wind turbines have become one of the essential sources of energy generation due to their contribution to energy security, economic development, job creation, and technological innovation. This work proposes a methodology for designing robust fault diagnosis systems based on a bank of zonotopic state estimators built upon Takagi–Sugeno (TS) models. The TS models with associated parametric uncertainty are obtained using a Multiple Output Adaptive Neuro-fuzzy Inference System (MANFIS), an extended and improved version of single-input, single-output ANFIS. Its main difference is its multi-output architecture, which allows generalized weighting functions to be obtained, reducing training times, uncertainties estimation, and reduced complexity. As a result, a set of Linear Matrix Inequalities is obtained with the H∞ criterion to adjust the parameter of the zonotopic estimator considering the modeling uncertainty. Overall, the work contributes to improving the safety of WT through diagnostic methods that improve its operability. A well-known certified reference case study of a wind turbine system is considered to validate the proposed method. •A Takagi–Sugeno (TS) zonotopic observer scheme for fault diagnosis is proposed.•The uncertain TS models are identified with a MANFIS using healthy data.•A set of TS zonotopic observers are designed to detect faults in the wind turbine.•A fault isolation module is implemented through fault signal matrices.•The methodology is implemented in a certified FAST wind turbine of 5 [MW].