Data driven fault detection and isolation: a wind turbine scenario

Abstract

One of the greatest drawbacks in wind energy generation is the high maintenance cost associated to mechanical faults. This problem becomes more evident in utility scale wind turbines, where the increased size and nominal capacity comes with additional problems associated with structural vibrations and aeroelastic effects in the blades. Due to the increased operation capability, it is imperative to detect system degradation and faults in an efficient manner, maintaining system integrity, reliability and reducing operation costs. This paper presents a comprehensive comparison of four different Fault Detection and Isolation (FDI) filters based on “Data Driven” (DD) techniques. In order to enhance FDI performance, a multi-level strategy is used where: the first level detects the occurrence of any given fault (detection), while the second identifies the source of the fault (isolation). Four different DD classification techniques (namely Support Vector Machines, Artificial Neural Networks, K Nearest Neighbors and Gaussian Mixture Models) were studied and compared for each of the proposed classification levels. The best strategy at each level could be selected to build the final data driven FDI system. The performance of the proposed scheme is evaluated on a benchmark model of a commercial wind turbine.