An approach for robust data-driven fault detection with industrial application

Abstract

This paper introduces a robust data-driven fault detection method and its application on a wind turbine benchmark. The benchmark is provided by a Simulink Model, which contains nonlinear wind turbine model and complex wind disturbances. The model-based fault detection technique is hardly to be applied to solve this problem because modeling this wind turbine is quite difficult. Besides, the unknown wind disturbances and the large measurement noises are two enormous challenges for most of the fault detection techniques. To overcome these difficulties, this paper applies a robust data-driven fault detection scheme, which is based on a standard residual generation and decision logic structure. In the residual generation step, a robust residual generator with an optimal parity vector is constructed directly from the measurement data. Moreover, a filter algorithm is used in the residual evaluation step to reduce false alarms rate. Simulation results show that the performance and effectiveness of the proposed scheme are satisfied.