No abstract

No abstract

Advanced Fault Detection (FD), diagnosis, and isolation schemes are necessary to realize the required levels of reliability and availability and to minimize financial losses against failures, explicitly in modern Wind Turbine Systems (WTSs) which are designed to generate electrical energy as efficiently and reliably as possible. This paper presents the design of a novel and inclusive FD framework on the basis of data-driven methods, it is intended for the online early detection of involuntary anomalies of various types and locations. Conventional methods are based on the exact model and/or signal patterns or hardware redundancy, and generally fail to address this issue. On the contrary, the presented algorithm is motivated by the availability of fast sensors and powerful computers yielding big data which can be explored to extract and exploit useful information. In a typical WTS, FD procedures face particular challenges attributed to high levels of measurement noise and sparse changes due to the fast dynamics as well as the switching control strategy along with transient phases. In this scope, a minimum and informative set of measured variables is proposed to accurately and completely describe the system behavior under its entire operation range; datasets are hence defined for appropriate dimensions and sampling time to satisfactorily model the measured data patterns and variability. Among data-based strategies, multivariate and univariate statistical analysis tools are recommended for this approach. A method based on Principal Component Analysis (PCA) is used in this study for its capabilities of dimensionality reduction, information de-correlation, and noise rejection. Multi-PCA-models are consequently used to train parallel statistical models through the proposed set of fault-relevant variables and used for online FD. Moreover, this method is reinforced by an adaptive threshold scheme based on an efficient modified EWMA control chart, the overall algorithm is robust to outliers and highly sensitive to small abnormalities and abrupt changes. Static and dynamic applications are investigated for modern wind turbine systems under different operating zones, and faults having different severity levels are studied ranging from sensors to actuators and up to system faults. The overall constructed algorithm shows significant potential applications compared to several methods recently reported in the literature in terms of applicability potential, robustness, and detection sensitivity.

No abstract