Nikolaos Dervilis

A Performance Monitoring Approach for the Novel Lillgrund Offshore Wind Farm

The use of offshore wind farms has been growing in recent years. Europe is presenting a geometrically growing interest in exploring and investing in such offshore power plants as the continents water sites offer impressive wind conditions. Moreover, as human activities tend to complicate the construction of land wind farms, offshore locations, which can be found more easily near densely populated areas, can be seen as an attractive choice. However, the cost of an offshore wind farm is relatively high, and therefore, their reliability is crucial if they ever need to be fully integrated into the energy arena. This paper presents an analysis of supervisory control and data acquisition (SCADA) extracts from the Lillgrund offshore wind farm for the purposes of monitoring. An advanced and robust machine-learning approach is applied, in order to produce individual and population-based power curves and then predict measurements of the power produced from each wind turbine (WT) from the measurements of the other WTs in the farm. Control charts with robust thresholds calculated from extreme value statistics are successfully applied for the monitoring of the turbines.

Aspects of structural health and condition monitoring of offshore wind turbines

Wind power has expanded significantly over the past years, although reliability of wind turbine systems, especially of offshore wind turbines, has been many times unsatisfactory in the past. Wind turbine failures are equivalent to crucial financial losses. Therefore, creating and applying strategies that improve the reliability of their components is important for a successful implementation of such systems. Structural health monitoring (SHM) addresses these problems through the monitoring of parameters indicative of the state of the structure examined. Condition monitoring (CM), on the other hand, can be seen as a specialized area of the SHM community that aims at damage detection of, particularly, rotating machinery. The paper is divided into two parts: in the first part, advanced signal processing and machine learning methods are discussed for SHM and CM on wind turbine gearbox and blade damage detection examples. In the second part, an initial exploration of supervisor control and data acquisition systems data of an offshore wind farm is presented, and data-driven approaches are proposed for detecting abnormal behaviour of wind turbines. It is shown that the advanced signal processing methods discussed are effective and that it is important to adopt these SHM strategies in the wind energy sector.

Novelty detection applied to vibration data from a CX-100 wind turbine blade under fatigue loading

The remarkable evolution of new generation wind turbines has led to a dramatic increase of wind turbine blade size. In turn, a reliable structural health monitoring (SHM) system will be a key factor for the successful implementation of such systems. Detection of damage at an early stage is a crucial issue as blade failure would be a catastrophic result for the entire wind turbine. In this study the SHM analysis will be based on experimental measurements of Frequency Response Functions (FRFs) extracted by using an input/output acquisition technique under a fatigue loading of a 9m CX-100 blade at the National Renewable Energy Laboratory (NREL) and National Wind Technology Center (NWTC) performed in the Los Alamos National Laboratory. The blade was harmonically excited at its first natural frequency using a Universal Resonant Excitation (UREX) system. For analysis, the Auto-Associative Neural Network (AANN) is a non-parametric method where a set of damage sensitive features gathered from the measured structure are used to train a network that acts as a novelty detector. This traditionally has a highly complex bottleneck structure with five layers in the AANN. In the current paper, a new attempt is also exploited based on an AANN with one hidden layer in order to reduce the theoretical and computational difficulties. Damage detection of composite bodies of blades is a grand challenge due to varying aerodynamic and gravitational loads and environmental conditions. A study of the noise tolerant capability of the AANN which is associated to its generalisation capacity is addressed. It will be shown that vibration response data combined with AANNs is a robust and powerful tool, offering novelty detection even when operational and environmental variations are present. The AANN is a method which has not yet been widely used in the structural health monitoring of composite blades.

Exploring Environmental and Operational Variations in SHM Data Using Heteroscedastic Gaussian Processes

The higher levels of Structural Health Monitoring (SHM)—localisation, classification, severity assessment—are only accessible using supervised learning in the data-based approach. Unfortunately, one does not often have data from damaged structures; this forces a dependence on unsupervised learning i.e. novelty detection. This means that detection is sensitive to benign environmental and operational variations (EOVs) in or around the structure. In this paper a two-stage procedure is presented: identify EOVs in training data using a nonlinear manifold approach and remove EOVs by utilising the interesting tool of heteroscedastic Gaussian processes (GPs). In Classical GPs models, the data noise is assumed to have constant variance throughout the input space. This assumption is a drawback most of the time, and a more robust Bayesian regression tool where GP inference is tractable is needed. In this work a combination of data projection and a non-standard heteroscedastic GP is presented as means of visualising and exploring SHM data.

Machine Learning Applications for a Wind Turbine Blade under Continuous Fatigue Loading

Structural health monitoring (SHM) systems will be one of the leading factors in the successful establishment of wind turbines in the energy arena. Detection of damage at an early stage is a vital issue as blade failure would be a catastrophic result for the entire wind turbine. In this study the SHM analysis will be based on experimental measurements of vibration analysis, extracted of a 9m CX-100 blade under fatigue loading. For analysis, machine learning techniques utilised for failure detection of wind turbine blades will be applied, like non-linear Neural Networks, including Auto-Associative Neural Network (AANN) and Radial Basis Function (RBF) networks models.

On damage detection in wind turbine gearboxes using outlier analysis

The proportion of worldwide installed wind power in power systems increases over the years as a result of the steadily growing interest in renewable energy sources. Still, the advantages offered by the use of wind power are overshadowed by the high operational and maintenance costs, resulting in the low competitiveness of wind power in the energy market. In order to reduce the costs of corrective maintenance, the application of condition monitoring to gearboxes becomes highly important, since gearboxes are among the wind turbine components with the most frequent failure observations. While condition monitoring of gearboxes in general is common practice, with various methods having been developed over the last few decades, wind turbine gearbox condition monitoring faces a major challenge: the detection of faults under the time-varying load conditions prevailing in wind turbine systems. Classical time and frequency domain methods fail to detect faults under variable load conditions, due to the temporary effect that these faults have on vibration signals. This paper uses the statistical discipline of outlier analysis for the damage detection of gearbox tooth faults. A simplified two-degree-of-freedom gearbox model considering nonlinear backlash, time-periodic mesh stiffness and static transmission error, simulates the vibration signals to be analysed. Local stiffness reduction is used for the simulation of tooth faults and statistical processes determine the existence of intermittencies. The lowest level of fault detection, the threshold value, is considered and the Mahalanobis squared-distance is calculated for the novelty detection problem.

Automatic Kernel Selection for Gaussian Processes Regression with Approximate Bayesian Computation and Sequential Monte Carlo

The current work introduces a novel combination of two Bayesian tools, Gaussian Processes (GPs) and the use of the Approximate Bayesian Computation (ABC) algorithm for kernel selection and parameter estimation for machine learning applications. The combined methodology that this research paper proposes and investigates offers the possibility to use different metrics and summary statistics of the kernels used for Bayesian regression. The presented work moves a step towards online, robust, consistent and automated mechanism to formulate optimal kernels (or even mean functions) and their hyperparameters simultaneously offering confidence evaluation when these tools are used for mathematical or engineering problems such as structural health monitoring (SHM) or system identification (SI).

An SHM View of a CFD Model of Lillgrund Wind Farm

Reliability of offshore wind farms is one of the key areas for the successful implementation of these renewable power plants in the energy arena. Failure of the wind turbine (WT) in general could cause massive financial losses but especially for structures that are operating in offshore sites. Structural Health Monitoring (SHM) of WTs is essential in order to ensure not only structural safety but also avoidance of overdesign of components that could lead to economic and structural inefficiency. A preliminary analysis of a machine learning approach in the context of WT SHM is presented here; it is based on results from a Computational Fluid Dynamics (CFD) model of Lillgrund Wind farm. The analysis is based on neural network regression and is used to predict the measurement of each WT from the measurements of other WTs in the farm. Regression model error is used as an index of abnormal response.

Envelope Analysis Using the Teager-Kaiser Energy Operator for Condition Monitoring of a Wind Turbine Bearing

Different signal processing methods are applied to experimental data obtained from a rolling element bearing rig in order to perform damage detection. Among these methods the Teager-Kaiser energy operator is also proposed as a more novel approach. This energy operator is an amplitude-frequency demodulation method used in this paper as an alternative to the Hilbert Transform in order to perform envelope analysis on the datasets analysed.

Comparative Study of Robust Novelty Detection Techniques

The central target of this work is to provide an alternative to machine learning approaches to structural health monitoring with one of robust multivariate statistic novelty detection. Damage detection and identification is a procedure that is hierarchical in nature. At its most sophisticated, diagnosis of the damage could include localisation, classification and severity assessment and even go so far as to estimate the time-to-failure of the structure. In this paper, robust multivariate statistics were investigated focused mainly on a high level estimation of the outliers which determines only the presence or absence of novelty - something that is of fundamental interest. These methods allow a diagnosis of deviation from normality and the option of identifying the presence of masking effects caused by multiple outliers. This paper is trying to introduce a new scheme for damage detection by adopting simple measurements and exploiting robust multivariate statistics.