Damage detection in operational wind turbine blades using a new approach based on machine learning

Abstract

The application of reliable structural health monitoring (SHM) technologies to operational wind turbine blades is a challenging task, due to the uncertain nature of the environments they operate in. In this paper, a novel SHM methodology, which uses Gaussian Processes (GPs) is proposed. The methodology takes advantage of the fact that the blades on a turbine are nominally identical in structural properties and encounter the same environmental and operational variables (EOVs). The properties of interest are the first edgewise frequencies of the blades. The GPs are used to predict the edge frequencies of one blade given that of another, after these relationships between the pairs of blades have been learned when the blades are in a healthy state. In using this approach, the proposed SHM methodology is able to identify when the blades start behaving differently from one another over time. To validate the concept, the proposed SHM system is applied to real onshore wind turbine blade data, where some form of damage was known to have taken place. X-bar control chart analysis of the residual errors between the GP predictions and actual frequencies show that the system successfully identified early onset of damage as early as six months before it was identified and remedied.

Statistical Structural Integrity Control of Composite Structures Based on an Automatic Operational Modal Analysis — a Review

“…The control chart is constructed by drawing two lines: the center line (CL) and an additional horizontal line corresponding to the upper limit (UCL), which are given by [68] CL (32) UCL CL X (33) where ψ is the value or variable on whose basis the control chart is constructed, and the overbar denotes the mean value. The subscript X denotes the X -bar chart, and ψ can be either the mean value of a DSF vector [56], the mean value of some similarity measure, for example, MD [68], or even the mean value of standard deviations of a DSFs σ (for S-control charts) [56]; α can be taken equal to 3, corresponding to the 99.7% confidence [68,93]. In [94], it is assumed that 1 4 5 : .…”

Operational Modal Analysis for Scour Detection in Mono-Pile Offshore Wind Turbines

“…Using similar notation as shown in [12], considering the noise present and assuming that the DSF vector used for training is a function of the EOV, f tr can be expressed in the form:…”

Certification Approach for Physics Informed Machine Learning and its Application in Landing Gear Life Assessment

“…• SHM and neural network data is taken from sensors placed at the points of interest, which in turn generates costs for the sensor's data storage in addition to the many sensors required themselves [14]. • The robustness of the system formulated to diagnose and regulate the data must be vigorous with low error rates, due to the nature of conditions faced by the aircraft's landing gear, in which landing strips constantly change due to the wind conditions and control surface changes.…”

“…• The robustness of the system formulated to diagnose and regulate the data must be vigorous with low error rates, due to the nature of conditions faced by the aircraft's landing gear, in which landing strips constantly change due to the wind conditions and control surface changes. • The method used for diagnosis must include analyses of cost-benefit scenarios due to the costs associated with downtime [14] The machine learning method used by Holmes et al, namely Gaussian Progress regression, to calculate loads on landing gear, resorted to the input sourced from sensors being attached to the landing gear of a singular aircraft while enveloping multiple surfaces of runways the aircraft interacts with, taken from a landing gear testing rig [15]. Concluding with the requirement of aircraft-specific model training, the machine learning model used would adapt to different landing conditions without the need for filtering of the data used for input.…”