Inertial measurements monitor wind turbines in action

Reliability is a significant issue for wind turbines, with estimates that per year, each one is down for 52–237 hours and experiences an average of 0.4–2.38 failures.¹ Some of these are infancy failures due to problems in manufacturing, quality control, or installation. However, transient and unpredictable loading from turbulence and gusts cause other failures. Therefore, researchers are developing operational monitoring systems to estimate the wind loading applied to the rotor blades. This information can be used to track turbine component conditions to improve performance. Such methods could also be used to detect damage and attenuate loading in future control systems.

There are two approaches to monitoring wind loading: aerodynamic methods (such as Pitot tubes, pressure taps, and light detection and ranging to get surface flow conditions) and structural ones (such as strain gages, fiber Bragg gratings, and accelerometers). Aerodynamic systems directly measure the wind flow conditions and infer the structural response, while structural monitoring infers the former from the latter. Currently, it is unclear which approach is superior because both must infer information they do not directly measure (although Fritzen² provided an overview of load estimation methods). We focused on developing an accelerometer-based structural monitoring system for operational wind turbine rotor blades.

Figure 1. A quasi-static acceleration signal in a turbine rotor blade.

Accelerometers mounted to turbine rotor blades are useful because they can directly estimate the forces used in the blade dynamic equations of motion. (In other words, force is proportional to the product of mass and inertia in Newtonian physics.) The acceleration signals are a summation of several sources of forced response, such as centripetal and gravitational acceleration, turbulence, yaw misalignment, coning, tilt, and tower motion. Each type produces a characteristic waveform that the sensor measures. Understanding how each of these sources affects the measured signal is critical for load estimation. For example, turbine blade rotation causes centripetal acceleration, which appears in the measured signal as a constant acceleration in the flap, lead-lag, and span directions when rotor coning and pitch are included (see Figure 1). Another example, gravitational acceleration (see Figure 2), appears in the signal as a once-per-revolution oscillation in the same directions when we include rotor tilt, coning, pitch, and rotation. The ultimate goal is to estimate individual loads by using spectral and spatial filters to decouple all sources.

Figure 2. A sensor measures an alternating gravitational acceleration field.

Figure 3. Centripetal acceleration in (a) undeformed and (b) deformed rotor blade. R: Radius. ω: Angular speed. θ: Angular position. Span-wise static acceleration. Flap-wise static acceleration.

Figure 4. (a) Two triaxial and one uniaxial accelerometers mounted near shear web termination at the 8m station during blade fabrication. (b) A sensored rotor blade (white box) mounted on a Micon 65/13 wind turbine at USDA in Bushland, TX prior to operational monitoring.

We are investigating online monitoring methods to estimate static and dynamic structural loading as well as to detect damage. Quasi-static loading to the turbine blade is caused by the mean wind, which we assume changes gradually over several minutes. This loading, when applied to the blade, deflects it downwind. Though the centripetal acceleration appears in the measured signal of an undeformed rotor blade as a function of the coning angle, the quasi-static loading increasingly deflects it. This process rotates the blade, leading the sensor to measure decreased span-wise and increased flap-wise acceleration (see Figure 3). We use this phenomenon to determine the quasi-static deformation and loading using optimized generalized estimators.³

Dynamic loading to the rotor blade is caused by turbulence and gusts that cause a spectrum of forced response component excitations. The higher frequency ones excite the rotor blade's modal dynamics. To estimate the dynamic contributions of each mode to the overall response, we use a spatial filter that relies on measured response signals and structural mode shapes to estimate the contributions of individual ones as a function of time. These contributions can track the magnitude and cycles of excitation of each mode throughout the turbine's life. They can also serve as control observers for attenuating damaging transient loads in next-generation smart turbines.

We have implemented these approaches on a 9m CX-100 rotor blade for the Sandia National Laboratories sensor blade project. In this work, we equipped the blade with seven triaxial and three uniaxial accelerometers during blade fabrication (see Figure 4). We tested it on the ground and then mounted it on a Micon 65/13 wind turbine at the United States Department of Agriculture facility in Bushland, TX. We acquired operational data and are currently processing it to determine the performance of these methods. In the future, we will compare the operational data with computational models of the same rotor blade using the analysis programs FAST and MSC.ADAMS for validation.