To further reduce the levelised cost of energy (LCOE), increasing the rated power of a single wind turbine is still a common and effective approach that is actively pursued by the wind energy industry. When developing larger wind turbines, it is of key interest to maximise the efficiency of wind energy utilisation. This efficiency, however, is not a single and constant parameter. Wind turbines operate in a wide working range that at most times deviates from the rated power and speed. Moreover, wind turbines are constantly subjected to stochastic wind conditions that directly or indirectly influence efficiency. The efficiency property of a wind turbine therefore has to be determined across the entire working range and under different conditions. The determined efficiency property provides an important basis for turbine optimisation.

The overall efficiency of the wind turbine consists of the aerodynamic efficiency of the rotor and the efficiency of the drivetrain. The drivetrain efficiency is affected by a number of factors that have to be considered in the design of the turbine. These factors include the setting and functionality of the cooling system, the structural deformation (especially the air gap change for direct drive turbines) due to external loads and temperature change, as well as control strategies of the generator and converter. In order to validate and optimise the turbine design, the influence of these factors on the drivetrain's efficiency needs to be determined both qualitatively and quantitatively. To do so, it is necessary to determine the efficiency with a high level of accuracy and with measurements traceable to national standards according to metrological rules . Traceability to national standards and the accuracy and precision defined by the calibration, which is a comparison of the measurements on hand with national standards, are of great importance for optimising efficiency. A performance comparison of different drivetrain components or working conditions is only meaningful if the measurement methods and conditions are sufficiently accurate to be able to compare the measurement results with each other.

The best place to determine the drivetrain efficiency of a wind turbine is on a nacelle test bench, where the design mechanical load cases and electrical grid conditions can be easily produced and replicated. This said, determining efficiency with sufficient accuracy is still very challenging even on nacelle test benches. Major reasons for this are the lack of traceability to national standards in both the mechanical and the electrical fields. Metrological traceability is defined as the “property of a measurement result whereby the result can be related to a reference through a documented unbroken chain of calibrations, each contributing to the measurement uncertainty” . It ensures the accuracy of a measurement described with the measurement uncertainty that is given by the calibration chain. On nacelle test benches, calibration of the torque measurement is especially challenging due to the high torque levels and the absence of torque standards above 1.1 MN ⋅ m . A few approaches have been suggested to avoid the need for torque measurement when determining efficiency on nacelle test benches, including the calorimeter method and the modified back-to-back method . Nevertheless, the most reliably reproducible method of efficiency determination that is traceable to national standards is still the “direct” method, i.e. measuring the input and output power directly using state-of-the-art equipment. In this case, a 5 MN ⋅ m torque transducer including a device for rotational speed measurement owned by Physikalisch-Technische Bundesanstalt (PTB, the German National Metrology Institute) was used to trace the mechanical power measurement. Additionally, a reference power measuring system (RPMS) calibrated by the Swiss National Metrology Institute METAS together with the PTB was used to trace the electrical power measurement to national standards.

The WindEFCY project provided the opportunity to determine the efficiency of a wind turbine drivetrain with traceable measurements of both the mechanical input power and the electrical output power. The wind turbine drivetrain was tested on the 10 MW DyNaLab nacelle test bench of Fraunhofer IWES in Bremerhaven, Germany. During the test campaign, the 5 MN ⋅ m torque transducer (also known as the torque transfer standard, TTS) and other mechanical and electrical sensors were used to produce traceable measurements of the input and output powers. The TTS is linked to a primary national torque standard via calibration and transfers the highly accurate and precise torque measurement into industrial applications; here it serves as a reference standard to ensure the accuracy and traceability in torque measurement of the test bench's own torque transducer.

The WindEFCY project, officially titled “Traceable mechanical and electrical power measurement for efficiency determination of wind turbines”, was a collaborative research initiative across disciplines, such as mechanical and electrical power measurement and wind turbine test bench operation, under the European Metrology Programme for Innovation and Research (EMPIR). The aim of the project was the development and validation of standardised test methods for the efficiency determination of wind turbine drivetrains and their components on test benches in a reliable, reproducible and comparable way for quality assurance. This required several calibration processes and adequate standards. In a calibration process, the unknown measuring instrument is compared with a known standard provided by national metrology institutes, and the accuracy of the unknown measuring instrument is specified within certain tolerances, the measurement uncertainty (MU). Without traceability of the measurands, the accuracy and precision of the measurands are neither known nor reliable. Consequently, the measured quantities cannot be compared with information in data sheets or measurement data acquired by other measuring equipment or in other test benches. To optimise the efficiency of wind turbine drivetrains, where already a good efficiency is achieved, measurements with high accuracy and precision are essential. Within the WindEFCY project, it was shown that – especially in the field of torque measurement in the mega-newton metre range – traceability is obligatory, as the torque measurement discrepancy in test benches was ±5 %. For discrepancies of this magnitude, it is important that the calibration not only determines the MU, i.e. the range within which the true value of the measured quantity is likely to lie, but also an adjustment to the torque measurement is made to align it with the standard.

The efficiency behaviour of the turbine drivetrain was determined on numerous working points at 5 MN ⋅ m and under different conditions. In addition to aiding the efficiency determination, the availability of the 5 MN ⋅ m TTS also offered a chance to calibrate the test bench's own torque transducer, which is instrumented on the shaft adapter connecting the test bench and the device under test (DUT), as shown in Fig. . For the calibration, a number of calibration profiles were also carried out during the test campaign. Since the 5 MN ⋅ m TTS is not designed for high levels of non-torque loads (also known as parasitic loads in some publications), the calibrated test bench transducer could be used instead in future tests with high non-torque loads.

The layout of the complete test setup is depicted schematically in Fig. . Two motors of the test bench are connected in tandem to provide the driving torque. A load application unit (LAU) can be used to generate the designed non-torque loads with a hexapod driven by hydraulics. A coupling is placed between the LAU and the motors to prevent non-torque loads being transferred backwards to the motors. The non-torque loads are transferred via a main bearing from the hexapod to the output shaft that also carries the torque. A combination of loads in 6 degrees of freedom can be applied to the DUT through the flange of the output shaft. To connect the test bench with the DUT, a shaft adapter is used to fit the flanges on both sides. This adapter (yellow in Fig. ) is also used as a robust way to measure loads, i.e. torque, bending moments, and axial and lateral forces, directly in front of the DUT. For the WindEFCY test campaign, the 5 MN ⋅ m TTS from PTB (red in Fig. ) was additionally integrated into the setup between the shaft adapter and the DUT with the help of specially designed adaptation structures. To protect the 5 MN ⋅ m TTS, the applied non-torque load was controlled to the minimum during the tests.

The non-rotating part of the DUT is fixed to the base of the test bench. The DUT's generator is electrically connected to a full power converter, which is in turn connected to the transformer. Via a switch gear, the transformer is connected to the medium voltage inside a junction box. For efficiency determination, the mechanical power is measured by the 5 MN ⋅ m TTS, and the electrical power is measured by the electrical power measurement system (EPMS) in the junction box. The efficiency is then determined for all the components in between, including the generator, the converter and the transformer.

The mechanical input power Pmech of the DUT is a function of the input torque M and the rotational speed n at the interface of the DUT, as shown in Eq. (). It is very important that the torque and rotational speed being measured at the same position as the power is calculated from both of them. To this end, the 5 MN ⋅ m TTS was augmented by an inclinometer to measure rotational speed. In this chapter, important details of both the torque and the speed measurement are presented. 1Pmech=M⋅n

Efficiency is the ratio of useful electrical output power Pelec converted from the available mechanical input power. The electrical output power is the sum of the power of all three phases P1–3, whereas the power per phase P is determined as the product of transient current i1–3 and voltage u1–3: 8Pelec=Pelec,phase1+Pelec,phase2+Pelec,phase3,9Pelec,phase 1–3=1T∫0Tu1–3(t)⋅i1–3(t)dt.

Since the electrical output power is intended to be fed to the electricity grid, only the electrical power at grid frequency is useful. Power quality phenomena such as other spectral components are relevant since they influence, for instance, the stability of the grid. These are also studied, but since they are not considered “useful output” for determining efficiency, the uncertainty requirements are much less stringent. The setup of the electrical gauges in the DyNaLab nacelle test bench is shown in simplified form in Fig. .

In nacelle test benches, electrical power is usually measured using an electrical power measurement system (EPMS) integrated into the test bench’s data acquisition (DAQ) system. This system is optimised for convenience and versatility, not for minimum uncertainty. Since it is integrated into the nacelle test bench, sending this system to calibration laboratories is difficult and time-consuming. For this reason, METAS and PTB calibrated a reference power measuring system (RPMS) for use in test benches. This system is used to determine the efficiency and to calibrate the EPMS measurement chain of the test bench on-site.

The RPMS is based on commercial off-the-shelf components such as the LMG671 power analyser and the DL 2000ID current sensors (Fig. ). As the planned reference voltage divider HST12-3 could not be used due to the risk of over-voltage and to test hall safety regulations, the more robust HILO voltage dividers were employed instead, and an extensive recalibration of the HILO sensors had to be carried out at PTB. The over-voltage risk stemmed from the fact that the DUT had to be connected to the medium-voltage grid of the local grid operator instead of to the grid simulator of the test bench. This had the disadvantage that in the event of a fault in the medium-voltage grid, high over-voltages could occur due to ground faults or lightning strikes.

The power analyser was used for measurements at the primary side of the transformer. The power analysers were calibrated using a modular primary power standard . Depending on the currents and voltages to be measured, transducers can be used to reduce the currents and voltages to levels that can be measured with the power analyser. These reference transducers are calibrated with an uncertainty of 300 µV V⁻¹ and 30 µA A⁻¹ at power frequency.

Figure  shows the current measurement chain of the EPMS, consisting of current transformers of type DS 2000 ICLA, burden resistors of type HBR1.0, and the HBM data logger GEN4tB with the current measurement card GN8103B, which could be calibrated on-site. The MU for the measurement chain is 0.01 % and is valid for various load cases. A statement about the long-term stability cannot be made here.

Due to the over-voltage risk mentioned above, calibration of the HILO sensors on-site was not possible, as the over-voltages could have damaged the voltage reference sensor of the RPMS. To calibrate the HILO voltage sensors, the entire measurement chain, consisting of voltage divider, connection cable, transmitter, fibre optic cable and receiver (Fig. ), was shipped to and calibrated in the laboratory at PTB following the test campaign. The calibration was performed as a comparison measurement against a PTB standard. Due to a high position dependence of the voltage divider on other voltage dividers, the calibration resulted in an expanded MU (k=2) of 0.8 % at power frequency, with the standard uncertainty being 0.4 %.

As shown in Table , the uncertainty in the voltage measurement U1–3 plays a dominant role in the overall MU of the electrical power. Thanks to the state-of-the-art sensors and measurement system, the current I1–3 can be measured with an extremely small uncertainty. Additionally, since the power factor λ1–3 of the turbine is kept at 1 during the test, the uncertainty in λ due to the phase errors of voltage and current sensors is negligible. The uncertainties of the three voltage measurements for the three phases are regarded as independent after the calibration and correction. As a result, the total power of the three phases has a smaller relative uncertainty compared to the power P1–3 of each individual phase.

For the determination of efficiency, the electrical power and mechanical power measurements need to be synchronised. While the mechanical power measurement system is synchronised to UTC (coordinated universal time) using the IRIG-B signal, the chosen RPMS model cannot be synchronised to an external time reference other than by manually setting the time like on a wristwatch. Mechanical imperfections of the nacelle cause a pattern of mechanical power with a period of one revolution. Since the electrical power shows the same pattern, the electrical power measurement is synchronised in a post-processing stage using the cross correlation of the two power measurements.

During the test campaign, numerous tests serving various purposes were carried out. Because the measurement range of the TTS (being 5 MN ⋅ m) is smaller than the rated torque of the DUT, all tests were carried out with the turbine operating below the rated power. The results of two tests are presented in this paper to demonstrate the method of efficiency determination. In both tests, the torque was held stable around the 5 MN ⋅ m level to utilise the maximum capacity of the transducer. In the first test, the rotational speed was kept constant to be able to analyse the long-term behaviour of the DUT and the temperature influence on the efficiency, while in the second test the rotational speed followed an operational curve in a stepwise manner upwards. For each test, the mechanical input power and the electrical output power were calculated to determine the efficiency. The uncertainty analysis was carried out for the first test with uncertainty budgets of the raw measurements determined according to the propagation principle. The turbine under test was a customer device in development mode. Since the rated rotational speed is a key parameter for a commercial wind turbine, it was agreed to be normalised, which consequently leads to the normalisation of the mechanical and electrical powers in this paper.

The efficiency determination for both of the tests discussed above achieved an accuracy of approximately 0.7 % uncertainty, thereby breaking the 1 % mark. The largest uncertainty contribution still comes from the torque measurement, despite the use of the best possible torque transducer and calibration machine. To further reduce the uncertainty, the transducer needs to be calibrated to a higher level of torque. PTB is commissioning a new torque calibration machine with a capacity of 5 MN ⋅ m, and this could help achieve better uncertainty.

The second largest contribution comes from the voltage measurement. Although in this case it stems from safety regulation requirements and could in particular circumstances be solved by a dedicated reconfiguration of the test bench, it still shows the importance of planning effort and investment in electrical power measurement. In practice, it should not be taken for granted that electrical power can be measured automatically with sufficient accuracy. Because testing time on a nacelle test bench is a limited resource (drivetrain efficiency would very likely be tested together with many other test items), it is not always possible to reconfigure the test layout just for one test. Therefore, it is important to plan the test in advance and take all relevant factors into consideration in order to achieve the best possible electrical measurement accuracy.

Rotational speed and electrical current measurements achieved very high levels of accuracy. For these two cases, suitable sensors with careful calibration were instrumented at the right positions on the drivetrain and integrated into well-calibrated measurement chains. All these factors combined produce satisfying results.

Owing to a number of discussed reasons, ripples on the measurement and deviations in the determined efficiency are inevitable. To achieve stable efficiency under certain conditions, it is recommended to average at least six full revolutions of measurement. Based on the results of this study, the uncertainty caused by the deviation in efficiency will only represent a minor contributor to the overall MU of efficiency if this recommendation is followed.

One limitation of the test layout presented in this paper is that the non-torque loads, such as bending moments and shear forces, could not be applied to the DUT because the 5 MN ⋅ m TTS from PTB is not designed to withstand high levels of non-torque loads. To overcome this limitation, a series of calibration profiles was carried out during the test campaign so that the DyNaLab transducer developed in-house at Fraunhofer IWES could be calibrated. This transducer was placed directly in front of the reference transducer from PTB. The torque calibration of this transducer has been reported by . This transducer is designed to withstand and measure loads in all 6 degrees of freedom and will be used for torque measurement in future test campaigns.

This paper reported an approach to determine the drivetrain efficiency of a modern multi-megawatt wind turbine drivetrain. Addressing the challenge of measurement accuracy, state-of-the-art sensors, measurement systems and calibration facilities were employed within the framework of the WindEFCY project. The results show that an overall MU of 0.7 % is achievable for an efficiency determination with torque measurement up to 5 MN ⋅ m. As expected, torque measurement contributed the largest share to the overall MU. Surprisingly, the electrical power measurement also played a significant role in the MU. This highlights the fact that although electrical measurements are generally considered to be much more accurate, equal care must be exercised when measuring both electrical and mechanical power. Speed measurement using an inclinometer, however, yielded very good results including a very small MU of 0.01 %. To achieve a stable efficiency result, the measurement of at least six full revolutions was averaged, resulting in nearly negligible contributions to the overall MU.