1. Introduction
In the future, the certification of wind turbines’ electrical characteristics will heavily depend on measurement results derived on test benches equipped with multi-physics Hardware-in-the-Loop (HiL) systems. Certification measurements on test benches will reduce, and eventually replace, resource-consuming field tests of wind turbines. Currently, Germany already approves the use of test benches for the certification of electrical characteristics by means of the technical guideline FGW TR3 [
1]. The technical specification IEC 61400-21-4, which complements the existing standard IEC 61400-21-1 [
2], will also make this option available internationally as of 2022 [
3]. Both standards allow for the carrying out of Fault Ride Through and Active and Reactive Power Control tests on test benches instead of in the field.
This paper focuses on nacelle test benches as depicted in
Figure 1. The setup of a nacelle test bench includes a complete, full-scale wind turbine nacelle with its mechanical and electrical drive train. Thereby, HiL systems operate at the wind turbine’s electrical, mechanical, and signal interfaces and enable the operation of the wind turbine as if in the field. While the tower and the rotor are missing, the electrical setup and the control software of the wind turbine remain unchanged.
HiL systems for nacelle test benches are an active field of research for both academia and the wind industry. Recent publications focus on design aspects of the HiL systems, ranging from different interface methods to control methods that couple a real-time model with the test bench. The effect of different interface algorithms for power-level HiL (PHiL) systems of a wind turbine test bench is investigated in [
4]. A similar investigation extended to PHiL simulations in general, and that was not restricted to wind turbine test benches, can be found in [
5]. A general method for the selection of an appropriate interface algorithm for PHiL applications is derived in [
6]. The authors of [
7] use a system identification method to update the knowledge of the hardware and, by this, improve the interface algorithm. To overcome challenges regarding the integration of distributed energy resources, the task-force paper [
8] proposes a benchmark system for controller HiL and PHiL testing. For mechanical-level HiL (MHiL), research is more focused on the control of wind turbine test benches. The authors of [
9] propose a proportional-integral controller for the emulation of missing inertia on a test bench. A similar approach for MHiL is introduced in [
10], but restricted to simulation results for validation. An emulation of rotor inertia and related eigenfrequencies was first introduced in [
11], where a state feedback controller shifts the eigenfrequencies of the test bench to the desired locations in the complex eigenvalue plane. The authors of [
12] use model-based control, namely an internal model control, and suggest a two-step procedure to design the HiL control algorithm, which also allows eigenfrequency emulation. In [
13], a model-based control method damps the test-bench-related eigenfrequencies and applies a reference trajectory. This control method is used to couple a mechanical real-time model for inertia-eigenfrequency emulation in [
14]. A similar approach can be found in [
15]. This approach controls the test bench’s speed control rather than the torque. These control methods have been extended to robust control with a genetic algorithm optimizing the weighting functions used within the controller synthesis in [
16], linear parameter varying control where a full-order dynamic output feedback controller adapts to the plant’s parameter-varying characteristics [
17], and nominal performance in [
18].
Most of the aforementioned designs are analyzed in simulation, and only a few were put into operation in full-scale experiments. Full-scale validations with a focus on experimental results are reported in [
19] for the initial commissioning of the test bench at the Center for Wind Power Drives at RWTH Aachen University, and in [
20] for the electrical faults on the same test bench. In [
21], an experimental validation of the rotor eigenfrequency emulation on a full scale test bench is demonstrated, in [
22] for mechanical results on the dynamometer at the Fraunhofer IWES, and in [
23] for electrical faults on the same test bench. Experimental results, with a focus on the grid emulator, of the wind turbine testing facility at NREL are given in [
24], and practical recommendations regarding the PHiL interface derived from these results are given in [
25].
Some of these publications even refer to certification as a possible use case [
20,
24,
26]. However, the specific impact HiL operated test benches can have on certification results has hardly been researched so far, as this is a topic on the fringes of academic interest.
However, the transfer from research results to industrially relevant technology and application requires discussing the impact and limitations that come along with HiL-operated test benches. Hence, sharing experiences made during test-bench-based certification campaigns seems mandatory now and in the future, in order to maintain the high quality of wind turbine certification. While certification on test benches becomes part of the standards, the publicly available data of, and experience with, certification measurement campaigns conducted at nacelle test benches are limited. To the authors’ knowledge, only the joint research project CertBench, which the authors were also part of, published such results at all [
27]. This project achieved the worldwide first wind turbine certification, solely based upon test bench measurements by means of the German FGW TR3 guideline. An overview of the results [
28], and some results on MHiL- [
29] and PHiL-specific [
30] aspects have been published already.
For the first time, this paper examines, in detail, the impact the HiL system has on certification results. In this way, we try to bridge the gap between the academic perspective on HiL systems and the certification perspective. For this, we investigate how HiL functionalities of a test bench affect the results of selected certification tests and if the observed limitations are significant. This is done by comparing results to field measurements with respect to the evaluation criteria stated in the standards. With that, rather than discussing details of the HiL systems, we focus on the results of certification-relevant measurements. Based on the findings, this paper also aims at deriving procedures which help to ensure the suitability of test benches with HiL systems for certification purpose. The basis for this work is a set of selected results of a measurement campaign carried out with an ENERCON E-115 E2 wind turbine as the device under test (DUT) at the Center for Wind Power Drives and in the field.
The structure of the paper is as follows:
Section 2 gives an overview of the general setup of a nacelle test bench before
Section 3 and
Section 4 introduce the PHiL and the MHiL systems, respectively, and discuss their accuracy in an isolated manner. The influence of the HiL systems on the tests Frequency Control, Synthetic Inertia, and Under Voltage Ride Through (UVRT) is discussed in
Section 5.
Section 6 presents initial ideas on possible validation procedures for HiL-operated test benches.
5. Analysis of the HiL Systems’ Impact on Selected Certification Tests
This chapter describes the HiL systems’ reaction during selected certification-relevant tests and discusses their influence on the measurement and, hence, the certification results. For this purpose, three different tests from IEC 61400-21-1 are considered. These are Under Voltage Ride Through (UVRT) events, Frequency Control, and Synthetic Inertia.
The UVRT event is one of the most important tests when it comes to certification and one of the main reasons for test-bench-based certification for the industry. The influence that the MHiL has on the measurement results is expected to be minor because the DUT is a full-scale power converter wind turbine, so the mechanical drive train and the electrical grid are not coupled directly. The impact of the PHiL system on the test results is expected to be more significant, as the emulated impedance determines how the grid voltage reacts on the wind turbine’s reactive current injection and, hence, the course of the test.
Beyond that, this chapter discusses Frequency Control and Synthetic Inertia tests. For both tests, the MHiL system can have a more significant influence on the measurement results because the tests include major and rapid active power changes. The PHiL system was not found to influence these test results in any form and is, therefore, not discussed for these tests.
5.1. Under Voltage Ride through Events
This section investigates the influence of the PHiL and the MHiL system, respectively, on the UVRT results in comparison to field measurements.
5.1.1. PHiL Impact on UVRT
Testing the UVRT capability of a wind turbine in the field usually requires a voltage-divider-based test device. Both IEC 61400-21-1 and FGW TR3 recognize the voltage-divider-based test equipment as state of the art and describe their functionality in more detail. During UVRT tests carried out with such a test device, the impedance at the PCC of the DUT changes with time [
38]. Before the actual fault is generated, a series impedance is switched into the circuit, while during fault generation an additional shunt impedance (also called short circuit impedance) is active. In order to show the extent to which the impedance control on the grid emulator enables the reproduction of state-of-the-art field results with a voltage-divider-based test device, an experimental series with two-phase UVRT tests is carried out. In a first measurement sequence, the real-time simulation of the PHiL setup maps the grid impedance and the series-connected impedance of the voltage-divider-based test equipment. In a repetition of the test sequence, the real-time simulation additionally includes the parallel connection of the shunt impedance. Subsequently, these measurements are compared with field data of an identical DUT type.
Figure 9 shows the comparison of the positive and negative sequence voltages determined from the field measurement and from the grid emulator running a PHiL simulation. The example measurement is a two-phase UVRT test to 0.25 pu residual voltage. For the results in the upper diagram, the PHiL simulation only includes the series impedance, while for the results in the lower diagram, it also simulates the shunt impedance. The exclusive representation of the series impedance in the PHiL simulation leads to a deviation of 4% in the positive sequence voltage. The negative sequence even shows a deviation of 8%. This deviation decreases to less than 1% if the PHiL simulation includes the shunt impedance in addition to the series impedance (
Figure 9 bottom).
In order to assess the performance of the DUT,
Figure 10 compares the reactive current injection for voltage support observed in the field and at the test bench. Again, the DUT shows identical characteristics on the test bench as in the field, provided that the real-time PHiL simulation maps the shunt impedance in addition to the series impedance of the test equipment. If only the series impedance is simulated, the deviation of the reactive current injected into the positive sequence system increases to 13%, while it is 10% in the negative sequence system. This example shows that by emulating the voltage-divider-based test method’s impedance in the PHiL simulation, the test bench is capable of reproducing field characteristics. This is even correct when the PHiL simulation does not emulate the switching of the impedances through the course of the test, but keeps the impedance constant at the impedance level required during fault generation, as well as before, during, and after the fault.
For a more comprehensive comparison over different operating points,
Figure 11 shows the calculated positive and negative sequence voltages for all tested voltage dip depths of the two-phase UVRT tests. Except for the dips <0.05 pu, all dip depths result in discrepancies of less than 1% between the field measurement and the test bench measurements that include series and short-circuit impedances in the PHiL simulation. For the <0.05 pu tests, a resulting voltage of 0.02 pu was set on the grid emulator, while this was 0 pu in the field. The different setting ultimately results in a 2% deviation. The differences between the measurements that include the shunt impedance and those that simulate only the series impedance is up to 0.15 pu in the positive sequence. This deviation decreases to 0.01 pu when the residual voltage increases to 0.73 pu. This is because of the decreasing discrepancy between the resulting impedances. The difference between the shunt and series impedances is 7.52 ohms for a residual voltage of 0.73 pu, while it is 28.55 ohms for a dip to 0 pu.
In conclusion, the DUT’s reactive current injection shows identical results for field and bench tests, as long as the real-time simulation of the grid impedance simulates both the series and shunt impedances of the voltage-divider-based test equipment.
5.1.2. MHiL Impact on UVRT
This section discusses the role the MHiL plays for an UVRT event tested at a test bench. Since the DUT is a type 4 wind turbine and equipped with a full-scale power converter, grid events are very much decoupled from the mechanical system and, hence, from the MHiL system. For type 3 wind turbines without a full-scale power converter, a much more significant interaction is expected.
To illustrate, how the MHiL reacts to a grid fault event, a three-phase voltage dip to 0 pu in partial load is chosen. Many other symmetrical and asymmetrical voltage dips with different residual voltage levels and different X/R ratio were tested and, in principle, lead to identical results.
Figure 12 shows the mechanical quantities of the DUT during the UVRT event, such as wind speed, pitch angle, various torques, rotor speed, and active power, calculated from measured current and voltage.
Figure 13 shows the corresponding measurements for voltage, active current (
), and reactive current (
). As the top plot in
Figure 12 shows, the test was carried out at a constant wind speed. The bottom plot in the same figure features the active power, calculated at low- and medium-voltage, and indicates that the UVRT event begins at 60 s and lasts 400 ms, which can also be observed in the voltage plot in
Figure 13. For assessing the MHiL’s impact, the speed values in
Figure 12 are of interest. Before the fault begins, the MHiL internal speed reference and the measured generator speed at the test bench match. As soon as the fault occurs, the actual generator speed differs from the reference speed set by the MHiL. The observed acceleration is caused by a drop of the generator torque at the fault occurrence. It takes the MHiL system approximately 1 s to control the generator speed oscillation and up to 2 s to reduce the control error significantly. In the moment when the generator torque drops, the DUT and the reference model change their rotational speed according to their inertias with approximately 0.045 pu/s and 0.00054 pu/s, respectively. This acceleration of the DUT corresponds to a 0.01 pu (24 kNm) drop of mechanical generator torque, which matches the generator torque drop in
Figure 12 (22 kNm). This generator torque, calculated from electrical power, does not drop immediately, but after a delay time of 0.11 s, whereas the mechanical generator torque drops after 0.02 s, indicated by the DUT acceleration. This delay time of the calculated generator torque, which is an input to the MHiL system, contributes to the observed deviation. However, the maximum deviation of the reference and the measured speed is less than 0.5%. During all UVRT tests, no deviation of the MHiL reference and measured generator speed higher than that was found.
As the rotation speed and torque are linked to active power, an influence of the MHiL system on the certification results would be expected only in the active current or power measurement. As the bottom plot in
Figure 12 shows, no significant impact can be found.
In conclusion, there exists a minor interaction between the grid fault and the mechanical quantities. The analysis shows that the occurring deviation of the rotation speed, controlled by the MHiL system, does not influence the certification-relevant quantities negatively. However, for type 3 wind turbines, such a deviation may have relevant consequences, which is why the issues with the speed deviation should be resolved.
5.2. Frequency Control
With the Frequency Control test, the DUT’s ability to reduce active power output in the case of over-frequency is tested. An increased grid frequency indicates a power surplus in the grid and leads to a power drop of the DUT. A lookup-table, implemented in the DUT control, specifies the relationship between power and frequency. The limitations for power change-rates are set as high as possible when decreasing power. The power increase is subject to change-rate limitations. In the certification, the resulting static active power is considered by the IEC 61400-21-1 standard. The German FGW TR3 standard also analyses the dynamic transition, i.e., settling time, of the power. As the desired behavior requires rapid changes of the wind turbine’s active power, the rotor emulation of the MHiL systems plays a crucial role. An impact of the PHiL system was not observed, and is thus not discussed further.
A comparable test with the same DUT, but at a different test bench with another MHiL system, was presented in [
29]. The results presented therein are very promising in terms of the MHiL system’s performance. Different from this work, discussing the influence of the MHiL on the results was not the focus of that paper and of less relevance, due to high MHiL performance.
At the nacelle test bench, the test was executed with laminar wind. According to the German standard FGW TR3 Rev. 24, the frequency varies from 50 Hz to 50.25 Hz, 50.7 Hz, 51.15 Hz, and 50.07 Hz, before it is set to the nominal 50 Hz again. Each frequency level is held for 30 s.
Figure 14 shows wind speed, pitch angle, several torques, generator speed, and active power for the complete test procedure. On a macroscopic scale, the DUT power reacts as expected, with a power decrease and an increase when the frequency rises or falls, respectively. In more detail, the MHiL internal signals’ reference speed and actual speed show relevant deviations during dynamic transitions between 75 s and 110 s. The deviation is up to 4%.
At 75 s, the frequency increases from 50.25 Hz to 50.7 Hz and causes the DUT to rapidly reduce its active power output. This decrease leads to a lowered generator torque and an undesired generator speed increase at the test bench. In the field, the immense rotor inertia of a wind turbine prevents this. At the test bench, the MHiL system should reproduce this impact of the inertia. As described in the previous chapter, the delay of the generator torque signal and the limited control bandwidth prevent an adequate reaction of the MHiL system. This is due to a motor torque change-rate limitation, which is used up by approximately 80% during this test. Since the MHiL is designed such that the maximum is not fully exploited for robustness reasons, the observed shortcoming is linked to that test bench limitation.
Another effect which influences the MHiL performance is the interaction of the MHiL control with the DUT control. Due to the MHiL’s limited bandwidth, both controls are not fully decoupled in the frequency range and interactions may lead to oscillations. This theory is supported by the fact that the pitch angle, commanded by the DUT control, reacts in accordance with the observed speed oscillations. The interaction of both controls acts similar to a positive feedback. The increase of the generator speed increases the power for a given torque, instead of decreasing. Consequently, a further reduction of the torque is commanded by the DUT, which, in turn, lets the generator accelerate and so on. Both the MHiL and the DUT’s speed control try to level the input torque to the same level as the generator torque, but interfere with each other when trying so. After a settling time of 5 s, the tracking recovers.
The presented issues do not influence the steady state power, i.e., the settled power after 30 s. Since the IEC 61400-21-1 only considers this, the certification outcome would be unchanged. Only with respect to the settling time, which at least the German FGW TR3 considers as certification criteria, the result does affect the certification. This leads to a stated performance in a turbine certificate, which is worse than it is in reality. While this may be acceptable from a certification point of view, since the results would be more conservative, it is not desirable for the manufacturer to misrepresent the product as being worse than it actually is.
5.3. Synthetic Inertia
In contrast to the Frequency Control test, the Synthetic Inertia test evaluates the DUT’s ability to provide additional power to the grid in case of an under-frequency event [
2]. The test must be carried out at different power levels, namely between 0.25 and 0.5 pu, 0.8 pu, and 1 pu. For how long a wind turbine must be able to provide additional power depends on the DUT specification and is not defined by IEC 61400-211.
In this paper, we discuss a measurement at 1 pu power, with turbulent wind conditions carried out at the test bench.
Figure 15 shows wind speed, pitch angle, several torques, generator speed, grid frequency, and power. At 6 s, the grid frequency decreases stepwise to 48.5 Hz and recovers to 49.3 Hz at 15 s. After the grid frequency drops, the power boost phase, where the power is increased with maximum change-rate, starts. After the power reaches a steady level of 1.07 pu, it decreases with a pre-defined rate to nominal power again. As with frequency control, the DUT reacts as expected in terms of power. Considering the MHiL internal signals speed reference and actual speed, it is apparent that, again, deviations during the dynamic transitions occur. As with the frequency control before, this is caused by rapid power increase (6 s) and power decrease (15 s), respectively. The root causes are identical to those discussed for Frequency Control. In principle, the MHiL’s bandwidth is not high enough to couple the rotor inertia tight enough to the test bench. Compared to the results in the Frequency Control chapter, the observed deviation of the speed signals is much less. This is due to more moderate power change rates compared to the frequency control test.
The effect this difference has on the certification depends on the evaluated quantities. The stationary power boost level, which is evaluated, is not influenced, but reproduced perfectly. In contrast, the dynamic response, in terms of response and settling time, could be influenced. This heavily depends on the gradients set within the DUT software. For the considered DUT, no significant oscillations of the power can be observed so that the settling time is not influenced.
In conclusion, the MHiL system has no negative impact on the measurement results of the synthetic inertia test carried out with this specimen.
6. Discussion and Conclusions
The previous chapters introduced the different HiL systems, which allow operating a wind turbine at a test bench in a realistic manner and carrying out certification measurements. The experimental results showed that, with the given HiL systems, it is possible to conduct certification measurements at the test bench, which lead to results comparable to results from field measurements. As shown in [
28], this is not only true for the few tests discussed in this paper, but also for many others and at other nacelle test benches. However, the analysis in this paper also illustrated in detail that a limited accuracy or errors of the HiL systems, their simulation models, and controls could easily compromise the quality of the certification result. In the previous analysis in
Section 5, we showed, for instance, that the UVRT results are sensitive to the correct emulation of the impedance by the PHiL, which correlates with the findings in [
35], and that, for example, frequency control and synthetic inertia tests can easily be jeopardized by a MHiL system whose bandwidth is limited due to test bench restrictions and poor feedback signals.
Thus, test-bench-based certification is an opportunity for the industry to decrease the time to market, but also carries the risk of a faulty certification if it is not done correctly. Whether the latter happens unknowingly or intentionally, it poses a risk to the entire industry. Standardized procedures for the qualification of a test bench and its HiL systems could help to minimize this risk. Such standard procedures do not yet exist, but need to be developed. Based on the authors’ findings in this paper, and experiences from several test bench measurement campaigns, standard tests for several aspects can be proposed.
Models of the grid or parts of the wind turbine, which are implemented in the HiL system, usually contain simplifications and transfer to a different simulation platform. In order to check the validity of these models, the tests recommended to verify the impedance control, dynamic and static test-bench-performance should be used. Beyond that, for the validation of aerodynamic and mechanic models, the final steps of the procedure described in [
39] are recommended in order to ensure consistent models.
In order to verify the correctness of the PHiL system impedance control, one should operate the wind turbine at different grid impedances and let the wind turbine feed in a fixed reactive current. The voltage at the PCC should be measured and compared to the theoretical voltage rise/fall at the PCC. Alternatively, UVRT tests can be performed with different grid impedances to verify the correctness of the PHiL system impedance control. For this, one calculates the impedance from the two states before and during the fault, and compares it to the specified value. In both cases, it is important to also consider the impedances of the test setup, such as the DUT transformer. These must be added to the set impedance.
The dynamics of the grid emulator can be specified by performing no-load tests before each UVRT test series and calculating the positive sequence voltage at the PCC for each no-load test. The observed voltage transition should, according to IEC 61400-21-1, last maximum 7.5 ms for the dynamic transition during the dip and 10 ms for the transient voltage transition during recovery [
2].
For the verification of the MHiL system’s static properties, the DUT can be run at different wind speeds from cut-in to cut-out, with a maximum step size of 1 m/s, in laminar or turbulent conditions. Thereby, each wind speed is held for at least 100 s (laminar) or 600 s (turbulent), respectively. The resulting static values for generator power, rotation speed, and pitch angle should match results derived in offline simulation. With respect to field results, and also allowing some deviations of different numerical aeroelastic solvers, the tolerable power and speed deviations can be considered up to 0.1 pu. The pitch should be accurate by at least 2°.
For verifying the dynamic performance of the MHiL system and the test bench, different tests are required. In a first test, the DUT operates at different wind speeds from cut-in to cut-out in turbulent conditions. The wind speed steps can be up to 4 m/s. For determining the dynamics, the spectrum of the generator speed or power and is calculated and compared to simulation results of the same test run. For a type 3 wind turbine, a UVRT test can be carried out to determine the dynamics and calculate the spectrum, since the voltage drop leads to a step-like excitement of the mechanical drive train. The difference of the resulting eigenfrequencies should be below 5% of its nominal value, which correlates with tolerances of the mechanical models itself.
The second test, which is required to assess the MHiL system and test bench’s dynamic performance is a rapid change of the active power during operation. For this, the maximum change-rate of the active power of the DUT, occurring during operation, is identified. While the DUT is operated at nominal power, the power reference is changed to 0.4 pu, with the identified maximum change rate. The difference of the generator speed between the internal simulation and the experiment should be 2% or less, since higher deviations were found to have a relevant impact on the frequency control results.
The validation steps proposed in the above paragraphs can only be considered as a first draft. Academia and industry are invited to refine and extend this validation procedure with experiences and results derived during test bench measurement campaigns. Ultimately, the national and international standards must include such validation tests and procedures for test benches in order to prevent a lowering of the certification standards and, hence, risking the reputation of a whole industry. Since the goal is to reduce the required DUT hardware and virtualize testing, it is mandatory to define procedures which ensure the capability of a test bench and its HiL systems to serve as a platform for certification.
Important to note at this point is that all experimental results discussed in this paper are based on a type 4 wind turbine. There are no publicly available results for the same test campaign with a type 3 wind turbine. It is very likely that some of the effects regarding the accuracy of the PHiL and the MHiL system are even more important with such a type of wind turbine.