**4. Conclusions**

The first part of the present study investigated the relationship among the number of yaw gear and motor failures and TI at all the wind turbines under investigation with the use of in situ data. The investigation revealed that wind turbine #7 (T7), which experienced a large number of failures, was affected by terrain-induced turbulence with TI that exceeded those presumed for the wind turbine design class for which T7 was designed. The frequency of occurrence of such terrain-induced turbulence at this wind turbine was also significantly higher than that at the other wind turbines. When the TI values evaluated for a wind turbine are high only in some of the examined periods, it is likely that upstream terrain mainly accounts for the high values in those periods and that the TI values become high in specific wind directions. Accordingly, the TI values at T7 were examined with respect various wind directions. The examination revealed that the TI values were high in westerly to northwesterly winds. Because complex terrain with ridges and surface undulations of various scales

exists in the vicinity of T7, it was speculated that wind flows separated due to these ridges, resulting in terrain-induced turbulence.

Subsequently, a CFD simulation was performed to examine if the abovementioned observed wind flow characteristics could be successfully simulated. The CFD software package that was used in the present study was RIAM-COMPACT, which was developed by the first author of the present paper. RIAM-COMPACT is a nonlinear, unsteady wind prediction model that uses LES for the turbulence model. RIAM-COMPACT is capable of simulating flow collision, separation, and reattachment and also various unsteady turbulence–eddy phenomena that are caused by flow collision, separation, and reattachment. A close examination of computer animations of the streamwise (x) wind velocity revealed the following findings: As we predicted, wind flow that was separated from a micro-topographical feature (micro-scale terrain undulations) upstream of T7 generated large vortices. These vortices were shed downstream in a nearly periodic manner, which in turn generated terrain-induced turbulence, affecting T7 directly. In addition, vertical wind profiles of the (instantaneous) streamwise wind velocity at the wind turbine site were studied from the time at which the abovementioned unsteady turbulence–eddy phenomena occurred. As a result, it was found that a complex wind flow field with unusual vertical profiles of streamwise wind velocity formed at the wind turbine. More specifically, these vertical profiles of streamwise wind velocity deviated significantly from the power law profile and also had negative wind shear. On the other hand, the time-averaged wind flow data showed that the average streamwise wind velocity increased locally at the wind turbine site due to terrain effects and that no significant wind velocity deficit occurred at the site. Thus, even when no significant wind velocity deficit existed in the vertical profile of the time-averaged streamwise wind velocity, significant velocity deficits were present in vertical profiles of the instantaneous wind velocity, and such instantaneous wind velocity deficits should not be overlooked. An examination of the values of the standard deviations of the three wind velocity components at the wind turbine hub center (60 m above the ground surface) revealed that the value for the vertical (z) direction was large. The ratio of the standard deviations of the three wind velocity components at the wind turbine hub height was σ1:σ2:σ3 = 1.0:0.7:0.65, which clearly showed the effect of terrain-induced turbulence. When wind flow with such properties passes through a wind turbine, it causes a large wind load on the wind turbine. Furthermore, because the structure of this complex wind flow is three-dimensional, wind loads imposed on the left and right side of the swept area are expected to differ. It was speculated that such wind loads would exert force on the wind turbine in such a way that they would forcibly rotate the nacelle of the wind turbine, which in turn would cause impact loads on both the yaw gears and motors and ultimately result in the failures of the yaw gears and motors.

Finally, the temporal change of the streamwise (x) wind velocity (a non-dimensional quantity) at the hub height of T7 in the period from 600 to 800 in non-dimensional time was re-scaled in such a way that the average value of the streamwise (x) wind velocity for this period was 8.0 m/s, and the results of the analysis of the re-scaled data were discussed. With the re-scaled full-scale streamwise wind velocity (m/s) data (total number of data points: approximately 50,000; time interval: 0.3 s), the time-averaged streamwise (x) wind velocity and turbulence intensity (TI) were evaluated using a common statistical processing procedure adopted for in situ data. Specifically, 10-min moving averaging (number of sample data points: 1932) was performed on the re-scaled data. Comparisons of the evaluated TI values to the TI values from the Normal Turbulence Model in IEC61400-1 Ed.3 (2005) revealed the following: Although the evaluated TI values were not as large as those observed in situ, some of the evaluated TI values exceeded the values for turbulence class A, suggesting that the influence of terrain-induced turbulence on the wind turbine was well simulated.

**Author Contributions:** The study idea, plan and design were conceived by T.U.; And, all authors prepared the manuscript.

**Funding:** This work was supported by JSPS KAKENHI Grant Number 17H02053. **Acknowledgments:** The present study was supported by collaborative research with Eurus Energy Holdings Corporation: (1) Collaborative research and development on a method for assessing the wind speed ratio for wind turbine sites on complex terrain, July 2013-March 2014, December 2015-March 2018; (2) Collaborative research on land use modeling and its effect on simulated local-scale wind fields, December 2015-June 2016, principle investigator: Takanori Uchida. The authors wish to express their gratitude to all the individuals involved.

**Conflicts of Interest:** All authors declare no conflict of interest.
