Greater blade lengths and higher tip speeds, coupled with a harsh environment, have caused blade leading edge erosion to develop into a significant problem for the offshore wind industry. Current protection systems do not last the lifetime of the turbine and require regular replacement. It is important to understand the characteristics of the offshore environment to model and predict leading edge erosion. The offshore precipitation environment has been characterised using up-to-date measuring techniques. Heavy and violent rain was rare and is unlikely to be the sole driver of leading edge erosion. The dataset was compared to the most widely used droplet size distribution. It was found that this distribution did not fit the offshore data and that any lifetime predictions made using it are likely to be inaccurate. A general offshore droplet size distribution has been presented that can be used to improve lifetime predictions and reduce lost power production and unexpected turbine downtime.
The offshore wind industry's need for larger rotors and higher tip speeds has caused blade leading edge erosion to develop into a major problem for the industry. Leading edge erosion is caused by raindrops, hailstones and other particles impacting the leading edge of the blade and removing material. This degrades the aerodynamic performance of the blade and requires operators to perform expensive repairs. The issue has grown in prominence recently with reports that Ørsted had to make repairs to up to 2000 offshore wind turbines after just a few years of operation (Finans, 2018).
The industry attempts to prevent the onset of leading edge erosion by applying protection systems, such as coating and tapes, to the blade leading edge. However, currently these do not last the lifetime of the turbine and require regular replacement.
Several analytical models that aim to estimate the expected lifetime of a protection system have been developed (Eisenberg et al., 2018; Slot et al., 2015; Springer et al., 1974). Finite-element models that can predict the stresses and strains in a protection system from an impinging water droplet have also been produced (Keegan et al., 2012; Doagou-Rad and Mishnaevsky, 2020). To model leading edge erosion, it is important to understand the characteristics of the impinging hydrometeors and, as rain is the most frequent hydrometeor, the droplet size distribution (DSD) of the impinging rain.
The aim of the industry is to develop a methodology that can predict the lifetime of a protection system on a wind turbine from rain erosion tests. The DNV GL project COBRA aims to address this, and Eisenberg et al. (2018) propose using the Springer model. Due to the lack of an offshore dataset, the project uses the onshore Best distribution published in 1950 (Best, 1950). However, the manual measurement techniques used by Best are outdated and have been found to provide inaccurate results (Kathiravelu et al., 2016). The lack of an offshore dataset introduces uncertainty into lifetime predictions, and, as a result, inaccuracies may exist. In this work, state-of-the-art measurement techniques have been used to characterise the offshore precipitation environment and provide the required offshore dataset. A general offshore DSD is presented.
The most widely used DSD is the Best distribution. Best takes the work of
several authors and converts them into a common DSD defined as
This is commonly presented in the literature as
In both measurement techniques, sampling can only occur in short intervals.
Best performs measurements using the stain method for a maximum of 2 min. During prolonged periods of sampling, the droplet stains and
pellets can overlap, making it difficult to accurately measure and count
individual drops. Furthermore, the techniques have a low resolution.
Best registers droplet sizes in 0.5 mm intervals. Given that the
distribution predicts that for a rain rate of 1 mm h
Two Campbell Scientific PWS100 disdrometers have been installed onto
Offshore Renewable Energy Catapult's offshore anemometry hub, which is
located 3 nautical miles (5.56 km) from the coast of Blyth, Northumberland. Figure 1 shows the position of the two disdrometers, with the first mounted on the existing platform 25 m a.s.l. (above sea level; disdrometer A) and the second mounted 55 m a.s.l. (disdrometer B). Each disdrometer consists of two photodiode-sensing heads, one near-infrared diode laser head, and one CS215
temperature and humidity sensor. The sensor heads are positioned
20
The optical disdrometers mounted to the platform
Optical disdrometers are non-intrusive and do not influence drop behaviour during measurement. They have also been shown to successfully resolve droplet break-up and splatter problems experienced by other measurement techniques (Kathiravelu et al., 2016). Agnew (2013) explored the performance of the PWS100 at a site in southern England, finding that the device slightly underestimates the number of droplets with a diameter below 0.8 mm. However, the measurement of larger, more damaging droplets was found to be accurate. Montero-Martínez et al. (2016) compared the performance of the disdrometer during natural rain events in Mexico City to results from a beam occlusion disdrometer and a reference tipping bucket. The PWS100 recorded greater amounts of precipitation than the reference, but the study was unable to back this up statistically and no significant differences in precipitation estimation was found between the disdrometers. Montero-Martínez et al. (2016) concluded that the two devices performed similarly and that the PWS100 provides reliable precipitation measurements. Johannsen et al. (2020) studied the PWS100 against a Thies CLIMA laser precipitation monitor and an OTT Parsivel at a site in Austria. In contrast to Montero-Martínez et al. (2016), the PWS100 recorded less than the reference rain gauge in all but two events. The PWS100 recorded 3 % less total precipitation than the rain gauge across the measurement period, outperforming the Thies and the Parsivel instruments which recorded 20 % and 30 % less, respectively, and the PWS100 was consistently closest to the rain gauge reading throughout the period. Similar drop sizes were recorded between the PWS100 and the Parsivel, with Johannsen et al. (2020) noting that the PWS100 tended to record slightly faster and larger drops. The studies show that there are uncertainties in the accuracy of all disdrometers, with the PWS100 used in this study performing comparatively with or better than the other examined disdrometers.
DSD data from 1 September 2018 up to and including the 31 August 2019 are presented to provide a 12-month period for analysis. This allows analysis to also be completed seasonally. Hydrometeor diameters have been recorded with a resolution of 0.1 mm. Data are available with a time interval of 1 min.
Raw data were received from the disdrometers, and, therefore, detailed quality
control was completed before subsequent analysis in line with recommendations from Hasager et al. (2020), Chen et al. (2016) and Vejen et al. (2018). Duplicate records were assessed by comparing timestamps, with any identical timestamps eliminated from the dataset. The meteorological parameters were also evaluated to remove entire duplicate records. It may be possible for a few parameters to be the same; however an entire row of identical parameters is extremely unlikely, and consequently duplication has almost certainly occurred. A gross value check was completed to remove unrealistic and impossible values. Certain parameters are constrained within limits, such as relative humidity, which is given as a percentage, whereas other parameters, such as droplet size, can be evaluated against sensible threshold values. Furthermore, precipitation events where the disdrometer recorded a rain rate of 0 mm h
The consistency between disdrometers was also explored. No sensible results
were recorded by disdrometer A from 23 November 2018 until its repair at the start of May 2019, whilst disdrometer B remained in operation throughout the year with short, infrequent gaps in data gathering. Of the available
recordings, the two disdrometers agreed on the occurrence of precipitation
97.40 % of the time, with this increasing to 99.74 % when evaluating
precipitation intensities above 0.5 mm h
The comparable data gathered by disdrometer A and B enabled some gaps in
disdrometer B's dataset to be filled with the respective data from
disdrometer A, where available. In total, 34.25 h were gap filled, of which 229 min experienced precipitation and 111 min experienced a precipitation greater than 0.5 mm h
Precipitation intensity during the measurement period.
Cumulative distribution of precipitation for the respective seasons.
Table 1 presents the percentage of available quality-controlled data for each month and the percentage of the available data in which precipitation was recorded. An estimation of the actual percentage of precipitation can be obtained by assuming that the same proportion of precipitation occurred across the unavailable data. A total of 82.89 % of the data were available during the entire measurement period. Precipitation was recorded in 8.71 % of the available data, giving a yearly precipitation estimate of 10.50 %. Winter had the highest estimation of total time with precipitation with 12.07 %, whilst spring saw the lowest with an estimation of 8.65 %. Including the missing data provides an annual accumulation of 500 mm, which is lower than the 650 mm average annual precipitation reported in Northumberland (WeatherSpark, 2020), indicating that the measurement year was a relatively dry year for the area.
Percentage of available data for each month.
The average precipitation intensity was recorded every minute. Figure 2
presents its variation across the measurement period, and Fig. 3 presents
the cumulative frequency of the recorded intensities. The median
precipitation intensity for the measurement period was 0.311 mm h
Precipitation intensity distribution for seasons and intensity categories.
Precipitation is classified according to its intensity with the following
categories defined by the Met Office (2007):
light – precipitation intensity less than 2.5 mm h moderate – precipitation intensity between 2.5 and 10 mm h heavy – precipitation intensity between 10 and 50 mm h violent – precipitation intensity greater than 50 mm h
The seasonal breakdown of precipitation categories is shown in Table 2. Summer had the highest median precipitation intensity with the highest amount of recorded heavy and violent precipitation. In contrast, winter and spring saw minimal heavy precipitation and no violent precipitation. Light precipitation dominated across the entire measurement period, accounting for
92.58 % of all precipitation. Furthermore, 78.31 % of the recorded
minutes had an intensity lower than 1 mm h
Therefore, a wind turbine in this location would experience less than 3.5 h yr
Figure 4 presents the number of recorded hydrometeors by type during the data collection period. The hydrometeor type is clearly dominated by rain droplets. “Errors” and “unknown” particles accounted for 17.93 % of the hydrometeors recorded. These may be caused by insects, particles between states or equipment failures and have been ignored in the subsequent analysis, with any records where they were the modal hydrometeor removed. Drizzle and rain droplets make up a combined 98.45 % of all hydrometeors recorded. The number of ice pellets, hail and graupel particles recorded was low, accounting for only 0.49 % of hydrometeors recorded.
Number and type of hydrometeors recorded during the total measurement period.
As expected, ice- and snow-based hydrometeors occurred most frequently in winter. Ice pellets, hail and graupel accounted for 0.94 % of the hydrometeors recorded in the season with snow grains and snowflakes accounting for 3.56 %. In contrast, only 0.16 % of hydrometeors recorded in summer were ice pellets or hail, with no graupel, snow grains or snowflakes. Spring and autumn recorded 0.31 % and 0.57 %, respectively, of ice pellets, hail and graupel.
The severity of a hydrometeor impact is governed by its kinetic energy. Whilst the blade speed provides most of the impact velocity, the hydrometeor fall velocity and mass are important. For each minute, the average diameter and velocity was plotted for the modal hydrometeor type. This is presented in Fig. 5.
Relationship between size and velocity for the modal hydrometeor at each minute.
There is a clear distinction between water particles and snow particles, with snow particles occurring across a wider range of diameters and lower velocities than rain particles. For the few cases where ice pellets were the model hydrometeors, they all occurred to the right of the rain droplet scatter, indicating that they have a lower fall velocity than rain droplets. There were no cases where hail or graupel were the modal hydrometeor, and they were found to be mixed in with rain particles. The presented velocities for water particles are in line with those predicted in models by Gossard et al. (1992) and Brandes et al. (2002). The data presented in the above figure are used in the subsequent analysis to estimate the number of droplets that impact the blade per second and inform lifetime prediction models.
To inform lifetime prediction models, a general equation for an offshore DSD is required. The Best DSD has been reproduced, both seasonally and non-seasonally, with updated constants for the offshore rain data presented. Only data where rain particles were the modal hydrometeor were examined.
For each recorded minute, the cumulative function,
Rearranging Eq. (1) gives
Evaluation of Eq. (4) for precipitation intensities 0.1058, 1.2708, 4.9865 and 10.2624 mm h
Rearranging Eq. (2) gives
Evaluation of Eq. (5) to derive the constants
The constants
Best concluded that the constant
Evaluation of Eq. (7) to derive the constants
The constants
Determined constants for the non-seasonal and seasonal offshore DSDs.
Reproducing Eq. (1) with the derived non-seasonal constants gives a general non-seasonal offshore DSD:
The non-seasonal offshore DSD at different precipitation intensities.
The sensitivity of the constants to the data selected has been evaluated.
The following cases have been examined:
Low and high precipitation intensity have been individually and collectedly neglected. Precipitation intensities below 0.1 mm h Precipitation intensities that account for a small number of the recorded intensities have been individually and collectively neglected. These are the bottom 1 % and the top 1 %.
Minutes where the measured precipitation intensity is low generally record
fewer droplets than those with higher precipitation intensities. Conversely, a
significant number of droplets are generally seen in heavy precipitation.
Low- and heavy-intensity rain may, therefore, have a high scatter that could
influence the determined constants. Figure 3 presents the cumulative
distribution of the recorded precipitation intensities. The bottom and top
1 % of precipitation intensities may also skew the data by providing a
point significantly different to the trend. The impact of these conditions
on the constants is shown in Table 4.
Sensitivity of constants to the selected cases.
In general, the constants are consistent across all the examined cases. The
constant
The general offshore DSD has been compared to the Best DSD at various
precipitation intensities in Fig. 10. The precipitation intensities 0.1, 1,
2.5, 5, 10 and 20 mm h
Comparison between the offshore DSD and the Best DSD at
precipitation intensities
Figure 10 reveals that the Best DSD significantly overestimates the diameter
of droplets. This is particularly true at the higher precipitation
intensities. The goodness of fit of the offshore and Best DSDs has been
evaluated across the range of precipitation intensities in Fig. 11. The
offshore DSD aligns well with the raw data and possesses a high coefficient
of determination (
Coefficient of determination of the offshore DSD and the Best DSD across a range of precipitation intensities.
The offshore DSD presented has two main limitations. Firstly, the presented measurement period may be a limiting factor. As the disdrometer continues to collect data, the DSD can be further refined. Secondly, data have only been collected at one point. Offshore DSDs may vary from location to location. To address this, a disdrometer has been positioned at ORE Catapult's Levenmouth offshore demonstration turbine for future comparison and validation.
The implications of the offshore DSD have been assessed using the Springer
model, which is used by Eisenberg et al. (2018) to predict a protection solution's in situ lifetime from leading edge erosion. The model uses the median droplet diameter for a given rain rate to determine the number of impacts to failure,
The rate of damage,
The exact number of impacts to failure is dependent on the protection system and substrate used. For a commercial erosion-resistant polyurethane coating system, the offshore DSD has been applied to the above equations, and the relative effect on leading edge erosion prediction of the DSD in relation to the Best DSD is presented in Fig. 12.
Percentage change in leading edge erosion damage values from implementing the offshore DSD relative to implementing the Best DSD.
The smaller median droplet diameter for precipitation intensities above 0.15 mm h
This dataset can be used to help to inform the lifetime of leading edge erosion protection systems installed offshore, helping to ensure maintenance is conducted early and further leading edge erosion can be combatted. The dataset can also be used to inform droplet impact models and rain erosion testing with the greater understanding of the environment facilitating the development of improved protection systems.
DSDs are important in predicting and modelling leading edge erosion. Currently, there is a lack of an offshore dataset and the industry uses onshore distributions in lifetime predictions. In this work, a disdrometer has been positioned 3 nautical miles (5.56 km) offshore to collect and characterise the offshore precipitation environment and to provide an offshore DSD for lifetime prediction models.
Heavy and violent precipitation was rare in the measurement period, accounting for less than 3.5 h of precipitation across the year. Therefore, erosion damage is not likely to be driven exclusively by heavy and violent precipitation. Rain was the most frequently occurring hydrometeor, whereas snow, ice and hail particles were scarce. A clear distinction was visible in the diameter–velocity plots for each hydrometeor, with snow particles occurring across a wider range of diameters and lower average velocities. The majority of raindrops observed had a diameter below 2 mm.
A general offshore DSD has been presented. The raw data were compared to the
presented DSD and the most widely used DSD proposed by Best. A statistical
The results presented address the lack of an offshore dataset and provide a general offshore DSD that can be used to inform lifetime prediction models for the offshore environment. A disdrometer has been placed at ORE Catapult's Levenmouth offshore wind turbine to provide further information about the precipitation environment and validate the presented DSD. The offshore dataset can be used to improve prediction and modelling techniques, helping to inform the design of new protection solutions and to combat leading edge erosion whilst reducing lost energy production and unexpected turbine downtime.
Please contact the corresponding author.
RH had the lead on paper writing and data analysis and derived conclusions. KD and PH were responsible for installing and setting up the disdrometers. KD and CW supervised the research.
The authors declare that they have no conflict of interest.
This article is part of the special issue “WindEurope Offshore 2019”. It is a result of the WindEurope Offshore 2019, Copenhagen, Denmark, 26–28 November 2019.
This work was supported by the Engineering and Physical Sciences Research Council through the EPSRC Centre for Doctoral Training in Composites Manufacture, project partner the Offshore Renewable Energy Catapult,
This research has been supported by the EPSRC Centre for Doctoral Training in Composites Manufacture (grant no. EP/K50323X/1) and the EPSRC Future Composites Manufacturing Hub (grant no. EP/P006701/1).
This paper was edited by Ignacio Marti Perez and reviewed by two anonymous referees.