Assessing the influence of lightweight generators on substructure cost and levelized cost of energy: a techno-economic model for large offshore wind turbines

Trends in wind turbine design are continuously aiming at achieving higher rated power levels, beneficially leading to both decreasing levelized cost of energy (LCOE) and increasing annual energy yield (AEY). A problem arises for conventional direct-driven synchronous generators with such upscaling to 10 MW rated power and beyond: Related to the wind turbine’s rotor diameter, electrical power is increasing with power of two while mass is increasing with power of three. To tackle this disproportional increase in mass, an innovative wind turbine concept was developed in the research project “MagnetRing” [1]. The design is shown in Fig. 1. It utilizes the relatively light permanent magnet generator concept with innovative magnet circuit arrangement, high force densities and a low mass structure, specifically a large diameter ring design resembling a spoke wheel. Picking up from there, the project “MagnetRing II” delivered a full-scale 10 MW lightweight direct drive permanent magnet ring generator concept as well as a scaled demonstration generator of the same design [2]. “MagnetRing III” pushed forward the development, optimizing and integrating the generator system in order to reduce overall cost, improve grid integration, and evaluate the limits of the design regarding rated power [3]. As part of this project, the influence of direct cost, efficiency and mass of the generator on wind turbine LCOE and substructure cost in particular was assessed. The methodology, modelling and results are presented and discussed in this paper.

In order to evaluate the impact of the final MagnetRing generator design on LCOE, a case study was carried out, comparing a wind turbine comprising the MagnetRing generator (MR-WT) to a principally similar reference wind turbine including a conventional direct drive permanent magnet generator (Ref-WT). In detail, the case study was performed for two complete 400 MW offshore wind farms in the German North Sea Exclusive Economic Zone. The wind farms consist of 40 wind turbines of each particular type, having 10 MW of rated power and being installed on tubular steel towers and monopile foundations. For the calculation of the resulting LCOE according to Eq. 1, the case study considers initial capital expenditure (CAPEX) for turbines and farm balance including installation vessel cost, operational expenditure (OPEX) for turbines and farm balance, annual energy yield (AEY), and decommissioning cost for the turbines, for a life span n of 20 years and a constant moderate interest rate i of 3%/a.

It is presumed that the MagnetRing generator design influences LCOE mainly through its purchase cost, its efficiency characteristics, and its mass; OPEX is presumed to be not particularly different compared to conventional designs, decommissioning cost are not considered. Figure 2 shows how LCOE are composed. While purchase cost and efficiency feed into LCOE very directly through CAPEX and AEY, generator mass affects LCOE only indirectly through the turbine’s rotor-nacelle-assembly (RNA) mass, a matching mechanical substructure design to achieve structural integrity for an according RNA mass, and consequently the CAPEX of that appropriate substructure. Concluding from this and looking for applicability beyond a singular case study, a detailed technical design and economics model was created for a wind turbine mechanical substructure, practically converting generator mass into substructure CAPEX in order to feed it into LCOE calculation. To reflect a designer’s typical aim to avoid unnecessary oversizing, economic optimization of a limited number of substructure geometry features is implemented utilizing the model, allowing identifying the most cost-efficient substructure fulfilling the given design constraints.

In order to keep the required substructure design and economic modelling manageable, of all mechanical substructure components only the tubular steel tower, transition piece and monopile foundation were considered, since they represent the most significant share of wind turbine substructure CAPEX [4] and consequently offer the most relevant potential for decreasing LCOE. Additionally, in order to facilitate automatic geometry optimization and limit optimization run time, the design modelling of these components was restricted to a limited number of basic design steps according to Fig. 3, reaching from the collection of “turbine and substructure data”, continuing with the definition of a “draft tower geometry”, and finally performing a “modal analysis”, “buckling/breaking strength checks” and “deflection checks”.

Deriving a solution from the complex task of calculating LCOE for a complete wind farm dependent on wind turbine RNA mass, an integrated software tool was created which interconnects known existent as well as new purpose-created models, reducing the set of required inputs to a minimum and adding automatic substructure design optimization, thus relieving the end user and allowing fast conclusions towards the economic benefit of a specified custom drive train design of 10 MW rated power. For the purpose of performing a particular case study, the tool would practically be run twice: once for the Ref-WT, and once for a wind turbine featuring a custom drive train featuring a potentially lightweight generator and uncommon auxiliary systems, in this case the MR-WT. For any custom drive train, a set of typical input parameters would have to be known, distinguishing that drive train from the reference, primarily including CAPEX, mass, efficiency characteristics and potentially OPEX.

The utilized models in the aforementioned software tool can in principle be separated into four groups: mass models related to wind turbine components, design models related to the wind turbine substructure, energy models related to electricity generation of wind turbines, and cost models related to wind turbines, substructures and wind farms.

A large number of mass models produced for a 10 MW offshore wind turbine design were incorporated into the tool for modelling masses of generic wind turbine components excluding the substructure [4]. Since the modelling of a conventional direct drive generator’s mass appeared questionable (low mass compared to other available sources), an alternative mass value was taken from a dedicated 10 MW permanent magnet direct drive generator design study ([6]; see Table 1). The mass of the wind turbine substructure components are a by-product and integrated part of substructure design models.

To accomplish a substructure design modelling according to the described methodology, the draft “tower” (in this case the whole substructure comprising monopile, transition piece and the actual tower) geometry modelling was taken from a publicly available 10 MW reference onshore wind turbine design [7] and a manual for designing monopiles for offshore wind turbines [8]. Main fixed parameters are 40 m water depth, 119 m hub height, 178.3 m rotor diameter, maximum rotational speed 9.6 min⁻¹, minimum rotational speed 6 min⁻¹.

The modelling to perform the structural integrity checks of the overall wind turbine structure was taken from various scientific resources: Modal analysis modelling using Rayleigh’s method is based on [8] as well as [9‐11], targeting a “soft-stiff” design between 1P and 3P excitation frequencies, and considering soil interaction in medium dense sand as well as damping. Interaction with water is not considered. After cross-checking the modal analysis results with a small number of publicly available wind turbine tower designs, the resulting 1st bending eigenfrequency is assumed to be overestimated by 10% and is reduced accordingly in the utilized model.

Global and local buckling modelling was taken from [12] and [13]. It appeared that for non-cylindrical structures as assumed for the wind turbine tower in this case study, global buckling modelling could not be adopted straight away. For the time being, the results of the global buckling check for the actual tower and the complete wind turbine substructure are therefore disregarded in this case study. Compensating for this momentary gap, a less conservative custom-made breaking strength model is utilized, using tower top load assumptions and safety factors from a 10 MW offshore wind turbine study [14].

Deflection checks for the monopile were modelled according to [8], aiming at a slender/elastic monopile design. The design is to maintain acceptable deflections of 0.12 m at sea floor level, 0.02 m at the bottom of the monopile [15], and an acceptable tilt angle at the top pf the tower of 0.5°.

Modelling of the AEY was taken from cost and energy models referring to a 10 MW offshore wind turbine [4], including parametrization with a Weibull shape factor k of 2.0 and a Weibull form factor C of 10.38 m/s. The drive train efficiency specifications of the Ref-WT as shown in Fig. 4 were custom-modelled.

Analogous to the mass models, CAPEX models for generic wind turbine components were also taken from [4]. Analogous to the questionable modelling of the conventional direct drive generator’s mass, its CAPEX (high compared to other available sources) was custom-modelled (see Table 1). CAPEX modelling and parametrization of the wind turbine substructure components was taken from [16], utilizing the mass results of the dedicated substructure design models. With steel price as the main parameter of the substructure CAPEX model being a volatile product of global market developments, the parametrization taken from the original source of the model has to be revised in the future, and the substructure CAPEX results have to be taken with a pinch of salt when observed singularly. However, for the purpose of performing this case study, questionable parametrization is seen less critical as long as potentially deviant sub-results affect both Ref-WT and MR-WT in the same way.

CAPEX modelling and parametrization related to wind farm balance and OPEX modelling related to both wind farms and single wind turbines were taken from [17]. Originally resembling a 400 MW offshore windfarm comprising 100 turbines of four Megawatt rated power, slight linear adaptations were made to parts of the model to resemble a 400 MW offshore windfarm comprising only 40 turbines of 10 MW rated power instead.

CAPEX modelling and parametrization related to the use of wind farm installation vessels was taken from [18] and [19]. Monopile mass was identified as a critical parameter in this model, creating a significant jump in CAPEX when exceeding the hoisting limit of typical installation vessels of commonly 1500 t; day rates of suitable vessels would more than double in such a case.

Miscellaneous CAPEX modelling related to foundation installation such as installation of scour protection, noise protection and grouting installation were taken from [18].

All models are implemented in Python and are formed into an interlinked unified tool of various sub-modules. For each drive train design to be assessed (including the reference case), the tool consecutively executes the sub-modules, most of them as part of an optimization loop. During optimization only three independent geometrical features of the wind turbine substructure are varied, specifically the outer diameter of the monopile and the diameter-to-thickness ratios of the tower and the transition piece, in order to find the appropriate substructure for each generator design causing minimum LCOE. Figure 5 shows the order of execution of the sub-modules, starting with the calculation of masses of all RNA components where not given as initial inputs, and then running the optimization. The optimization is realized through a minimize function using the COBYLA method, allowing the use of constraints. The loop includes the setting of the aforementioned geometry parameters, the calculation of CAPEX for the RNA, the current substructure design, offshore installation of the complete wind turbine, and proportionate CAPEX for an offshore wind farm in which the turbine is embedded. Furthermore, the calculation of OPEX, downtime, AEY and finally the resulting LCOE are included in the loop. The tool could take into account special OPEX and downtime models for unconventional wind turbine designs if required.

Accompanying the optimization are value boundaries regarding substructure geometry on the one hand, and crucial mechanical engineering constraints related to valid substructure design on the other. In particular, the utilized constraints are a valid 1st bending eigenfrequency of the overall substructure, subcritical local and global buckling strength of the steel shells (with mentioned limitations), subcritical breaking strength of the steel shells, and an acceptable deflection of the substructure.

Table 1 shows the main results for this case study, comparing the LCOE performance of Ref-WT and MR-WT drive train designs. LCOE, CAPEX and OPEX numbers at the top of the table represent the results for the given complete wind farms, broken down to a single wind turbine.

The specified MagnetRing generator design can indeed create a drop in overall LCOE, primarily through a significant direct CAPEX reduction of the generator, and secondly through indirect CAPEX reduction of the substructure; the economic impact of the MR-WT drive train’s slightly higher AEY performance is low. The massively reduced generator and consequently overall RNA mass apparently leads to a significant, yet disproportionate reduction of turbine substructure mass and consequently only to moderate indirect CAPEX savings there.

The limitation in disproportionate CAPEX reductions related to the substructure, despite achieving massive RNA mass reductions, is connected to the necessary substructure design steps and applied constraints which ensure a valid substructure design (see Figs. 3 and 5). Figure 6 shows the resulting first bending eigenfrequency of the MR-WT after soft-stiff substructure geometry optimization: The bending eigenfrequency—a constraint being highly sensitive to RNA mass—lies well above the allowed minimum close to the 1P excitation frequency at rated rotor speed. In theory, this remaining frequency margin could be used up by an even softer—and consequently lighter and lower-cost—substructure, shifting the eigenfrequency even closer to the allowed lower boundary. However, Fig. 7 shows that the interdependent designs of the MR-WT’s substructure elements have already reached the minimum allowed safety factor of 1.0 regarding breaking strength—a constraint being practically insensitive to RNA mass; a softer, lighter and lower-cost substructure would in consequence be invalid.

A limited number of sensitivity analyses were conducted with the techno-economic tool and the Ref-WT parameter set in order to assess the LCOE leverage effect of relevant generator and drive train properties. When properties need to be balanced during a design process, those property changes having a larger effect on LCOE could be preferred over less effective ones, and counter-acting property changes could be traded off against each other.

Figure 8 shows the result for RNA mass reduction. The trend indicates that an LCOE decrease of approximately 1% can be achieved through a 15% lighter RNA. However, mass reduction beyond that value is without effect. The reason is that the design of the mechanical wind turbine substructure could not become any lighter and therefore cheaper due to surpassing its critical breaking strength. The result shows both the value and the limitation of this particular design path: From the actually achieved RNA mass reduction for the MR-WT as shown in Table 1, it can be concluded that mass reductions in the area of 15% could possibly be achieved with relative ease. In connection with the relatively high LCOE reduction lever of 1%, such a reduction appears to be economically very attractive. Extraordinary high RNA mass reduction targets should to be questioned, though, at least if they are not coupled to significant direct cost reductions.

The leverage effect of such cost reductions is shown in Fig. 9. The trend shows an LCOE reduction of 1% at approximately 25% lower generator CAPEX.

From the actually achieved generator CAPEX reduction for the MR-WT as shown in Table 1, it can be concluded that this design path is economically highly attractive and LCOE reductions in the area of 2% could be achieved with dedicated low-cost generator designs. Generator CAPEX reduction is likely to be linked to low degree generator mass reduction measures, while high degree generator mass reduction measures could lead to CAPEX increase instead.

The efficiency related LCOE trend as depicted in Fig. 10 shows an LCOE decrease of approximately 1% at an efficiency offset for the overall PTO of approximately + 1%. This lever is economically highly attractive, and its increase has been the top priority of historical generator development measures. Especially when attempting to reduce generator mass and CAPEX, negative trade-offs in efficiency should be reviewed carefully.

Figure 11 shows that an LCOE increase of 1% is suffered with 10% higher wind turbine OPEX. This negative leverage effect appears to be extremely high, is difficult to foresee in a design process, and could therefore have a decisive impact on the overall success of a drive train design. It can be assumed that an OPEX increase of that magnitude could very easily suffered, especially with novel and complex generator designs. This leverage effect must not be underestimated.

In this paper an integrated techno-economic software tool is described which is capable of feeding offshore wind farm cost and energy yield into an LCOE calculation. The tool is composed of a number of mass and cost related sub-models, i.a. modelling relevant drive train properties of 10 MW wind turbines such as generator CAPEX, drive train efficiency and RNA mass. In order to feed the RNA mass into the LCOE calculation, a wind turbine substructure design model is utilized, modelling a tubular steel tower, transition piece and monopile foundation while considering basic technical design constraints. The models of the tool are integrated into a substructure geometry optimization loop, generating the most cost-efficient substructure for a given RNA mass and therefore finding the optimum CAPEX saving potential induced by the mass of a lightweight generator. The integrated character of the tool reduces the set of required inputs to a minimum, relieves the end user and allows fast conclusions towards the economic benefit of a specified custom wind turbine drive train design of 10 MW rated power.

With the help of the tool, a case study is performed, comparing the LCOE performance of the lightweight MagnetRing drive train design to that of a conventional reference design in the context of being applied in the turbines of an offshore wind farm. The main results as presented in Table 1 show that the MagnetRing design has massive advantages compared to the reference in terms of generator mass (74.4% reduction) and generator CAPEX (44.9% reduction). A huge potential for secondary CAPEX savings through a softer, lighter and lower-cost wind turbine substructure is created by the according RNA mass reduction through an effective increase of the overall wind turbine’s 1st bending eigenfrequency. With the limitation of breaking strength being very insensitive to RNA mass, the potential cannot be fully exploited though, resulting in moderate substructure CAPEX savings (7.0 to 10.7% decrease). Still, the achieved LCOE reduction is significant (3.7% reduction down to 5.73 €-cent/kWh).

Besides the conclusion that RNA mass induced substructure CAPEX reductions can be quite limited due to insensitive substructure design constraints, the indicated LCOE sensitivities with respect to RNA mass and generator CAPEX show that drive train design decisions which reduce direct CAPEX are more effective in decreasing LCOE than decisions reducing mass. For the current MagnetRing generator design, these circumstances in connection with the untapped bending eigenfrequency margin suggest that future designs could effectively economize on lightweight construction measures, eventually reducing direct CAPEX and therefore the overall LCOE even further while still leading to viable mechanical wind turbine substructures.

The techno-economic models of the described tool are currently being amended in order to assess the influence on LCOE of a fully super-conducting generator system design for large offshore wind turbines [20], featuring low mass at the cost of high technological complexity.