In practice, rotor blade design starts with material selection, and to this end, candidate materials may be inventorised based on performance in manufacturing as well as in operation, availability, and cost. A selection of these materials is then subjected to standardised material testing. For a typical blade, ca. 5–10 materials may be submitted to such a test programme, with subjecting the materials to several different quasi-static test types (eg, tension, compression, in-plane shear, interlaminar shear), each of which is standardised to yield a prescribed description of material parameters (typically a stress–strain diagram and Poisson contraction). Each configuration is tested 6–10 times under identical conditions to obtain an impression of inherent scatter in the material or properties. Information regarding scatter is used in a later stage to obtain either a pre-defined safety factor or, in probabilistic design, a targeted reliability. From the best results, a few materials are then often tested in fatigue, where typically three R-ratios are investigated up to 10⁷ load cycles (the R-ratio is defined as the ratio of minimum to maximum cyclic load applied in a fatigue test; typically R = 0.1 (all tension), R = −1 (reversed loading), and R = 10 (all compression) are used). In all, a material testing programme performed in the framework of a blade certification can amount to 500 individual tests and last a few months.

A systems engineering insight that has been prevalent in (notably) aerospace engineering and is being investigated for wind turbine design is the concept of the verification escalade. This entails the reduction of uncertainties in design by testing at the material, subcomponent and full-scale levels, consecutively, for a schematic see Fig. 5.5.

In design practice up to the early 2000s, it was common practice to essentially base a rotor blade design on knowledge of the ‘bulk’ material properties, ie, material properties obtained from laboratory tests on standard coupons. This approach to a large extent disregards the interaction between different materials (composite or hybrid combinations of composites and metals) that exist in and around bondlines and connections, and in sandwich panels. This has led to designs where design details rather than laminates were critical. In European projects such as UPWIND [10], and the more recent IRPWIND [11], the notion of subcomponent testing development was elaborated [12]. A subcomponent test is a mechanical experiment on an isolated part of the structure where failure mechanisms are expected (or known to exist from full-scale blade experience), with the aim of inferring these failure mechanisms and learning about their behaviour. Typically, these are structural details for which (numerical) models cannot accurately predict their behaviour.

As such, the subcomponent testing level seems to occupy a highly necessary intermediate step between material and full-scale testing. A full-scale test raises many questions (such as: “What is the predictive power of the result of a single full-scale blade with regards to the entire production?”), and in practice fails at the critical design detail (as opposed to failing everywhere simultaneously in ideal optimal design). In this light it is striking that the development of subcomponent testing has lagged behind with respect to the aerospace industry, and has only recently been targeted for inclusion in some of the design guidelines [13]. A possible explanation for this relatively late introduction is the fact that a wind turbine rotor blade is in many ways a more integral structure than, eg, an airframe, due to the structural layout and the materials used. An aircraft can, at least from a structural point of view, be separated into a large amount of the structural parts, consisting of isotropic materials, which are each connected in a distinct fashion, often through mechanical fasteners such as bolts and rivets. This simplifies to a certain extent the division in representative test specimens, as well as the appropriate representation of boundary conditions (a riveted connection to the rest of the structure can be replaced by a bolted or riveted connection with the test machine). Other reasons may be the fast-changing designs and design scales (rendering previous test programmes on structural details less useful for next-generation designs). Furthermore, standardisation is of significant influence on the introduction of subcomponent testing in common design practice. A certain degree of standardisation of subcomponent testing is necessary for useful inclusion in the blade certification process, but the intellectual property represented in structural details specific to certain blade designs makes them unsuitable for publication in a standard, whereas generic subcomponent specimen designs and test set-ups can only be relevant for a limited range of design details.

In a recent study described in Ref. [14], the experimentally determined quasi-static and fatigue characteristics of three different laminates are used, for each laminate, to optimise the internal structure of a reference blade. The three laminates are a full-glass reinforced epoxy; a full-carbon reinforced epoxy; a hybrid laminate consisting of both glass and carbon layers. The one-million cycle stresses measured in fatigue tests in three R-ratios are shown in Fig. 5.6.

Several design limits and boundary conditions are taken into account in the blade design. For optimisation, the masses of the blade's load-carrying elements are converted to cost (applying a straightforward multiplier), giving a general impression of the potential of using alternative laminates.

An important boundary condition in this study is that only the internal structure of the blade could be modified, ie, the outside geometry was fixed to an available reference geometry. This is analogous to a design dictated by aerodynamic performance requirements. In integrated design, it is advisable to optimise the blade more globally, including a more thorough interaction between aero-elastic design and structural design.

Basic design limits were taken into account, ie:

a limit on blade bending during the maximum operational gust.

a limit on fatigue damage in the main spar flanges and trailing edge reinforcements (if any).

avoidance of buckling in the spar flanges – buckling of skin panels was not taken into account.

The optimisation of blade design for this study consisted of finding the best distribution of spar cap thickness, ie, resulting in a minimum amount of material required to resist the design loads (see Fig. 5.7).

The resulting mass and cost from the spar cap thickness distribution optimisation are shown in Fig. 5.8. Using carbon, due to its higher stiffness and better fatigue properties, the spar cap thickness can be reduced by approximately 50% by going to a hybrid laminate. Further ‘removal’ of glass to obtain an all-carbon spar cap leads to a relatively smaller decrease in mass (ca. 35%), which can be attributed partly to buckling constraints. In terms of cost, the reductions are significantly smaller than in mass, due to the higher material cost. Note that in this study a reference blade design is used, but since the hybrid and carbon versions contain significant ±45 degree layers to improve coupon test performance and infusability, the glass spar cap was modified in the simulation software optimisations to also contain ±45 degree layers. This approach facilitates comparing apples to apples, but the ‘optimised’ glass spar cap (‘Glass blade’ in Fig. 5.7) was actually 2.7 tonnes heavier than the reference (all-UD) spar cap (‘Reference blade’).

In this study, only one parameter was optimised at a time. More detailed routines might include multi-parameter optimisation, including spar cap width and laminate configuration. A reasonable amount of ‘engineering judgement’ was used in the partly manual optimisation executed for this single study. However, in a design department the more extensive optimisation routines available in some integrated design packages may be employed to maximise optimisation of automisation. The study discussed here reflects the state of the art in the sense that blade designs are only to a certain extent ‘integrated’. In most blade designs, if not all, separate elements fulfil separate tasks. As design as well as manufacturing automation might become more prevalent, further integration of functions and optimisation may occur.

More detailed parameterisation of (parts of) rotor blades for optimisation can easily result in a very large number of parameters. Even a relatively simple thickness distribution of a UD spar as shown in Fig. 5.9 has some locations where the thickness needs to be defined, even with the simplification that individual ply-drops are not taken into account.

In addition, parametrisation of an existing blade design may be hard and time-consuming and the blade definition may need to be simplified to end with a reasonable number of parameters in the optimisation. For some recent optimisation projects where layups with a complex thickness distribution had to be parametrised, a novel parametrisation method was implemented in the FOCUS6 wind turbine design software [15]. Instead of a direct parameterisation of the layup, a shape modifier was applied on the thickness distribution of the various layups in the blade. The used method is similar to the ‘Adjust Color Curves’ function as commonly found in photo-editing applications such as Photoshop and Gimp (Fig. 5.10).

With a few degrees of freedom (the markers in the graph), the distribution can be changed. The shape modifier is applied to the existing thickness distribution resulting in a new thickness distribution. The advantage of this is that the starting point of the optimisation can be a complex thickness distribution, while the shape modifier has only a few degrees of freedom, thus less parameters need to be included in the optimisation routine.

For the shape function, a third-degree dimensionless polynomial function is used, that runs from j = 0 to 1 in the domain x = 0 to 1:

If the thickness distribution in the spanwise direction is defined as t(s), the thickness distribution with the shape function applied is

The function j modifies the spanwise distribution of the thickness, while α_h scales the thicknesses.

Figs 5.11–5.13 demonstrate graphically how the thickness distribution can be altered by changing the curve of the shape modifier by adjusting α₁ and α₂ (upper left graph) and the adjustment of the scaling (α_h). In case α₁ = α₂ = α_h, the original thickness distribution is obtained. For reasonable distributions α₁ and α₂ must be kept between −1/3 and +1/3. The method is not limited to the third order polynomial function as used here.