Failure Test and Finite Element Simulation of a Large Wind Turbine Composite Blade under Static Loading

Transcription

1 Energies 2014, 7, ; doi: /en Article OPEN ACCESS energies ISSN Failre Test and Finite Element Simlation of a Large Wind Trbine Composite Blade nder Static Loading Xiao Chen 1,2,3, *, Wei Zhao 3, Xiao L Zhao 1,2,3 and Jian Zhong X 1,2, Institte of Engineering Thermophysics, Chinese Academy of Sciences, No.11 Beisihan West Road, Beijing , China; s: xlz@iet.cn (X.L.Z.); xjz@iet.cn (J.Z.X.) National Laboratory of Wind Trbine Blade Research & Development Center, No.11 Beisihan West Road, Beijing , China Engineering Research Center on Wind Trbine Blades of Hebei Province, No.2011 Xiangyang North Street, Baoding , Hebei, China; zhaoweigirl@sino-wind.com.cn * Athor to whom correspondence shold be addressed; drchenxiao@163.com; Tel.: ; Fax: Received: 9 December 2013; in revised form: 26 March 2014 / Accepted: 27 March 2014 / Pblished: 10 April 2014 Abstract: This stdy presented a failre analysis of a 52.3 m composite wind trbine blade nder static loading. Complex failre characteristics exhibited at the transition region of the blade were thoroghly examined and typical failre modes were indentified. In order to predict mltiple failre modes observed in the tests and gain more insights into the failre mechanisms of the blade, a Finite Element (FE) simlation was performed sing a global-local modeling approach and Progressive Failre Analysis (PFA) techniqes which took into accont material failre and property degradation. Failre process and failre characteristics of the transition region were satisfactorily reprodced in the simlation, and it was fond that accmlated delamination in spar cap and shear web failre at the transition region were the main reasons for the blade to collapse. Local bckling played an important role in the failre process by increasing local ot-of-plane deformation, while the Brazier effect was fond not to be responsible for the blade failre. Keywords: wind energy; blade failre; composite; delamination; local bckling; progressive failre

2 Energies 2014, Introdction With the expansion of wind energy in recent years, the sizes of wind trbine blades have become increasingly large in order to captre more power from wind and frther redce the cost of energy [1,2]. When blades are small tip deflection is a major driver in strctral design and blade failre is of less concern. However, as blade sizes grow, the types of failre change and three-dimensional stresses become important [3]. In order to investigate the failre behavior of large blades installed on mlti-megawatt (MW) wind trbines, some researchers have carried ot failre tests on fll-scale blades. Among them, Jorgensen, et al. [4] performed a failre test on a 25 m blade nder flap-wise bending emphasizing the geometrical nonlinearity at large deflections. Jensen et al. [5 7] tested a 34 m blade and its load-carrying box girder nder flap-wise bending and fond that the Brazier effect indced a large deformation in the spar cap and the frther delamination bckling were the case leading to strctral collapse. Overgarrd et al. [8,9] tested a 25 m blade to failre nder flap-wise bending and conclded that the ltimate strength of the blade was governed by instability phenomena in the form of delamination and bckling instead of the Brazier effect. Yang et al. [10] stdied the strctral collapse of a 40 m blade nder flap-wise bending and fond that debonding of aerodynamic shells from adhesive joints was a main reason for the blade to collapse. Cho et al. [11] investigated a typhoon-damaged composite blade with a blade length close to 39.5 m and showed that the blade failed by delamination and laminate cracking. It can be seen that failre cases varied among blades and little consenss has been reached. Althogh several failre incidents of 50+ m blades have been reported in recent years [12], no experimental stdy focsed on the failre of blades with lengths longer than 40 m has been pblically reported so far according to the athors knowledge. It is considered that larger blades may exhibit more complex failre behavior which has neither been observed from the existing stdies nor been paid enogh attention to in crrent blade design. Therefore, there is an rgent need of experimental stdies on large blades to gain more nderstanding on their failre behavior. Frthermore, nmerical models capable of predicting blade failre in a large strctral scale are also essential to complement experimental stdies de to the expensive cost associated with fll-scale blade tests. Althogh some researchers [13 15] have sccessflly investigated failre behavior of composite blades sing the Finite Element (FE) method, they primarily focsed on in-plane failre mode, i.e., laminate fractre (LF), for which failre criteria based on a two-dimensional stress field are readily available in most commercial FE programs. As observed from the existing failre tests on large composite blades, throgh-thickness failre mode in the form of delamination appears to be more crcial than the in-plane one. Only a few works, i.e., [8,9], have investigated delamination of blades nmerically in a strctral scale and the FE program they sed was developed in-hose. It is therefore of significance to develop a modeling method based on commonly sed FE programs with which blade designers are familiar for predicting the failre behavior of large blades. In order to partially meet the need of experimental and nmerical stdies on the failre of large blades, this stdy presents a fll-scale failre test of a 52.3 m blade with particlar focs on its complex failre characteristics which were associated with both in-plane and throgh-thickness failre modes and have not been investigated by other stdies focsing on smaller blades. In order to predict the complex failre characteristics of the blade, this stdy proposes a FE modeling method based on a

3 Energies 2014, commercial FE program and frther investigated the mechanisms leading to the blade failre. It was expected that more nderstanding on the failre of large wind trbine blades cold be obtained from this stdy. The contents of this paper are organized as follows: general information and experimental procedres of the blade were described and then failre characteristics were investigated in Section 2; in Section 3 a FE modeling method which was capable of predicting complex failre characteristics of the blade was proposed taking advantage of Progressive Failre Analysis (PFA) techniqes and a global-local modeling approach; after assessing nmerical models by comparing predicted strctral response with available experimental measrements, failre process and failre characteristics of the blade were reprodced in Section 4; based on failre test and nmerical simlation the root case of failre and the Brazier effect were examined to nderstand the failre mechanisms of the blade in Section 5; finally, the findings obtained from this stdy are smmarized in Section Fll-Scale Failre Test Commercial wind trbine blades have to pass fll-scale tests reqired by blade certification bodies [16,17] in order to start a series prodction of those blade types. Static bending loads are applied in the test to simlate extreme wind loads the blades are expected to sstain in their design lives. This section presents general information and the experimental procedre of the blade nder concern and investigated its complex failre characteristics throgh post-mortem observations General Information of the Blade The blade was designed for 2.5 MW wind trbines and had a total length of 52.3 m. It was made of glass fabrics and vacm infsed with epoxy resin. Typical constrction of the blade cross-sections is shown in Figre 1. Spar caps were made of composite laminates and designed to carry bending moments, while leading edge (LE) panel and aft panel were made of sandwich constrctions with PVC foam cores and designed to provide an aerodynamic airfoil shape. Shear webs were also sandwich constrctions and designed to spport two spar caps and transfer shear forces. Figre 1. Typical constrction of blade cross-section. Spar cap SS Aft panel TE Shear webs LE LE panel Spar cap PS Throghot this paper span-wise locations of the blade were normalized by the total length of the blade and denoted as Lxxxx. For example, a span-wise location of 4 m was denoted as L0.076 in

4 Energies 2014, Figre 2, which shows composite layps at representative blade locations. The blade srfaces contained two types of triaxial laminates denoted as Triaxial_7 and Triaxial_1 ; spar caps contained a large amont of nidirectional laminates and shear webs contained biaxial laminates as sandwich skins. Figre 2. Representative composite layps of the blade. Normalized blade span (-) LE panel Spar cap Aft panel Shear web L0.076 L0.268 Triaxial_7 Triaxial_1 PVC foam Unidirectional Biaxial 2.2. Experimental Procedre In general, two sets of static bending tests were performed on this blade, the first set was based on design loads for 2.5 MW wind trbines, and it followed the test reqirements for fll-scale wind trbine blades according to certification bodies. The second set was based on design loads for 3.0 MW wind trbines and it was performed with the intention to overload the blade and examine possible failre. All test cases and their seqence are shown in Table 1. Loading directions are schematically shown in Figre 3 with typical cross-sectional profiles of the blade sperposed. It is noted that the test seqence of the blade was determined to comply with an order of conterclockwise rotation of the blade axis. The blade was root-fixed on a test stand as a cantilever beam and cranes applied pward plling forces as shown in Figre 4 for the test case F_Max_3.0, after one test case was finished, the blade was rotated along its longitdinal axis to a prescribed position for a sbseqent test case. Table 1. Load test history of the blade. Test seqence Test case Trbine Class Loading direction 1 F_Min_2.5 Flap_Min 2 E_Min_2.5 Edge_Min 2.5 MW 3 F_Max_2.5 Flap_Max 4 E_Max_2.5 Edge_Max 5 F_Min_3.0 Flap_Min 6 E_Min_3.0 Edge_Min 3.0 MW 7 F_Max_3.0 Flap_Max 8 * E_Max_3.0 Edge_Max * Test case E_Max_3.0 was not tested de to the final blade failre in F_Max_3.0.

5 Energies 2014, Figre 3. Loading direction of test cases. Flap_Max Cross section at Root Cross section at Max. Chord Edge_Max Edge_Min Cross section at 0 degree initial twist Cross section at Tip Flap_Min Figre 4. Test scene of the blade in the test case F_Max_3.0. Three cranes and for cranes were sed for 2.5 MW loading and 3.0 MW loading, respectively. Rbber pads with approximately 200 mm width and 15 mm thickness were installed between blade srfaces and loading saddles to redce stress concentration at loading locations. The target test bending moments applied to the blade were normalized by the maximm root moment in the test case F_Max_2.5 and they are shown in Figre 5. It can be seen that the target test loads of F_Max_2.5 were basically same as those of F_Max_3.0. Test loads were applied in a stepwise form following a loading procedre of 0%, 40%, 60%, 80%, 100% and then an nloading procedre of 80%, 60%, 40%, 0% of the target loads in each test case. There was no commnication among cranes which applied plling force simltaneosly to obtain each prescribed load level and then held for ten seconds as reqired by [16,17]. Dring the test, applied loads were recorded by load cells monted on the cranes, deflections of the blade were measred at loading saddle locations and at blade tip sing draw-wire displacement transdcers, longitdinal strains were recorded by strain gages located along the middle axis of spar caps. Average vales of measrements dring ten seconds were sed to represent the strctral response of the blade at the prescribed load levels.

6 Energies 2014, Figre 5. Applied target bending moments. 1.2 F_Min_2.5 F_Min_3.0 Normalized test moment (-) E_Min_2.5 F_Max_2.5 E_Max_2.5 E_Min_3.0 F_Max_3.0 E_Max_ Post-Mortem Observations and Discssion Normalized blade span (-) After the sccessfl completion of the 2.5 MW loading set for blade certification, the 3.0 MW loading set was followed. At 100% of the target test loads in the test case E_Min_3.0, trailing edge (TE) and the adjacent aft panels failed at a transition region where the cross-sectional geometry of the blade transited from a circlar shape at the blade root to an airfoil shape at the maximm chord. After nloading the test case F_Max_3.0 was followed and the blade collapsed de to a final failre at the transition region at arond 90% of the target test loads. Becase the target test loads of F_Max_2.5 and F_Max_3.0 were basically the same, the final failre load of the blade was actally 10% smaller than the load once carried by the blade. Thorogh investigation was then performed on the entire blade. It was fond that major failre regions were located from L0.067 to L0.105 as shown in Figre 6a. Althogh failed regions exhibited a complex form of failre, some typical failre modes, i.e., laminate fractre (LF), composite delamination (DL), and sandwich skin-core debonding (DB) cold be identified as shown in Figre 6b,c where visally apparent debonding fronts were depicted with dashed crves, and intersections between spar cap and sandwich panels were depicted with dashed lines. A close-p of the sction side (SS) of the blade at L0.076 shows local failre characteristics in Figre 6d, and it can be seen that LF occrred at the intersection between the spar cap and aft panel, and skin laminates of spar cap were sbjected to LF as well as DL.

7 Energies 2014, Figre 6. External failre of the blade. (a) Final failre of the blade at the transition region; (b) View#1 Typical failre modes at SS; (c) View#2 Typical failre modes at SS; (d) Local failre featres at L Top: SS Front: LE side Bottom: PS (a) DL & LF DB L0.076 DB TE side LF DB LF L0.096 LE side Spar cap (b) DB DL & LF LF L0.076 TE side DB LF Aft panel DL L0.096 LF LF LF DB Spar cap (c) (d) Inspection on the blade interior was also made, and it was fond that the rear shear web at the location from L0.076 to L0.096 was completely failed with an approximate fractre angle of 45 to the blade axis, see Figre 7a. TE at L0.092 exhibited considerable failre with combined failre modes of LF, and DB as shown in Figre 7b.

8 Energies 2014, Figre 7. Internal failre of the blade. (a) Rear shear web failre; (b) TE failre. (a) (b) Frthermore, the blade was sectioned at L0.076 to facilitate examination on its cross-section. It was fond that DL of spar cap occrred not only at triaxial laminates constitting blade skins bt also at nidirectional laminates as shown in Figre 8a. Investigation on aft and LE panels revealed DB and associated core shear failre (CF) at sandwich constrctions and the failre srface of DB propagated at a region roghly 1 to 2 mm nderneath the interface between skin laminates and core materials as shown in Figre 8b. Figre 8. Failre observed at L0.076 cross section. (a) The SS spar cap; (b) LE panel. DL in triaxial laminate CF Laminate skin Foam core DB DL in nidirectional laminate (a) (b) From post-mortem observations, it was obvios that the transition region was pervaded by mltiple failre modes. Among them DL and DB were essentially related to mechanical properties and stress states of interfaces between constitent layers and they cold be categorized as interfacial (or throgh-thickness) failre. Another failre mode LF althogh commonly considered to be a form of in-plane failre was fond at the intersection where geometric and material discontinities indced interlaminar stresses inevitably affected the characteristics of LF. Becase both interfacial and interlaminar stresses are closely associated with throgh-thickness stresses, it can be conclded that

9 Energies 2014, throgh-thickness stresses played an important role in the complex failre characteristics with mltiple failre modes of the blade. Frthermore, it is noted that spar caps with a large amont of nidirectional laminates governed the bending stiffness and the overall strength of the blade, and shear webs spported two halves of aerodynamic shells and transferred shear forces in the blade. From post-mortem observations, it can be seen that among varios failre modes exhibited at the transition region delamination of nidirectional laminates in the SS spar cap and the failre of shear web were most responsible for the complete loss of the load-carrying capacity of the blade. 3. FE Modeling Method Althogh post-mortem observations provided some nderstanding of the failre characteristics of the blade, the failre process and failre mechanisms have not been clarified yet. In order to gain more insights into the failre of the blade, a nmerical model is necessary to complement the experimental stdy. This section, aimed at developing a FE modeling method for nmerical stdy, first presents the considerations that have to be taken into accont in the simlation and then describes material models sed in the PFA techniqes and a global-local approach for blade modeling. Frthermore, nmerical implementation of the simlation in a commercial FE program is also presented Considerations in FE Modeling Based on the failre observations from the test, it was evident that two featres have to be incorporated in FE modeling in order to accrately simlate the failre process and failre characteristics of the blade. The one featre is that the FE model to be constrcted shold consider the effect of loading history on strctral performance as the final failre load was fond to be smaller than the load once carried by the blade, implying the adverse effect of loading history on its ltimate load-carrying capacity. This featre can be achieved by sing well-established PFA techniqes [18] which consist of two parts, i.e., failre criteria indicating occrrence of material failre and material property degradation rles representing residal properties of material once failre criteria are satisfied. Althogh the PFA techniqes have been sed in many works [19,20] to investigate failre processes of composite materials and strctral components, they were rarely applied to analyze large composite blades on a strctral scale. In the crrent stdy the PFA techniqes were sed to simlate the loading history of the blade with material failre and property degradation. The other featre that needs to be inclded in the FE model are the three-dimensional stresses which enable analysis of complex failre characteristics associated with both in-plane and throgh-thickness stresses, and this reqires solid elements to be sed in the model. However, modeling and analyzing an entire blade strctre sing solid elements is extremely difficlt de to the large differences in size scales of the component details and the entire blade. Considering that the blade failre occrred primarily at the transition region, only a local part of the blade had to be modeled and analyzed sing solid elements. Althogh the local solid element model appeared to be a promising soltion to redce modeling difficlty, the actal loads applied to the entire blade instead of the transition region were explicitly known. In order to obtain loads sstained by the transition region, it was decided to first constrct a global shell element model of the entire blade representing its load stats in the test and

10 Energies 2014, back-calclate loads sstained by the transition region, and then these loads were applied to the local solid element model and detailed failre analysis was sbseqently performed Material Models Material Properties Experimental reslts of in-plane properties from material tests are shown in Table 2. De to commercial concerns from blade designer, all modls and strength parameters of materials were normalized by the corresponding longitdinal properties of each material. For the local solid element model, throgh-thickness properties were needed to perform the analysis. However, material properties in the throgh-thickness (or 3 ) direction were not tested in this stdy, so some assmptions were sed to estimate these vales as shown in Table 3 where the properties of nidirectional composites were regarded to be proportional to their in-plane ones, and the proportional factors were determined according to [21], in which a similar glass/epoxy nidirectional thick laminate has been systematically tested. The strength properties of two types of triaxial composites in the 3 direction were assmed to be the same as those of nidirectional composites considering that there was no fiber reinforcement aligned in this direction and strength properties were largely controlled by the matrix in composites. Table 2. Material in-plane properties sed in both the global shell and local solid element models. Name Elastic constants Strength parameters E 11 E 22 G 12 μ 12 σ 11t σ 11c σ 22t σ 22c σ 12 Unidirectional Triaxial_ Triaxial_ Biaxial PVC foam Where sbscripts 11 and 22 mean longitdinal and transverse direction, respectively; sbscript 12 means in-plane shear direction; sbscripts t and c mean tensile and compressive direction, respectively; sperscript indicates ltimate strength parameters. Table 3. Material throgh-thickness properties sed in the local solid element model. Name Elastic constants Strength parameters E 33 G 13 G 23 μ 13 μ 23 σ 33c σ 33t σ 13 σ 23 Unidirectional E G 12 G 12 μ μ σ 22c 0.71σ 22t 0.97σ σ 12 Triaxial_7 E 22 G 12 G 12 μ 12 μ 12 = = = = Triaxial_1 E 22 G 12 G 12 μ 12 μ 12 = = = = Biaxial N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. PVC foam E 22 G 12 G 12 μ 12 μ 12 σ 22c Note: = means that the vale is the same as that for nidirectional composite; N.A. means not applicable, it is becase biaxial materials are sed in shear webs, which are modeled with shell elements, and their throgh-thickness properties are not needed in the FE model. In order to predict interfacial failre, i.e., DL and DB, in the solid element model, material properties of interfaces need to be known. In this stdy, all elastic constants and strength parameters of σ 22t σ 12 σ 12

11 Energies 2014, interfaces between two composite lamina layers were assmed to be eqal to the average vales of material properties of two lamina layers. As for interfaces between composite lamina and PVC foam core, material properties were assmed to be the same as PVC foam considering that the failre srface of skin-core debonding occrred in PVC foam as observed in the test Failre Criteria Three typical failre modes, i.e., LF, DL, and DB, observed in the failed blade were considered in FE modeling. For LF, the Tsai-W failre criterion [22] in a three-dimensional stress field, as expressed in Eqation (1), was employed in this stdy. The criterion is widely sed to predict LF and has several advantages. It allows for interaction among stress components (analogos to the von Mises criterion for isotropic materials), it takes into accont differences between tensile and compressive strength, and it is operationally simple and readily amenable to comptational procedres [23]: F.I. 3d = F 1 σ 11 + F 2 σ 22 + F 3 σ F 12 σ 11 σ F 13 σ 11 σ F 23 σ 22 σ 33 + F 11 σ F 22 σ F 33 σ F 44 σ F 55 σ F 66 σ 12 2 (1) where, F 1 = 1/(σ 11t σ 11c ), F 2 = 1/(σ 22t σ 22c ), F 3 = 1/(σ 33t σ 33c ), F 12 = 0.5/(σ 11t σ 11c σ 22t σ 22c ) 0.5, F 13 = 0.5/(σ 11t σ 11c σ 33t σ 33c ) 0.5, F 23 = 0.5/(σ 22t σ 22c σ 33t σ 33c ) 0.5, F 11 = 1/(σ 11t σ 11c ), F 22 = 1/(σ 22t σ 22c ), F 33 = 1/(σ 33t σ 33c ), F 44 = 1/(σ 23 ) 2, F 55 = 1/(σ 13 ) 2, F 66 =1/(σ 12 ) 2 ; and F.I. 3d was defined as failre index in a three-dimensional stress field, and a vale larger than 1 indicates a composite failre; σ 11, σ 22, and σ 33 are normal stresses of composites in longitdinal, transverse and thickness direction, respectively; σ 12, σ 23, and σ 13 are shear stresses in in-plane, transverse, and axial direction, respectively. The sperscript means ltimate strength of the material, and sbscripts t and c mean tensile and compressive direction, respectively. This failre criterion is mode-independent as it does not directly identify the natre of material damage. In order to apply appropriate material degradation rles, the natre of LF needs to be differentiated. According to Christensen [24], the term F 1 σ 11 + F 11 σ 11 2 indicates a fiber-controlled failre and it was employed to identify the fiber (or matrix)-controlled failre mode if it is larger (or smaller) than the smmation of other terms in Eqation (1) when failre indices exceed 1. DL and DB were regarded as two different failre modes emphasizing their locations of occrrence, whereas they shared a similar failre featre that two constitent layers separated and the properties of their interface dominated failre behavior. Therefore, a same qadratic failre criterion based on Ye [25] as expressed in Eqation (2) was employed to predict the occrrence of DL and DB: F.I. iff = (σ 33 + /σ 33t ) 2 + (σ 13 /σ 13 ) 2 + (σ 23 /σ 23 ) 2 (2) where, F.I. iff was defined as a failre index of interfacial failre, and a vale larger than 1 indicates an interfacial failre, σ 33 +, σ 13, and σ 23 are normal tensile stress, axial shear stress, and transverse shear stress, respectively; σ 33t, σ 13, and σ 23 are the corresponding ltimate strength. It is noted that all stress variables and strength parameters refer to the interface of two constitent layers.

12 Energies 2014, Material Degradation Rles Once failre criteria were satisfied, material properties were degraded according to certain rles representing the effect of material damage. Some material degradation rles in the form of mltiplying factors [26 28] as shown in Table 4 have been sed in the failre analysis of composites, and these mltiplying factors were applied to the intact material properties to obtain the degraded ones once material failre criteria were met. In the present stdy, the degradation rle adopted by Apalak et al. [27] was sed with changing G 13 vale for the matrix-controlled LF from 0 to 1 regarding insignificant effect of matrix failre on transverse normal shear modls. The vale of zero was replaced by a very small vale in order to avoid nmerical errors in the FE analysis. The mltiplying operation was only performed once at the first occrrence of the corresponding failre mode and the properties of materials hereafter remained constant, representing the irreversible natre of material damage. Table 4. Material degradation rles sed for PFA. Failre mode Model E 11 E 22 E 33 G 12 G 13 G 23 μ 12 μ 13 μ 23 Nagesh [26] LF Z.G. Apalak et al. [27] (Fiber-controlled) K.I. Tserpes et al. [28] Present LF (Matrix-controlled) DL Nagesh [26] Z.G. Apalak et al. [27] K.I. Tserpes et al. [28] Present Nagesh [26] Z.G. Apalak et al. [27] K.I. Tserpes et al. [28] DL&DB Present FE Models The Global Shell Element Model The global shell element model was constrcted in order to obtain loads sstained by the transition region by simlating a fll-scale blade in the test. In this stdy, the general FE program Abaqs [29] which is commercially available was sed. In total, abot 47,300 S4R shell elements with a typical mesh size of 100 mm 100 mm were sed in the model. S4R is a 4-node, qadrilateral, stress/displacement shell element with redced integration and a large-strain formlation. Before the final mesh size was determined, a mesh convergence stdy has been condcted. It was fond that when the blade was modeled with typical mesh sizes of 150 mm 150 mm, 100 mm 100 mm, and 50 mm 50 mm, the reslts of the first natral freqency between 100 mm 100 mm mesh and 50 mm 50 mm mesh were below 1%, and the reslts of the first bcking eigenvale between two meshes were below 2%, therefore, a mesh size of 100 mm 100 mm was deemed sfficient.

13 Energies 2014, Layered orthotropic materials were applied to the blade model according to the layp scheme for blade manfactring. A fixed bondary was applied at the blade root and point loads with resltants eqal to the forces recorded from load cells were eqally distribted on the srfaces where spar caps of the blade and rbber pads in the loading saddles are contacted representing actal loading setp sed in the test, as shown in Figre 9. Loads were applied incrementally p to the target test loads and geometric nonlinearity was taken into accont in the model to captre large deformation. Three reaction forces and three reaction moments at the blade root were extracted in every load step of each test case. Figre 9. The global shell element model with applied loads in the test case F_Max_3.0. Applied load Applied load Applied load Applied load Fixed bondary The Local Solid Element Model In the present work, major blade failre of primary interest was located at the transition region approximately from L0.067 to L0.105 of the blade. Therefore, only the inboard part of the blade p to L0.134 was constrcted in a local FE model. Both shell elements for shear webs and solid elements for the rest were sed in the model. Nevertheless, this FE model was regarded as the local solid element model considering its different capacity to predict blade failre from the global shell element model. There were abot 25,900 solid C3D8RC3 elements ot of the total 33,000 elements sed to mesh the local part. C3D8RC3 is a 8-node, three-dimensional, linear, continm stress/displacement element with redced-integration. A typical mesh size was 45 mm 45 mm for shell elements and 45 mm 45 mm t i for solid elements, and t i varied according to the local thickness of blade external shells. Only one solid element was sed in the throgh-thickness direction of the blade shells. The root of the local part was fixed and mltiple-point constraints (MPC) [29] were applied to a loading section at L0.134 to introdce appropriate loads throgh a reference point, which was positioned to have same cross-sectional coordinates as the center of the circlar root section as shown in Figre 10.

14 Energies 2014, Figre 10. The local solid element model with applied loads back-calclated from shell element model. Root section Fixed bondary TE side Reference point for loading F z LE side MPC Loading section F y M y M z M x F x 3 point loads 3 point moments Three point loads and three point moments were applied to the reference point in sch a way that each reaction force and reaction moment yielded at the blade root was eqal to the corresponding ones obtained from the global shell element model which was loaded according to the actal experimental setp. Layered orthotropic materials were assigned to the model. In one solid element, composite laminates were sbdivided into several lamina layers with actal ply thicknesses and two types of interface layers with a negligible thickness of mm were embedded between lamina layers, i.e., Type 1 interface layer, as well as between composite material and PVC foam core, i.e., Type 2 interface layer, as shown in Figre 11. Interface layers were employed in the solid element model in order to captre interfacial failre modes, i.e., DL and DB, in the FE analysis. Figre 11. Schematic presentation of layered featres in a solid element. 3 direction 3 direction Lamina layer Type 1 interface layer Type 2 interface layer Core Local layers One solid element Layered materials Lamina layer Type 1 interface layer Local layers

15 Energies 2014, Nmerical Implementation of Simlation The PFA techniqes were implemented in the solid element model to captre the complex failre characteristics observed at the transition region by incorporating ser sbrotines into Abaqs according to the proposed FE modeling method. The occrrences of material failre were predicted sing prescribed failre criteria and material properties pon failre were degraded according to degradation rles in a three-dimensional stress field. All seven test cases were analyzed in a single nmerical rn following the loading history of the blade. The Riks method was adopted in order to help static eqilibrim in the analysis with considerable geometric and material nonlinearities. Material failre indices larger than 1 were pdated and recorded in each load increment resembling the progress of material failre in the blade. 4. Simlation Reslts and Discssion This section presents reslts obtained from FE simlation. Two FE models sed in the global-local modeling approach were assessed by comparing predicted reslts with experimental measrements available in the 2.5 MW loading set, and then failre process and the final failre characteristics of the transition region were predicted. Frthermore, taking advantage of the local solid element model, failre mechanisms was investigated and the Brazier effect was clarified in this section Model Assessment Shell Model Assessment In order to ensre that the global shell element model was able to represent the loading stats of the blade in the test, nmerical predictions of blade deflections and strains were compared with experimental measrements in the 2.5 MW loading set. Representatively, deflection measrements of the test case F_Max_2.5 are shown in Figre 12a, where nmerical predictions from the global shell FE model are also plotted. Longitdinal strains along spar caps at both pressre side (PS) and sction side (SS) of the blade are shown in Figre 12b. For for test cases in the 2.5 MW loading set, longitdinal strains at SS of the blade at L0.126 are shown in Figre 12c, where arrow signs on dashed lines indicate loading and nloading paths. From these comparisons, it was fond that the predicted global deflections and longitdinal strains were in satisfactory agreement with experimental measrements despite relative large difference in strains at one location as indicated with a dashed circle in Figre 12b. Therefore, the global shell element model was considered to be well constrcted and can be sed to calclate loads sstained by the transition region.

16 Energies 2014, Figre 12. Comparison of strctral response of the blade in the 2.5 MW loading set. (a) Deflection in the test case F_Max_2.5 ; (b) Longitdinal strains in the test case F_Max_2.5 ; (c) Longitdinal strains at the SS spar cap at L Deflection (m) FEA_Shell,40% FEA_Shell,60% FEA_Shell,80% FEA_Shell,100% Test, 40%, Loading Test, 60%, Loading Test, 80%, Loading Test, 100%, Loading Test, 80%, Unloading Test, 60%, Unloading Test, 40%, Unloading Normalized blade span (-) (a) 120 Strain ε 11 (μ) 5,000 PS 3,000 1,000-1, ,000 SS -5,000 Normalized blade span (-) (b) Applied load fraction (%) Solid Model Assessment Flap_Min_Test 60 Edge_Min_Test Flap_Max_Test 40 Edge_Max_Test Flap_Min_FEA Edge_Min_FEA 20 Flap_Max_FEA Edge_Max_FEA 0-6,000-3, ,000 6,000 9,000 Strain ε 11 (μ) (c) After applying three point loads and three point moments which back-calclated from the global shell element model to the reference point in the local solid model, the PFA was carried ot to reprodce the loading history. In order to assess the capability of the solid element model to represent the transition region of the blade in the test, the model shold have been validated against experimental reslts, however, de to test data availability at this region, the solid element model was alternatively compared to the shell element model which has been proved to be able to predict deflections and longitdinal strains of the blade with reasonable accracy. The test case F_Max_2.5 was considered for the comparison representatively as shown in Figre 13 where available experimental measrements were plotted. It was fond that both blade deflections and longitdinal strains obtained from the global shell element model and the local solid element model showed general agreement, despite relative large differences in strain at a few locations. Based on the comparison, it was regarded that the solid element model was loaded appropriately and was able to represent the transition region in the actal test.

17 Energies 2014, Figre 13. Comparison of reslts obtained from shell and solid element models. (a) Blade deflection; (b) Longitdinal strains on spar cap FEM_Solid,40% FEM_Solid,60% 5,000 PS Deflection (m) FEM_Solid,80% FEM_Solid,100% FEM_Shell,40% FEM_Shell,60% FEM_Shell,80% FEM_Shell,100% Strain ε 11 (μ) 3,000 1,000-1, , Normalized blade span (-) -5,000 SS Normalized blade span (-) (a) (b) 4.2. Failre Prediction Predicted Failre Process Using the proposed modeling method, progressive failre process of the transition region was predicted. De to massive reslts generated in the simlation, only reslts at the peak load of each test case were presented in this stdy. The locations with failre indices larger than 1 indicating material failre are shown in Figre 14. Figre 14. Progressive failre at the peak load of each test case. Test seqence Test case F_Min_2.5 E_Min_2.5 F_Max_2.5 E_Max_2.5 F_Min_3.0 E_Min_3.0 F_Max_3.0 LF DL & DB

18 Energies 2014, It was fond that LF was negligible ntil the test case E_Min_3.0, whereas interfacial failre in the form of DL and DB has been developed at a preceding test case F_Max_2.5 when SS of the transition region was primarily sbjected to compressive force de to applied flap-wise bending. These interfacial failre contined to progress arond TE and spar cap at a sbseqent test case E_Max_2.5. When edge-wise bending was applied in the test case E_Min_3.0, LF can be distinctly observed at TE and aft panels and interfacial failre spread over the transition region at varios locations. Sbseqently, when SS of the transition region was once again sbjected to compressive force in the test case F_Max_3.0, LF progressed rapidly and interfacial failre was also frther developed Predicted Final Failre Characteristics The predicted final failre of the transition region was presented in a perspective view in Figre 15 for more discssion. It can be seen from Figre 15a that major regions with LF were located at SS of the blade from L0.061 to L0.086, which agreed satisfactorily with the experimental observations. Moreover, some complex failre characteristics fond in the test were also well captred by the simlation, sch as LF at the intersection between spar cap and aft panel. LF was also predicted to occr at intersecting regions of blade root, spar cap and aft panel althogh this was not visally detected in the test. Figre 15. The final failre predicted from FE simlation. (a) LF at blade shells; (b) DL and DB failre; (c) LF at the rear shear web <1.00 #1 #3 #4 # <1.00 #2 #1 #4 #3 Failre location: #1. Spar cap #2. Aft panel #3. Intersection of spar cap and aft panel #4. Intersection of blade root, spar cap and aft panel (a) l Failre location: #1. Spar cap #2. Aft panel #3. TE and aft panel #4. Intersection of blade root and aft panel (b) 1.00 <1.00 Front shear web 0.80 <0.80 Front shear web L0.096 Rear shear web 50 Rear shear web Blade axis (c) Potential failre line

19 Energies 2014, For interfacial failre as shown in Figre 15b, significant DL was fond at spar cap, TE and the adjacent aft panel where similar failre characteristics were observed in the test althogh the scale of failre locations appeared to be overestimated. LF of the rear shear web was predicted to initiate at the top of the rear shear web at L0.096 as shown in Figre 15c. The failre region with failre indices larger than 1 was relatively small, bt a contor with failre indices larger than 0.8 showed that a potential failre line originated from the initial failre location and inclined approximately 50 degree to the blade axis, which basically agreed with failre characteristics fond from the post-mortem observations. In order to examine the effect of different material degradation rles on the final failre prediction, the present material degradation rle was replaced with that sed by Nagesh [26] and one new simlation was performed. The contors of the final failre obtained from two analyses were compared in Figre 16. It can be seen that failre locations predicted from the analysis with the Nagesh rle were relatively larger than those with the present one. For LF the locations with larger failre were spar cap and aft panels, and for interfacial failre, they were TE and the adjacent aft panel. Althogh there was a difference in the scale of the failre locations, the two analyses predicted similar reslts in terms of major failre locations and failre characteristics fond at the transition region of the blade. Figre 16. Effect of different material degradation rles on the final failre prediction. (a) LF_Present; (b) FL_Nagesh [26]; (c) DL&DB_Present; (d) DL&DB_Nagesh [26] < <1.00 (a) (b) (c) (d) 5. Failre Mechanisms This section presented investigation on the root case of blade failre according to experimental observations and nmerical simlation. The Brazier effect, which has been arged for its role in the blade failre among several stdies [6,8,10] was also examined sing the nmerical model Root Case of the Blade Failre Taking advantage of experimental observations and nmerical simlation, the mechanisms leading to the final failre were investigated. In the 2.5 MW loading set, the transition region accmlated material damage in the form of interfacial failre, i.e., DL of nidirectional laminates in spar cap and DB in sandwich constrctions. When the 3.0 MW loading set with large load amplitdes in the first

20 Energies 2014, two test cases, i.e., F_Min_3.0 and E_Min_3.0, were applied to the blade, interfacial failre was frther developed in spar cap and TE, and LF also occrred at aft panel and TE. At the test case F_Max_3.0, in which SS of the blade was once again sbjected to compressive force in flap-wise bending, the previosly developed LF contined to progress arond the spar cap region with significant interfacial failre reslting in a drastic decrease of the bending stiffness and the overall strength. Meanwhile, the rear shear web started to fail frther redcing its capacity to transfer shear forces. As a conseqence, the blade lost its load-carrying capacity at the transition region and the final failre was witnessed. Therefore, accmlated DL of nidirectional laminates in spar cap and the following failre of shear web were considered to be the root case of the blade failre The Brazier Effect The Brazier effect was examined sing the nmerical model. The deformation of blade cross sections at the final failre are shown in Figre 17. The global flattening of these cross sections, which illstrates the Brazier effect, was not fond in the deformed cross-sectional profiles possibly de to the constraint from shear webs between two spar caps, whereas the relative large ot-of-plane deformation indicating local bckling was observed in the form of aft panel blging otward and TE caving inward. Figre 17. Deformation (scaled by 5) of cross-sectional profiles at the final failre. (a) L0.067; (b) L0.076; (c) L0.086; (d) L (a) L0.067 (b) L0.076 (c) L0.086 (d) L0.096 Undeformed Deformed The enlarged deformation of spar cap is shown in Figre 18, and it was fond that the SS spar cap, which was initially convex in the transverse direction, became flattened and exhibited bckles where interfacial failre was significant. The relative ot-of-plane deformation was defined as (d i d 0 )/d 0 100%, where d i and d 0 are the ndeformed and deformed arc heights, respectively, of the spar cap at SS as shown in Figre 19. Using this relative ot-of-plane deformation, the local bckling response of spar cap was assessed qantitatively, and reslts at for locations of spar cap are shown in Figre 20, where initial slopes of the for crves were also plotted. It was observed that at certain load levels the crves of the relative ot-of-plane deformation changed their trends and their nonlinearity became significant, and these load levels were determined as local bckling loads which had applied load fractions ranging approximately from 70% to 80%. The local bckling was expected to significantly increase ot-of-plane deformation and throgh-thickness stresses which sbseqently deteriorated DL of spar cap. Therefore, nonlinear

21 Energies 2014, ot-of-plane deformation de to local bckling affected the failre process adversely and contribted to the blade failre. Figre 18. The SS spar cap from L0.049 to L0.086 at the final failre (deformation scaled by 10). (a) Undeformed; (b) Deformed with LF; (c) Deformed with DL. (a) (b) (c) Figre 19. Parameters for the relative ot-of-plane deformation of spar cap at SS. Undeformed Deformed d i d0 1/2 L t 1/2 L t Transverse span of spar cap L t Figre 20. Relative ot-of-plane deformation of the SS spar cap in the test case F_Max_3.0. Applied load fraction (%) Local bckling load L0.067 L0.076 L0.086 L0.096 Blade final failre load Relative ot-of-plane deformation (%)

22 Energies 2014, Conclsions This paper presented a thorogh investigation into the failre of a 52.3 m composite wind trbine blade sbjected to a series of static loads. Particlar focs was placed on its complex failre characteristics observed in the test and a nmerical model developed to predict failre behavior of the blade. Main conclsions can be drawn as follows: the blade failed catastrophically de to a total loss of load-carrying capacity nder flap-wise bending. The final failre load was only 90% of the load once carried by the blade. Major failre was located at the transition region and characterized by mltiple failre modes associated with both in-plane and more significant throgh-thickness stresses. Typical failre modes observed were delamination of the spar cap at the sction side, sandwich skin-core debonding, laminate fractre, and shear web failre. Among these failre modes spar cap delamination and shear web failre were indentified to be the root case of the blade failre. The loading history of the blade was nmerically reprodced and the complex failre characteristics observed at the transition region were satisfactorily predicted in the FE simlation sing the PFA techniqes and a global-local modeling approach. It was fond that the blade accmlated failre dring its loading history. Local bckling contribted to the failre process by increasing the local ot-of-plane deformation of spar cap, while the Brazier effect was not fond in the transition region and it was not responsible for the failre of the blade. This stdy emphasized the significance of throgh-thickness stresses on the failre of large composite blades. The proposed global-local modeling approach was proved to be effective to captre failre behavior of the blade while considerably redcing difficlties in modeling an entire blade. Acknowledgments We are gratefl to several anonymos reviewers for their constrctive comments that have led to significant improvement of the paper. Nomenclatre Blade geometry Failre mode LE Leading edge LF Laminate fractre TE Trailing edge DL Delamination SS Sction side DB Sandwich skin-core debonding PS Pressre side CF Core failre Others MW Megawatt FE Finite Element PFA Progressive Failre Analysis Athor Contribtions The analysis and interpretation of experimental reslts and FE modeling were carried ot by Xiao Chen, who was also responsible for writing the manscript of this paper. Experimental test was performed nder the spervision of Wei Zhao. This work was carried ot nder the advisement of and with reglar feedback from Xiao L Zhao and Jian Zhong X, who also revised the manscript critically.

23 Energies 2014, Conflicts of Interest The athors declare no conflict of interest. References 1. Griffith, D.T. Technology Trends Presented at the 5th Sandia Wind Trbine Blade Workshop, Albqerqe, NM, USA, 30 May 1 Jne Ashwill, T.; Laird, D. Concepts to Facilitate Very Large Blades. In Proceedings of the 45th AIAA Aerospace Sciences Meeting and Exhibit, Reno, NV, USA, 8 11 Janary Jensen, F.M.; Kling, A.; Sorensen, J.D. Change in Failre Type when Wind Trbien Baldes Scale-Up. Presented at the 5th Sandia Wind Trbine Blade Workshop, Albqerqe, NM, USA, 30 May 1 Jne Jorgensen, E.R.; Borm, K.K.; McGgan, M.; Thomsen, C.L.; Jensen, F.M.; Debel, C.P.; Sorensen, B.T. Fll Scale Testing of Wind Trbine Blade to Failre Flapwise Loading; Riso-R-1392 (EN); Riso National Laboratory: Roskilde, Denmark, Jensen, F.M.; Falzon, B.G.; Ankerson, J.; Stang, H. Strctral testing and nmerical simlation of a 34 m composite wind trbine blade. Compos. Strct. 2006, 76, Jensen, F.M.; Weaver, P.M.; Cecchini, L.S.; Stang, H.; Nielsen, R.F. The Brazier effect in wind trbine blades and its inflence on design. Wind Energy 2012, 15, Jensen, F.M.; Pri, A.S.; Dear, J.P.; Branner, K.; Morris, A. Investigating the impact of non-linear geometrical effects on wind trbine blades Part 1: Crrent stats of design and test methods and ftre challenges in design optimization. Wind Energy 2011, 14, Overgaard, L.C.T.; Lnd, E.; Thomsen, O.T. Strctral collapse of a wind trbine blade Part A: Static test and eqivalent single layered models. Compos. A 2010, 41, Overgaard, L.C.T.; Lnd, E. Strctral collapse of a wind trbine blade Part B: Progressive interlaminar failre models. Compos. A 2010, 41, Yang, J.S.; Peng, C.Y.; Xiao, J.Y.; Zeng, J.C.; Xing, S.L.; Jin, J.T.; Deng, H. Strctral investigation of composite wind trbine blade considering strctral collapse in fll-scale static tests. Compos. Strct. 2013, 97, Cho, J.S.; Chi, C.K.; Hang, I.K.; Chi, K.N. Failre analysis of wind trbine blade nder critical wind loads. Engin. Fail. Anal. 2013, 27, Facts, Analysis, Exposre to Indstrial Wind Energy s Real Impacts. Avalilable online: (accessed on 18 Febrary 2014). 13. Marin, J.C.; Barroso, A.; Paris, F.; Canas, J. Stdy of damage and repair of blades of a 300 kw wind trbine. Energy 2008, 33, Kong, C.; Bang, J.; Sgiyama, Y. Strctral investigation of composite wind trbine blade considering varios load cases and fatige life. Energy 2005, 30, Rajadrai, J.S.; Thanigaiyaras, G. Strctral analysis, failre prediction, and cost analysis of alternative material for composite trbine blades. Mech. Adv. Mater. Strct. 2009, 16, Wind Trbines-Part 1: Design Reqirements, 3rd ed.; IEC standard ; International Electrotechnical Commission (IEC): London, UK, 2005.