Hindawi Shock and Vibration Volume 2017, Article ID 3210271, 17 pages https://doi.org/10.1155/2017/3210271 Research Article Numerical Simulation of Fluctuating Wind Effects on an Offshore Deck Structure JinMa,1DaiZhou,1,2,3ZhaolongHan,1KaiZhang,1,4JenniferNguyen,5 JiabaoLu,1andYanBao1 1SchoolofNavalArchitecture,OceanandCivilEngineering,ShanghaiJiaoTongUniversity,Shanghai200240,China 2CollaborativeInnovationCenterforAdvancedShipandDeep-SeaExploration(CISSE),Shanghai200240,China 3StateKeyLaboratoryofOceanEngineering,ShanghaiJiaoTongUniversity,No.800,DongchuanRoad,Shanghai200240,China 4DepartmentofCivilEngineering,GraduateSchoolofUrbanInnovation,YokohamaNationalUniversity,Yokohama2408501,Japan 5CullenCollegeofEngineering,UniversityofHouston,Houston,TX77004,USA CorrespondenceshouldbeaddressedtoDaiZhou;[email protected] Received 5 November 2016; Revised 2 February 2017; Accepted 28 February 2017; Published 30 March 2017 AcademicEditor:MarcoBelloli Copyright©2017JinMaetal.ThisisanopenaccessarticledistributedundertheCreativeCommonsAttributionLicense,which permitsunrestricteduse,distribution,andreproductioninanymedium,providedtheoriginalworkisproperlycited. Theoffshorestructuresthatplayavitalroleinoilandgasextractionarealwaysundercomplicatedenvironmentalconditionssuch astherandomwindloads.Thestructuraldynamicresponseundertheharshwindisstillanimportantissueforthesafetyand reliabledesignofoffshorestructures.Thisstudyconductsaninvestigationtoanalyzethewind-inducedstructuralresponseofa typicaloffshoredeckstructure.Anaccurateandefficientmixturesimulationmethodisdevelopedtosimulatethefluctuatingwind speed,whichisthenintroducedastheboundaryconditionintonumericalwindtunneltests.Largeeddysimulation(LES)isutilized toobtainthetimeseriesofwindpressuresonthestructuralsurfacesandtodeterminetheworstworkingcondition.Finally,the wind-inducedstructuralresponsesarecalculatedbyANSYSParametricDesignLanguage(APDL).Thenumericallypredictedwind pressuresarefoundtobeconsistentwiththeexistingexperimentaldata,demonstratingthefeasibilityoftheproposedmethods.The wind-induceddisplacementshavethecertainperiodicityandchangesteadily.Thestressesatthetopofthederrickandconnections betweendeckandderrickarerelativelylarger.Thesemethodsaswellasthenumericalexamplesareexpectedtoprovidereferences forthewind-resistantdesignoftheoffshorestructures. 1.Introduction tocarryoutthewindfieldanalysesofoffshorestructuresto guaranteeareasonablesafetydesign. Numerous offshore structures have been constructed to In general, the wind-induced structural responses are supply oil and gas all over the world in recent years. The numerically investigated by applying wind pressures on the offshore structures are designed for resisting random wind structural surfaces by the time-domain analysis method andwaveloads.Inthecurrentdesignexperience,thelateral [5, 6]. Meanwhile, the numerical simulation in the present randomwindloadinthedesignoffixedoffshorestructures study could override the two-way fluid-structure interac- isoforderof10%inthetotallateraldesignloadsand25%in tion with the wind-induced structural response with small thecaseoffloatingplatforms[1,2].Manyaccidentsinvolving deformation [7]. Wind pressures on structures could be deck components such as a derrick and crane due to the extracted by the full-scale measurement, wind tunnel test, random wind loads have already been reported in damage and computational fluid dynamics (CFD) methods. Each investigation [3]. However, practical estimation and assess- of these methods is complementary with another in view ment of the fluctuating wind and wind-induced dynamic of their own advantages and limitations [8]. The full-scale response of the complex offshore deck structures are still measurementcanrecordthefirsthanddataofwindpressures involved problems [4]. Therefore, with the development of on prototype structures. However, it is generally conducted theoilexploitationtodeepsea,itisnecessaryandimportant with the limitations of several objective conditions such as 2 ShockandVibration weatherconditions,highcost,andsiteconditions[9].Wind Table1:Thesizeparametersofthesemisubmersibleplatform. tunnel test is a relatively mature and effective technique in wind engineering on scaled models of complex structures Members Dimensions(length×width×height)/m [10, 11]. However, there are also several limitations in wind Thelowerpartofderrick 17×17×42 tunneltestsforaccurateinvestigationsofwindpressureson Theupperpartof 17×17×22 structuresasfollows:(1)Reynoldsnumbersofscaledmodel derrick(wedge) tests are smaller than those of prototype structures; (2) the Deck 74.42×74.42×8.6 wind field conditions can hardly be reproduced exactly in Column 17.39×17.39×21.46 windtunneltests[10]. Owingtotherapiddevelopmentsofcomputerpowerand computationalmethods,CFDpredictionshavebeenwidely columns, and floating bodies. The main parameters of this usedtoanalyzethewindflowsaroundprototypestructures platformarelistedinTable1. inengineeringpractice[12].Largeeddysimulation(LES),as The tested model in Figure 1(b) was scaled as 1:100 in one of the main CFD methods, has been widely applied in [18]. The full-scale CFD numerical model of this platform thesimulationsofwindfields.WithpowerfulCFDtoolssuch in the present CFD simulations is built based on the tested as LES, it is possible to conduct full-scale size simulations scaled model [18] and numerical full-scale model [19]. In ofprototypestructuresinhighReynoldsnumbersflowsand view of the major consideration of wind field analyses, the provide abundant detailed data and information, such as offshoreplatformstructuresunderthewaterarehiddenand pressuredistributionsonstructuralsurfaces[13]. thederrickistotallyenclosedasshowninFigure1(c)whichis It is understood that proper generation of the inlet demonstratedthattheextractedwindpressuresareapplicable boundaryconditionsforCFDsimulations,especiallyforLES in the wind-induced dynamic response analyses [19, 20]. in computational wind engineering (CWE), is essential in However,thestructuralcomputationalmodelofthederrick the analyses of wind pressure distributions and the dyna- in the following time-domain analysis is similar to a real mic responses of prototype structures [14]. Therefore, the structureandnotenclosed,asisplottedinFigure1(d). improvement of simulating fluctuating wind speed in the Onethingtonoteisthattwotypesofnumericalmodels, inletboundaryisprovidedinthepresentstudy.Combining theCFDnumericalmodelandthestructuralcomputational withtheadvantagesofweightedamplitudewavesuperposi- model,areappliedinthispresentstudy.TheCFDnumerical tion(WAWS)[15]andautoregression(AR)[16,17]methods modelisaplatformmodelusedforLESsimulations,whichis withhighaccuracyandlesstimecost,respectively,amixture establishedbyusingacombinationofRhino(RobertMcNeel simulation(MS)methodisproposedanddemonstratedasan & Assoc, USA) and ICEM CFD (ANSYS, Inc., USA). The efficientandaccuratesimulationmethodinthepresentwork. structuralcomputationalmodelisafiniteelementmodelof In summary, the main objective of this present study is this platform for structural dynamic response analyses in toconductwindfieldanalysesoftheoffshoredeckstructures time domain, which is built by ANSYS Parametric Design whicharecarriedoutbyajointapplicationofCFDtechnique Language(APDL)(ANSYS,Inc.,USA).Thefiniteelementsof and structural finite element analysis. Furthermore, time thelocalcomponentsarelistedasfollows:thederrickadopts seriesofwindpressuresonthestructuralsurfacesatdifferent beamelement,thedeckusesshellelement,andthecolumns wind directions and inclined angles of the platform are adoptsolidelement. extracted in CFD wind tunnel tests and then applied to the finite element model to investigate the wind-induced 3.NumericalSimulation structuralresponse. Thispaperisorganizedasfollows:thestructuralmodels Inthissection,togetherwiththecommercialCFDsoftware innumericalsimulationandwindtunneltestarepresented Fluent 14.5 (ANSYS, Inc., USA), the unsteady-state analysis in Section 2. Methods for conducting CFD predictions are of the LES approach is used to predict the time series providedinSection3.AnaccurateandefficientMSmethod of wind pressures on the surfaces of structures above the forsimulatingfluctuatingwindspeedasboundaryconditions water surface at different working conditions. Fluent 14.5 isproposedinSection4.Computationalproceduresforthe conducts LES simulations based on finite volume method wholewindfieldanalysesoftheoffshoredeckstructuresare (FVM) [23]. Then, the wind-induced structural responses includedinSection5.ThevalidationtestsfortheMSandLES are explored by the extracted wind pressures acting on the methodsandnumericalcasesofwindfieldinvestigationsare structures using APDL and time-domain analysis method carriedoutinSection6. cooperatively [24]. The detailed computational algorithms and treatments of wind field analyses are addressed in the 2.TheModelSemisubmersiblePlatform followingsubsections. The semisubmersible platform in this study is the 6th- 3.1.CFDPredictions generationdeepwaterdrillingplatform,showninFigure1(a), andpossessesthefunctionofintelligentdrillingtechnology 3.1.1. LES Turbulence Model. In the present LES study, LES attaining the advanced level all over the world. The major separatesflowmotionsintolargeandsmalleddieswhichis structuralcomponentsofthisplatformincludederrick,deck, calculated by using a combination of the direct simulation ShockandVibration 3 (a) (b) (c) (d) Figure1:(a)Thepracticalsemisubmersibleplatform.(b)Thetestedmodelofwindtunneltests[17].(c)ThenumericalmodelusedinCFD windtunnelpredictions.(d)Thederrickmodelforwind-inducedanalyses. method and general model theory [25]. In view of these where V𝑖 is the large-scale wind velocity field tensor, 𝑃 the studiesintheLESsimulationswhoseproposedmethodsare resolvedpressure,𝑆𝑖𝑗theresolvedstrain-ratetensor,𝜌 theair appliedforfull-scaleCFDnumericalmodelunderReynolds density,𝜐thekinematicviscosity,𝑘 thesubgridscale(SGS) sgs number greater than 108 [10], these methods are adopted kinematicenergy,𝜐𝑠theSGSeddyviscosity,and to carry out the present LES simulations under Reynolds numberofapproximately6×107.Itisworthmentioningthat 𝜐 =𝐶 Δ √𝑘 , thesemethods,suchasone-equationmodel,aresuitablefor 𝑠 V V sgs relativelycoarsegridcaseandhighReynoldsnumberflows Δ [26].Theaccuracyandreliabilityoftheproposednumerical Δ = , framework have been validated by the consensus between V 1+𝐶 (Δ2𝑆2/𝑘 ) 𝑘 sgs numerical and experimental results in the previous studies [10,26]. 𝜕𝑘 𝜕V 𝑘 𝑘3/2 The governing equations of an incompressible flow for sgs + 𝑗 sgs =−𝜏 𝑆 −𝐶 sgs LESsimulations[26]areexpressedas 𝜕𝑡 𝜕𝑥 𝑖𝑗 𝑖𝑗 𝜀 Δ (2) 𝑗 𝜕 𝜕𝑘 + [(𝐶 Δ √𝑘 +𝜐) sgs] 𝜕𝜌V𝑖 + 𝜕𝜌V𝑖V𝑗 =− 𝜕 (𝑃+ 2𝜌𝑘 ) 𝜕𝑥𝑗 𝑑 V sgs 𝜕𝑥𝑗 𝜕𝑡 𝜕𝑥 𝜕𝑥 3 sgs 𝑗 𝑖 𝜕𝑘 𝜕𝑘 −2𝜐 sgs sgs, 𝜕 𝜕𝑥 𝜕𝑥 + [2𝜌(𝜐 +𝜐)𝑆 ], (1) 𝑖 𝑗 𝜕𝑥 𝑠 𝑖𝑗 𝑖 where𝐶V,𝐶𝑘,𝐶𝜀,and𝐶𝑑 aretheconstant[27,28]andΔis 𝜕V 𝑖 =0, thelengthscaleoffilterandcanbedeterminedaccordingto 𝜕𝑥 HuangandLi’swork[26]. 𝑖 4 ShockandVibration (a) (b) Figure2:Themeshingresults:(a)Block1;(b)Block2. TheimprovedLESapproachproposedbyHuangandLi thecomputerperformance,computingtime,andengineering [26]hasbeendevelopedasaFluentUserDefinedFunction requirements,theminimum,maximum,andmeanvaluesof (UDF) library, which has been applied in conducting LES 𝑦+are2.53308,4.89245,and3.48798,respectively.According simulations of tall buildings and long-span complex roof tothepreviousstudy[29],althougharecommendedvaluefor structuresinthepreviousstudies[10,12]. 𝑦+isabout1,valueslowerthan5areacceptablebecausethe cellstaysintheviscoussublayer. The total mesh number is nearly 11 million. According 3.1.2. Numerical Algorithms. In the LES predictions, all 9/4 to LES theory, the mesh quantity should reach Re in the discrete equations are solved by the pressure implicit order to provide very accurate numerical results. And the with splitting of operators (PISO) method. The time and more the mesh elements in the fluid domain, the higher spatial discretization are both treated by second-order dis- the computational accuracy that will be reached. However, crete schemes. Bounded central differencing is adopted for it requires much more computing time cost and higher momentum discretization. The LES simulations are per- computerperformancewhichwouldbeaheavytaskforCFD formed together with ANSYS/Fluent 14.5 and the parallel simulation. In order to make it feasible to reach moderate calculationsof16centralprocessingunits.Thetotaltimesteps accuracy while simultaneously satisfying the engineering arenearly41000with0.05spertimestep,whicharesufficient requirement (which can allow 10∼30% error for structure toreachthestatisticalconvergenceforLESsimulations[10, safety design), the grid resolution has to be reduced. For 12]. obtaining relative accuracies of the wind pressures, forces, andcomparingwindcoefficients,weconductedagridreso- 3.1.3. Computational Domain and Mesh Arrangement. The lutiontestinthefollowingvalidationtestandfoundthatthe numerical model of this offshore platform is established in presentedmeshbalancestimecostandnumericalaccuracy, full-scalesizes,assketchedinFigure1(c).Thedimensionsof satisfyingtheengineeringrequirements. thecomputationaldomain,sketchedinFigure2,are2600m (𝐿𝑥)×600m(𝐿𝑦)×200m(𝐿𝑧)asthecomputationaldomain 3.1.4. Boundary Conditions. The boundary conditions are is35(𝐿𝑥)×8(𝐿𝑦)×2.5(𝐿𝑧)[11].TheblockingrateofCFD described as the wind velocity profile in accordance with numerical model in the computational domain is less than an exponential rule, which can satisfy the Navier-Stokes 3%,whichisgenerallyacceptableinCWE[9].Accordingto equationsinLESsimulations[10,18]: themultiblocktechniqueinthepreviousstudy[12],theflow 𝐻 𝛼 field consists of two blocks: The small cylindrical flow field 𝑉𝑧 =𝑉0(𝐻 ) , (3) 0 andCFDnumericalmodel,whicharecombinedtogetherby aBooleanoperation,aredefinedas“Block1”inFigure2(a), where𝑉𝑧and𝑉0aretheaveragewindvelocitiesattheheight and the remaining part (i.e., the big rectangular flow field) of𝐻and𝐻0and𝛼isrelatedtogeomorphologicfeatures[22]. Inunsteady-stateanalysisoftheLESsimulations,theFlu- is defined as “Block 2” in Figure 2(b). One thing is to note entcodeprovidesadirectandconvenientmethodtogenerate thattheCFDnumericalmodelisnestedin“Block1.”Rotating a time-dependent turbulent fluctuating wind velocity field. “Block1”thespecifiedangleimmediatelycancarryoutLES However,itisreportedthatthegeneratedwindfieldbythe simulations at the specified wind direction, by which LES Fluentcodedecaysrapidlyintheinertialrangeandisnotwell simulations can be conducted at multiple wind directions approximated as the marine random wind [26]. The power conveniently. Numerous tetrahedral grids are generated in spectraldensity(PSD)ofthefluctuatingwindisfittedbyNPD theneighborhoodofthestructuralsurfaces,especiallyinthe (NorwegianPetroleumDirectorate)spectruminthepresent vicinity of the derrick, and the grid stretching ratio is kept study[22]: tobe<1.05.Themaximumandminimumgridlengthsinthe neighborhoodofthestructuralsurfacesare0.8mand0.2m, 320(𝑉/10)2(𝑧/10)0.45 𝑆(𝑓)= 0 , respectively. Hexahedron grids are adopted in the regions (1+1.5𝐹)3.56 farawayfromthecalculatingmodelandthegridstretching (4) ratioisrelativelyenlargedforreducingthetotalmeshnumber 𝑧 2/3 𝑉 −3/4 𝐹=172𝑓( ) ( 0) , for the sake of saving time cost in computation. In view of 10 10 ShockandVibration 5 Wind Inlet Outlet Figure3:Thesketchoftheinletboundaryofthefluctuatingwind. where𝑆(𝑓)isthePSDfunction,𝑓thewindfrequency,𝑉0the pressure, 𝜌 is the air density, and 𝑉0 is the average of wind averagewindspeedattheheightof𝐻0,and𝑧 theheightabove speedattheheightof10.0m. sealevel. AccordingtoChinaClassificationSocietycommonrules Given a clear explanation of the wind characteristics, and[19],theequationoftheshapecoefficientofastructural the fluctuating wind speed of the inlet boundary shown in memberisexpressedas Figure 3 is simulated by the MS method proposed in the 160 0.2 followingsection.Thisnewsimulationmethod,MSmethod, 𝑠𝑐 =𝐶 ( ) , can generate the time series of wind velocities of a num- 𝑖 𝑝 𝑧 (7) ber of spatial nodes accurately and efficiently. The pressure ∑ 𝑠𝑐𝑆𝐴 fluctuationonstructuralsurfacescanbesolvedbyimporting 𝑆𝐶= 𝑖 𝑖 𝑖, 𝑆𝐴 the inlet boundary conditions in LES simulations in the commercialsoftwareANSYS/Fluent14.5.Therefore,velocity where𝑖isameasuringpointonaspecifiedstructuralmember, andpressurecomponentstogethersatisfytheNavier-Stokes 𝑠𝑐𝑖istheshapecoefficient𝑖,𝐶𝑝themeanpressurecoefficient, equations. 𝑧 the height of a measuring point 𝑖, 𝑆𝐴𝑖 the area of the In order to guarantee the convergence property and structuralsurfacewhichameasuringpoint𝑖belongsto,and efficiencyinLESsimulations,thefluctuatingwindvelocities 𝑆𝐴 the total area of multiple structural surfaces. The ave- ateachtimestepattheinletaremodifiedthroughtheequal rage pressure coefficients, 𝑆𝐶, are changing with the wind totalflowoftheinletandoutletonthebasisoftheprevious directions. Therefore, the shape coefficients vary with wind studies[20,30].Theequationisdescribedas directions. (∑𝑛 V𝐴 −Ω) V =V − 𝑖=1 𝑖 𝑖 , (5) 𝑖+1 𝑖 𝐴 3.2. Wind-Induced Structural Response Analyses. The time series of the extracted wind pressures are applied to the whereV𝑖 isthefluctuatingwindspeedattimestep𝑖,𝐴𝑖 the offshorestructuresandthewind-induceddynamicresponses areaofthecorrespondinggridcell,𝐴thetotalareaoftheinlet arecalculatedbytime-domainanalysismethodtogetherwith boundary,andΩthetotalflowoftheoutletboundaryatthe APDLsolvingthestructuraldynamicequationasfollows: correspondingtimestep. According to the work of Lu et al. [20] and Shirani et [𝑀]{𝑢̈}+[𝐶]{𝑢̇}+[𝐾]{𝑢}=[𝐹], (8) al.[30],thewindvelocitiesateachtimestepbeingreadand updatedinFluent14.5arefoundtobeslightlysmallerthanthe where [𝑀], [𝐶], and [𝐾] are, respectively, structural mass, initialwindvelocitiesgeneratedbytheself-madeMATLAB damping,andstiffnessmatrices,{𝑢̈},{𝑢̇},and{𝑢}are,respec- program. tively, acceleration, velocity, and displacement vectors, and [𝐹] are fluctuating wind loads. Meanwhile, (4) is solved by theNewmark-𝛽approachinthispresentstudy[22]. 3.1.5.DataAnalyses. Theanalysesofwindpressuredistribu- In structural dynamic response analysis, the damping tionarecarriedoutintermsofnondimensionalcoefficients, 𝐶𝑝and𝐶𝑝rms,ofameasuringpoint𝑖 defined,respectively,as matrix[𝐶]isdeterminedbythefollowingequations: [𝐶]=𝛼[𝑀]+𝛽[𝐾], 𝑃 𝐶 = 𝑖 , 𝑝 0.5𝜌𝑉2 𝛼 𝛽𝜔 (9) 0 𝜉 = + 𝑖, (6) 𝑖 2𝜔 2 𝑃 𝑖 𝐶 = 𝑖,rms , 𝑝rms 0.5𝜌𝑉02 where𝛼and𝛽aretheconstantsinRayleighdampingwhich arecalculatedbythedampingratioineachmodeshape𝜉𝑖[20] where𝐶𝑝 and𝐶𝑝rms arethemeanandRMSpressurecoeffi- whichisassumedasaconstantoverarangeoffrequencies. cients,respectively,𝑃𝑖isthemeanpressure,𝑃𝑖,rmsistheRMS And𝜔𝑖isthestructuralfrequencyineachmodeshape. 6 ShockandVibration PSD:S(휔)=S1(휔)W1(휔)+S2(휔)W2(휔) Wavelet method Parts of energy concentrationS1(휔) Parts of energy dispersionS2(휔) WAWS method AR method Generate wind speed(cid:2)1(t) Generate wind speed(cid:2)2(t) Time series of fluctuating wind:(cid:2)(t)=(cid:2)1(t)+(cid:2)2(t) Figure4:TheflowchartforsimulatingthefluctuatingusingMSmethod. 4.TheMethodfor Simulating TheapplicationoftheWAWSandARmethodsisbriefly FluctuatingWind presented in the following formula derivation, which has fullyconsideredtemporalandspatialcorrelationintheself- Amixturesimulation(MS)methodisproposedtogenerate compiledprograms. the fluctuating wind speed as the boundary conditions in CFD predictions, combining with the weighted amplitude (1)TheWAWSMethodintheMSMethod.UsingtheWAWS wavesuperposition(WAWS)method[15]andautoregression method,thecorrelationfunction𝑅𝑗𝑚isexpressedas (AR) methods [16, 17]. The new approach overcomes the defects of the abovementioned traditional methods and 𝑅 (𝜏)=𝐸(V (𝑡)V (𝑡+𝜏)) inheritstheirownadvantageswhichguaranteethecomput- 𝑗𝑚 𝑗 𝑚 ingaccuracyandefficiencysimultaneously. stocAhasssutimcGinagussthiaantptrhoecesflsu,cthtueasttioncghawstiincdchaisracatesrtiasttiicosncaarny =Δ𝑓∑𝑗 ∑𝑁 𝐻𝑗𝑘(𝑓𝑘𝑙)𝐻𝑚𝑘(𝑓𝑘𝑙) (11) 𝑘=1𝑙=1 be described by the power spectral density (PSD) function [31]. Then, the energy distributions of the PSD function of ⋅cos(2𝜋𝑓𝑘𝑙+𝜃𝑗𝑘(𝑓𝑘𝑙)−𝜃𝑚𝑘(𝑓𝑘𝑙)), thefluctuatingwindareexploredbythewaveletmethod[32] adopting the Hanning windows [33]. Therefore, the main thoughts of the MS method are listed as follows: the parts where𝐸isexpectation,Visthetimeseriesofwindspeed,𝐻 of the energy concentration are simulated by the WAWS is obtained by Cholesky decomposition, and 𝜃𝑗𝑘(𝑓𝑘𝑙) is the method [15] and the parts of the energy dispersion are phaseanglebetweenspatialnodes𝑖and𝑗. generatedbytheARmethod[16].Thefinalsimulatingresults Assumingthat𝑁→∞andΔ𝑓→0, ofthefluctuatingwindareobtainedbythesuperpositionof tshimeusliamtiuolnatipnrgocreesssulitsscboynctlhuedetwdoinmFeitghuordes.4.ThTheescehqeumataiotinc 𝑅𝑗𝑚(𝜏)=∫𝑠ℎ1 ∑𝑗 𝐻𝑗𝑘(𝑓𝑘)𝐻𝑚𝑘(𝑓𝑘) ofthecalculatingprocessbytheMSmethodisdescribedas 𝑠𝑙1 𝑘=1 (12) follows: ⋅cos(2𝜋𝑓𝑘+𝜃𝑗𝑘(𝑓𝑘)−𝜃𝑚𝑘(𝑓𝑘))𝑑𝑓. 𝑆(𝑓)=𝑆1(𝑓)𝑊1(𝑓)+𝑆2(𝑓)𝑊2(𝑓), (10) Thecorrelationfunctionistransformedfromtimedom- where 𝑆(𝑓) is the PSD function, 𝑓 is the wind frequency, aintofrequencydomainonthebasisofconvolutiontheorem: 𝑊1(𝑓)and𝑊2(𝑓)are,respectively,theHanningwindowsof athree,erneesrpgeycctiovneclye,nstirmatuiolanteadndbdyistpheersWioAnW,aSndm𝑆e1t(h𝑓o)danadnd𝑆2(A𝑓R) 𝑅𝑗𝑚(𝜏)⊗𝑊1(𝑡)=∫𝑠ℎ1 ∑𝑗 𝐻𝑗𝑘(𝑓𝑘)𝐻𝑚𝑘(𝑓𝑘) method. 𝑠𝑙1 𝑘=1 The Hanning window number is determined by the energy distribution of the PSD function in the wind field. ⋅𝑊̂1(𝑓)cos(2𝜋𝑓𝑘+𝜃𝑗𝑘(𝑓𝑘)−𝜃𝑚𝑘(𝑓𝑘))𝑑𝑓 (13) Accordingtoourexperience,thewindownumberusingthe WmeAtWhoSd.mIneththoidswisayt,htrheeeMtimSemselathrgoedrctohualndirteduusicnegththeetiAmRe =∫𝑠ℎ1 ∑𝑗 𝑆𝑗𝑘(𝑓)⋅𝑊̂1(𝑓)cos(2𝜋𝑓𝑘)𝑑𝑓, costincomputationwithnoaccuracylosttosimulateakind 𝑠𝑙1 𝑘=1 ofloadswiththerandomproperties,whichisdemonstrated inthefollowingvalidationtestsinSection6.1. where⊗representsconvolutionoperation. ShockandVibration 7 Time series of wind speed simulated by the WAWS Step3. ConstructtheCFDnumericalwindtunnel. methodV1(𝑡)isobtainedby Step3.1.CreatethecalculationmodelandmeshfilesbyRhino andICEMCFD. V (𝑡) 1 Step 3.2. Set the initial and boundary conditions in =√2Δ𝑓∑𝑗 ∑𝑁 𝐻𝑗𝑘(𝑓𝑙)cos(2𝜋𝑓𝑙𝑡+𝜃𝑗𝑘(𝑓𝑙)+𝜑𝑘𝑙), (14) ANSYS/Fluentandstartthecalculation. Step3.3.Calculatetheresultantforcestodeterminetheworst 𝑘=1𝑙=1 operatingcondition. 𝑗=1,...,𝑚, Step 3.4. Extract the time series of wind pressure on the structural surfaces by the UDF program for the following where𝜑isarandomnumberfrom0to2𝜋. dynamicresponseanalyses. Step4. Carryoutthewind-induceddynamicresponseanal- (2)TheARMethodintheMSMethod.Similarly,thecorrela- ysesbyAPDL. tionfunctionintheARmethodcanbeexpressedas Step4.1.Importthefiniteelementmodeloftheplatformand 𝑅𝑖𝑗(𝜏)=∫𝑠ℎ2 𝑆𝑖𝑗(𝑓)𝑊̂2(𝑓)cos(2𝜋𝑓⋅𝜏)𝑑𝑓, (15) deckstructures. 𝑠 𝑙2 Step4.2.Settheinitialandboundaryconditionsandstartthe calculationbytheNewmark-𝛽methodintimedomainusing where𝑅isthecorrelationfunction. thesecondarydevelopmentapplicationprograms. TimeseriesofwindspeedsimulatedbytheARmethod Step4.3.Extractthecalculationresultssuchasthedeforma- V2(𝑡)iscalculatedby tionsandstressesofnodesandmembers. 𝑗 𝑝 6.ResultsandDiscussion V (𝑡)=−∑∑𝑎 V (𝑡−𝑘⋅Δ𝑡)+𝑢(𝑛), (16) 2 𝑖𝑗𝑘 𝑖 𝑖=1𝑘=1 6.1.ValidationTests. Inthissubsection,thenumericalresults are compared to verify that the methods and self-made whereΔ𝑡isthetotalsimulatingtimestep,𝑎thecoefficientof programsmentionedinSections3and4areapplicableand autoregressionmatrix,𝑢(𝑛)theexcitedwhitenoisethrough effectiveinthefollowingnumericalexamples. all-zerosfilter,and𝑝theorderofARmodel. Overall,thetotalwindspeedisobtainedbyV(𝑡)=V1(𝑡)+ 6.1.1.ValidationfortheMSMethod. Testsareconstructedto V2(𝑡),whichisshowninFigure4. demonstrate the accuracy and efficiency of the MS method comparedtothetraditionalsimulationmethods:theWAWS 5.TheProcedurefor theWindField and AR methods. The time series of the wind speed at any spatialpointissimulatedbytheMS,WAWS,andARmeth- AnalysesofthePlatform ods, respectively. The comparison between the simulated and theoretical results could determine the best simulation In this section, the details for conducting LES predictions method. and wind-induced dynamic responses analyses of the deck structures under the action of the fluctuating wind are (1) Validation for the Accuracy of the MS Method. The summarizedasfollows. fluctuatingwindspeedatonesinglespatialpointissimulated bytheMS,WAWS,andARmethods,respectively,inthesame Step1. Establishthegeometricandfiniteelementmodelsby initial condition listed as follows: the average wind velocity VC++andAPDL. 𝑉0 = 31.654m/s, the number of the sampling points 𝑛 = Step2. Generatethefluctuatingwindspeed. 1200, time stepΔ𝑡 = 0.1 s, and the measuring point (in meters)islocatedat(0,0,5).Theerrorbetweentheenergy Step2.1.Simulateandcorrectthetimeseriesofthewindspeed [34] of the simulated and theoretical results solved by (17) by using the MATLAB (MathWorks, USA) self-compiled can assess the accuracy of the simulation method by fast programs. Fouriertransform(FFT)[35]andfrequency-timeconversion methods[36,37]: Step 2.2. Convert these data into data files in the format tbhyaAt NhaSsYSa/.FtxluteenxttuensisniognthfeorUmsaetr.DThefienseeddaFtuanfictlieosnasr(eUrDeaFd) Ω= 𝑋(𝑗𝜔)2 = ∫𝑥(𝑡)𝑒−𝑖𝜔𝑡2, (17) program. 𝑇 𝑇 8 ShockandVibration 104 104 102 102 D D S S P 100 P 100 10−2 10−2 10−3 10−2 10−1 100 101 10−3 10−2 10−1 100 101 Frequency (Hz) Frequency (Hz) Simulated results Simulated results Target results Target results (a) (b) 104 102 100 10−2 10−3 10−2 10−1 100 101 Simulated results Target results (c) Figure5:ThecomparisonbetweensimulatedandtargetresultsbyMS,AR,andWAWSmethods:(a)MSmethod;(b)ARmethod;(c)WAWS method. Table2:Thecomparisonofcalculatingaccuracybetweentheresults Table3:Thecomparisonofthecomputingtimeandaccuracyat36 ofsimulatedPSDandtheoreticalPSD. and108spatialpointsusingtheMS,WAWS,andARmethods. MSmethod WAWSmethod ARmethod Numberofnodes Method Error Computingtime(s) Theoreticalenergy 12.6923 12.6923 12.6923 WAWSinMS 2.34% 256 Simulatingenergy 12.3927 13.6073 14.2073 ARinMS 6.24% 75 Error 2.36% 7.21% 11.94% 36 MS 4.79% 331 WAWS 4.24% 27465 AR 22.96% 75 where𝑥(𝑡)istimeseriesofwindvelocity,𝑋(𝑗𝜔)theampli- tudespectrumobtainedfrom𝑥(𝑡)byFFT,and𝑇thelasting WAWSinMS 2.07% 6766 timeoftimeseriesofwindspeed.TheenergyΩiscalculated ARinMS 9.40% 489 bythesquareoftheamplitudespectralmodedividedbythe 108 MS 4.85% 7255 lastingtime𝑇. WAWS 4.24% 647836 Thecomparisonsbetweensimulatedandtargetresultsby AR 32.85% 489 the MS, AR, and WAWS methods are plotted in Figure 5. The blue and red curves are target and simulated PSD, respectively.AsseenfromFigure5,thesimulatedresultsby the MS method are more accurate and closer to the target wind of numerous spatial points rather than one single results. The numerical accuracy by the AR method in the point.Timecostincomputationbecomesanimportantcon- lower and higher regions is the lowest compared to those sideration in the numerical examples. The comparisons of by the MS and WAWS methods. Table 2 conducts further thesimulatingtimeusingtheMS,WAWS,andARmethods, numerical accuracy analyses by the energy comparisons respectively, are carried out under the same computing betweenthesimulatedandtheoreticalresultsusingtheMS, conditionsmentionedinValidationfortheAccuracyofthe WAWS, and AR methods. Obviously, the error of the MS methodismuchlowerthantheothersunderthesameinitial MSMethod. conditionsas2.36%versus7.21%and11.94%. The fluctuating wind velocities of 36 and 108 adjacent These comparisons demonstrate that the MS method is grid nodes in the inlet boundary sketched in Figure 3 are amoreaccuratesimulationmethodtogeneratetherandom simulated by three abovementioned simulation methods. loadscomparedtotheWAWSandARmethods. Table 3 gives a series of comparisons of computing time and accuracy of generating the fluctuating wind velocities (2)ValidationfortheEfficiencyoftheMSMethod.Ingeneral,a at36and108adjacentspatialpointsusingthreesimulation numericalwindtunnelneedsthetimeseriesofthefluctuating methods. ShockandVibration 9 0.0 −0.5 nt e ci ffi e −1.0 o c e ur ess −1.5 pr d n Wi −2.0 −2.5 −4 −3 −2 −1 0 1 2 3 4 X (m) Total grid number 7×105 2.8×106 1.4×106 4.2×106 (a) (b) 0.0 0.4 −0.5 nt nt fficie fficie 0.3 oe −1.0 oe ure c ure c d press −1.5 S press 0.2 n M Wi R −2.0 −2.5 0.1 −4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4 X (m) X (m) The experimental data The experimental data The simulated results The simulated results (c) (d) Figure6:(a)ThesketchofflowfieldandFB16model.(b)Meshindependenttests.(c)Thecomparisonofthewindpressurecoefficients between the simulated and experimental results [21] of the roof on the windward side. (d) The comparison of the RMS wind pressure coefficientsbetweenthesimulatedandexperimentalresults[21]oftheroofonthewindwardside. InTable3,itisobvioustoseethattheincreasingnumber 6.1.2.ValidationfortheMethodofCFDNumericalWindTun- of the spatial points could raise the computing difficulty nel. InordertovalidatetheabovementionedCFDmethods including the computing time and accuracy. Although the inSection3quickly,CFDpredictionsofFB16inFigure6(a) calculating time by the AR method is always the shortest, [21], whose dimensions are smaller than those of the off- the computing error increases rapidly from 11.94% of one shore platform, are conducted as an initial solution of LES single point to 32.85%. The simulating accuracy of the simulations.Thecomputationaltreatmentandalgorithmare WAWS method is acceptable all the time, but time cost in similartoSection3.ThedimensionsofFB16modelare12m× computationoftheWAWSmethodisnearly100timesmore 6.7m×5.3m.Thedimensionsofthecomputationaldomain than those of the MS method. Therefore, it is concluded aredefinedas420m×55m×15m[12]andtheblockingrate that the MS method is a more efficient simulation method of the calculating model is less than 3%. The flow field and comparedtotheWAWSandARmethods. FB16aresketchedinFigure6(a). Based on the two abovementioned validation tests, the Throughaseriesofmeshindependenttestswiththetotal MS method is an efficient and accurate simulation method grid number of 7.0 × 105, 1.4 × 106, 2.8× 106, and 4.2× 6 integratingtheadvantagesoftheWAWSandARmethods. 10 , the wind coefficients of the roof on the windward side 10 ShockandVibration (a) (b) (c) ∘ ∘ ∘ Figure7:Thesketchofthecalculatingmodelwiththreeinclinedangles:(a)0,(b)5,and(c)10. are displayed in Figure 6(b). Since the total grid number 6.2. Numerical Examples. In this section, attention is fixed increases from 7.0 × 105 to 4.2 × 106, the wind coefficients onthewindfieldanalysesoftheoffshorestructuresoverthe havelittlechangeinvalue.Forobtainingrelativeaccuraciesof water as plotted in Figure 1 under the action of the harsh thewindpressures,forces,andcomparingwindcoefficients, fluctuatingwind.CFDnumericalwindtunneltestsandwind- itisfoundthatthepresentmeshinthepresentgridresolution induced structural response analyses are presented in the test(i.e.,thetotalgridnumber7.0×105)canhaveabalance followingnumericalstudies. on time cost and numerical accuracy, which can satisfy the engineering requirements by and large. Therefore, the total 6.2.1. CFD Wind Tunnel Predictions. The present work in gridnumber7.0×105 isadoptedinconsiderationofsaving thissubsectionistoanalyzethewindpressuredistributions computingtimewhileprovidingaccurateresults. andwindresultantforcesofthisplatformwiththeinclined Figures6(c)and6(d)describethenumericalresultsand angles of 0∘, 5∘, and 10∘ plotted in Figure 7 at different experimentaldataofthewindpressurecoefficientsandRMS wind directions. As the symmetry of the platform, 7 wind ∘ ∘ pressurecoefficients.Meanwhile,theredlinerepresentsthe directions are defined within the range of 0 and 90 with ∘ present CFD results and the dark line is extracted as the intervalsof15 .Similartothemeshindependentsimulations experimental data from [21]. The comparison shows that in the validation tests, the total grid number in the present the trends of the curves are roughly in agreement and the LESsimulationsisapproximately1.1×107. maximumerrorsofwindpressurecoefficientandRMSwind Accordingto[10,20,30],CFDsimulationsadoptthewind pressurecoefficientbetweenthenumericalandexperimental speedupdatingmethodbythejointapplicationofMATLAB resultsare,respectively,lessthan15%and21%.Therefore,the and Fluent. The wind speeds of the grid nodes in the inlet LESmethodwiththesestandardsisavailabletocarryoutthe boundary are simulated by the MS method mentioned in followingnumericalexamples. Section 3 using self-compiled MATLAB codes. Using the
Description: