Jtam-A4.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 55, 1, pp. 353-368, Warsaw 2017 DOI: 10.15632/jtam-pl.55.1.353 HIGH ORDER SENSITIVITY ANALYSIS OF A MISTUNED BLISK INCLUDING INTENTIONAL MISTUNING Linus Pohle, Sebastian Tatzko, Lars Panning-von Scheidt, Jörg Wallaschek Leibniz Universität Hannover, Institute of Dynamics and Vibration Research, Hannover, Germany e-mail: linuspohle@gmx.de Small deviations between turbine blades exist due to manufacturing tolerances or material inhomogeneities. This effect is called mistuning and usually causes increased vibration am- plitudes and also a lower service life expectancy of bladed disks or so called blisks (bladed integrated disk). The major resulting problem is to estimate the maximum amplitude with respect to these deviations.Due to the probabilitydistribution of these deviations, statistical methods are used to predict the maximum amplitude. State of the art is the Monte-Carlo simulationwhich is based on a high number of randomly re-arranged input parameters.The aim of this paper is to introduce a usefulmethod to calculate the probability distribution of the maximum amplitude of amistuned blisk with respect to the random input parameters. First, the applied reduction method is presented to initiate the sensitivity analysis. This reduction method enables the calculation of the frequency response function (FRF) of a Finite Element Model (FEM) in a reasonable calculation time. Based on the Taylor series approximation, the sensitivity of the vibration amplitude depending on normally distribu- ted input parameters is calculated and therewith, it is possible to estimate the maximum amplitude. Calculating only a single frequency response function shows a good agreement with the results of over 1000Monte-Carlo simulations. Keywords: turbine blades, mistuning, sensitivity analysis Nomenclature d0 – structural damping σ – standard deviation k – blade index I – identity matrix u – modal displacement Kb – stiffness matrix of blades x – geometric displacement Kd – stiffness matrix of one disk segment EO – engine order KAb – stiffness matrix of blades, type A N – number of blades [̃·] – reduced system matrices δk – mistuning factors [̂·] – absolute amplitude α – tolerance interval [·]H – Hermitian transpose 1. Introduction A blade integrated disk (blisk), i.e. blades and disk made out of one piece, has a lot of tech- nical benefits like weight reduction or omission of fretting between the blade and disk. This causes lower damping and higher structural coupling between the blades. Due to small devia- tions between the blade properties, energy localization can occur causing considerably higher vibration amplitudes of a few blades.Whitehead (1966) gave a limit for themaximumvibration amplitude of amistuned blisk. This theoretical maximum is existent, if all modes have the same eigenfrequency and are in phase at one blade position. 354 L. Pohle et al. The dynamical behavior of one blade mode can be simulated with a cyclic lumped-mass model (see e.g. Griffin and Hoosac, 1984). The benefits of such a simple model is the simple dynamical behavior and the low calculation time. However, differentmode families andmultiple mode shapes cannot be described by this model. On the other hand, high computational cost is inevitable to solve a full Finite Elementmodel of amistuned blisk. Especially for parameter stu- dies, a high number of high-resolution FRF’s is necessary. Therefore, a lot of different reduction methods are developed to calculate the FRF of a mistuned blisk with a lower computational effort. There are two ways that are frequently used to reduce the number of DOF of a mistuned blisk. On the one hand, the model can be described by the system modes (called Subset of Nominal Modes, SNM), which is described in Yang and Griffin (2001), extended to geometric mistuning in Sinha (2009) and compared to the first version of the SNM in Bhartiya and Sin- ha (2011). The model of the whole blisk is divided into single sectors which could be reduced by a modal transformation using the modes of the segment. Afterwards, the mistuned stiff- ness matrix can be calculated very efficiently, and a good agreement to the full model can be shown. If there is only one mode describing the blade dominant vibration, this method can be extended to the Fundamental Model of Mistuning (FMM), which was introduced in Feiner and Griffin (2002). The main condition is that the mode shape remains almost unchanged and only small difference occurs between the blades eigenfrequencies. The main benefit is that the knowledge of the system matrices of the Finite Element model is not required. This method is very useful for the identification of mistunig, see Shuai and Jianyao (2010), Feiner and Griffin (2004a,b). On the other hand, one sector can be divided into a blade and a disk sector. After mo- dal reduction of these components, the matrices are coupled and the dynamical behavior of the whole mistuned blisk can be calculated. This so-called Component Mode Synthesis (see e.g. Bladh et al., 2001a,b) needs more computational effort to regard mistuning, but is much more flexible (see Moyroud et al. (2002), for comparison, Castanier and Pierre (2006) for an overview). It is assumed that the disk is cyclic symmetric. Thus it can be described by cyclic boundary conditions (see Thomas (1979) for a description). The DOFs of the bla- de are reduced by the well-known Craig Bampton method (see Craig and Bampton, 1968; Craig, 2000). This method was extended in Hohl et al. (2011) using a reduction of the co- upling nodes using Wave-Based Substructuring Čermelj et al. (2008) and a secondary modal reduction. To analyze the influence of mistuning, Monte-Carlo Simulation (MCS) can be used to find an optimal pattern or to find themaximum amplitude of a given probability distribution of the eigenfrequencies, see Hohl et al. (2011). In Bladh et al. (2001), it is shown that only 50 MCS are necessary to fit the probability density function (PDF) of themaximumvibration amplitude over all frequencies. ThePDF is fittedwith aWeibull distribution. InMignolet andHu (1997), a direct predictionmethod describes how to find themaximumamplitude for one frequency using the cumulates of the normal distributions. The distribution of a lumped-mass model with one DOF per segment results in a Gaussian distribution. Thepresentedpaper introduces anewmethod tofindtheGaussiandistributionof amistuned blisk. The aim of the paper is to show the benefits of thesemethods and its possible application. First, theused reductionmethod is describedand theuniversal validity is given.Thereafter, high order sensitivity analysis is shownand compared to theMCS.The second case study emphasizes its versatile application spectrum based on an intentional mistuned example. Therewith, the dynamicalbehavior of amistunedblisk canbeanalyzeddependingondifferentmistuning factors, engine order, or blade pattern. High order sensitivity analysis of a mistuned blisk... 355 2. Reduction method The first example analyzed is a simplified blisk with N = 12 blades (see picture in Fig. 1). This blisk exhibits the typical dynamical behavior of a blisk without spinning effects like spin softening, centrifugal stiffening or Coriolis effects. The blisk is cyclic symmetric with 12 blades ignoring manufacturing tolerances. Thus, cyclic symmetry can be used to analyze the full blisk regarding only one sector. Fig. 1. Picture of the analyzed blisk Fig. 2. Finite Elementmesh with special nodes of interest In Fig. 2, the mesh of such a sector is shown. The mesh of the whole system has about 170000 degrees of freedom (DOF) and a harmonic analysis of the vibration amplitudes cannot 356 L. Pohle et al. be solved at a commonPC in a reasonable time, especially if parameter variations are processed or optimization problems are considered. Regarding cyclic symmetry, the number of DOFs is reduced to 14841 for the calculation. The excitation node and the monitor node are located at the blade tip, and the direction of excitation and monitoring is the z-direction. Of course, the following shownmethod can be used to analyze a blisk with any arbitrary geometry. In Fig. 3, the nodal diameter diagram is illustrated for the first four mode families. The firstmode family is well separated, and will be used for the analysis in this paper. Nevertheless, the results are verified for other mode families as well. Some chosen mode shapes of the tuned system are depicted in Fig. 3. Fig. 3. Nodal diameter diagram In general, the equation of the dynamical system can be written as Mẍ(t)+Dẋ(t)+Kx(t)=F(t) (2.1) withM,D, andK as themass, damping, and stiffnessmatrix. The excitation force vectorF(t) is assumed to be harmonic in timewith a constant phase shift between the blades depending on the engine orderEO F(t)= F̂eiΩt =   1 ... e−ikφ ... e−i(N−1)φ   f̂eiΩt (2.2) with the phase shift φ= 2πEO N = const (2.3) The damping is modeled as structural damping depending on the stiffness matrix D= d0 Ω K (2.4) High order sensitivity analysis of a mistuned blisk... 357 Using the modal transformation into the frequency domain with the harmonic approach x(t)= x̂exp(iΩt), Eq. (2.1) reads x̂(Ω)= [−MΩ2+(1+id0)K]−1F̂ (2.5) The attentive reader notices that the equation of motion is build up for a linear model. The following process will be described for such a model. Nevertheless, it will be very simple to include rotating effects like centrifugal or Coriolis forces. Therefore, the matrices have to be checked for nonsymmetry and varity over the rotation speed. The following reduction method is limited to linear models, but other reduction methods for nonlinear models can be used. For a largenumberofDOFsa lotof computational effort is required to calculate thecombined matrix (−MΩ2+(1+id0)K)−1. The chosen reduction method is based on the ComponentMode Synthesis (CMS) extended by theWave-Based Substructuring (WBS) (seeHohl et al., 2011). Firstly, the structuralmatrices are divided intomaster nodes (m) and slave nodes (s) asdescribed inCraig andBampton (1968). Themaster nodes are the coupling nodes between the blades and the disk (see Fig. 2). The slave nodesare transformed toa small basis ofmodeshapes.Therefore, the reducedstiffnessmatrix K̃b of one blade b is obtained by K̃b = [ Kmm K̃ms K̃sm K̃ss ] = [ I 0 Ψ Φ ]H[ Kmm Kms Ksm Kss ][ I 0 Ψ Φ ] (2.6) where I is the identity matrix, (·)H denotes the Hermitian transformation and Ψ=−K−1ss Ksm (2.7) andΦ is themodalmatrixwhich contains the eigenvectors of the generalized eigenvalue problem KssΦ=MssΦλ (2.8) Accordingly, themassmatrix is subdivided and reduced. The disk can be reduced using a cyclic symmetric model as shown in Thomas (1979). Thus, all tuned sectors have the same dynamic properties. To couple the segments, it is assumed that the nodes on the right hand side uk,R have the same motion as the nodes on the left hand side uk,L, except for the phase shift (see Fig. 2). Hence, the displacement of the nodes can be rewritten in cyclic coordinates by uk =   uk,L uk,M uk,R  =   0 exp ( − i2πk N ) I 0 0 I   [ uk,M uk,R ] (2.9) for each nodal diameter k. Due to this transformation, themassmatrix and the stiffnessmatrix can be formulated for each segment. After the Craig Bampton reduction of all blades, see Eq. (2.6), the reduced matrices can be written in the matrix of the full disk by using a subset of modes and themaster nodes K̃d = [ K̃mm[diag(K̃d,ms)] [diag(K̃d,sm)] [diag(K̃d,ss)] ] (2.10) with [diag(K̃d,ss)] =   K̃d,ss,1 · · · 0 ... ... ... 0 · · · K̃d,ss,N   (2.11) 358 L. Pohle et al. and just for [diag(K̃d,sm)] and [diag(K̃d,ms)]. The coupling nodes have to be transformed into the full system using the theory of cyclic systems K̃mm =   K̃1,1 K̃1,2 · · · K̃2,1 K̃2,2 · · · ... ... ...   (2.12) with the submatrices K̃i,j = N−1∑ h=1 (wh(i−1))HK̃hw h(j−1) (2.13) where w=e−i 2π N (2.14) As a consequence, the size of the structuralmatrices of the disk is strongly decreased. Neverthe- less, the master nodes are in physical notation and can include a lot of DOFs. These coupling nodes are reduced by the single-value decomposition (SVD) of the modal matrix of the whole segment including the disk and the blade KsegΦseg =MsegΦsegλ (2.15) The SVDof a subset of themodalmatrixΦsub =Φseg(cpl, ·), including only the couplingDOF, is given by Φsub =UΣV T (2.16) whereΣ is a square diagonal matrix with non-negative real numbers on the diagonal. Using the matrix U to reduce the system with Ured =U(w, ·), the behavior of the coupled nodes can be projected in an accurate way. w is the number of waves chosen to describe the motion of the coupling nodes. The reduction method is called Wave-Based Substructuring (WBS) which is described in Čermelj et al. (2008). In this way, the system matrices for the blades and the disk can be written as [ K̃mm K̃ms K̃sm K̃ss ] = CMS︷ ︸︸ ︷[ I 0 Ψ Φ ]H WBS︷ ︸︸ ︷[ Ured 0 0 I ]H[ Kmm Kms Ksm Kss ][ U∗ 0 0 I ][ I 0 Ψ Φ ] (2.17) Thus, the physical DOF are described by [ um us ] = [ U∗ 0 0 I ][ I 0 Ψ Φ ][ xm xs ] (2.18) with a significantly smaller number ofDOFs than theFinite Elementmodel. After the assembly of the blades and the disks, the system matrices are reduced by the second modal reduction. This reduction is a goodpossibility to decouple the equation ofmotion, too.Therewith, equation of motion (2.5) can be written as a scalar function ui = ˜̂ F i −M̃i,iΩ2+(1+id0)K̃i,i (2.19) One possibility to take into account the mistuning is to detune the single blades by a factor of Kb,k = δkKb,0 ⇒ K̃b,k = δiK̃b,0 (2.20) High order sensitivity analysis of a mistuned blisk... 359 This can be a variation of Young’s modulus. These factors are randomly built up by a standard deviation σ0 and a mean value of 1. Afterwards, the full system can be reassembled using the subsystems K̃sys =   [diag(K̃b,ss)] [diag(K̃b,ms] 0 [diag(K̃b,sm)] K̃mm K̃d,ms 0 K̃d,sm K̃d,ss   (2.21) where [diag(K̃b,ss)]=   δ1K̃b,ss · · · 0 ... ... ... 0 · · · δNK̃b,ss   (2.22) as the reduced diagonal matrix of all slave nodes of the mistuned blades. K̃mm is the diagonal matrix of the reduced coupling nodes and K̃d,ss is the reduced matrix of the blisk. The mass matrix is assembled in the same way. With the second modal reduction, the system is reduced to a very small number of DOFs and the equations are decoupled. In Figs. 4a and 4b, a comparison between the reduced order model (ROM) and the FEM is given. The ROM is reduced from overall 170000 DOF to 84 DOF for the whole system considering mistuning. To prove the capability of the reduced order model as an example a strongly mistuned blisk is used. Young’s modulus of one blade is 20% higher than all other. To calculate 131 frequency points, the time needed for the FEM is about 5 minutes on a standard personal computerwith a comercial FiniteElement code.The calculated points are concentrated around the eigenfrequency to guaranty the absolute amplitude. The calculation of the FRF of the ROM (Eq. (2.19)) with 2000 frequency points requires only 5 seconds with a very good accuracy. Fig. 4. Comparison betweenROMandFEM: frequency response function of (a) 7th blade, (b) 9th blade 3. High order sensitivity analysis One of the main points of interest is to calculate the maximum amplitude with respect to the standard deviation of themistuned blades. In this way, themaximum stress can be determined. 360 L. Pohle et al. Whitehead (1966) described the theoretical possible maximum of the amplitude of a mistuned blisk by Amax Atuned = 1 2 ( 1+ √ N ) (3.1) withN denoting the number of blades. Here it is assumed that all vibration energy of all mode shapes is concentrated in one blade only. However, this case seems to be unreasonable for real bladed disk geometries. The common way to analyze the maximum amplitude of a mistuned bladed disk is the use of Monte-Carlo simulation (see Petrov, 2011; Siewert and Stüer, 2010; Beck et al., 2012). The disadvantages are the very high calculation time and furthermore the lack of reliability of the results, e.g. there is no statement howmanyMCS are necessary to have convincing results. Using a Weibull estimation, the number of necessary MCS can be defined. Nevertheless, a high number of simulations is needed to obtain a good agreement (50 MCS in Bladh et al. (2001) and500MCS inCastanier andPierre (2002)). Due to these problems, another method to calculate themaximumamplitude is given by a sensitivity analysis. This approach is motivated by Sextro et al. (2002), who described an extension of the theory of Sinha (1986) and Sinha and Chen (1989). To calculate the maximum amplitude with a given standard deviation of the stiffness of the blades, the sensitivity of the frequency response function is developed. By this means, an estimation of the maximum amplitude is very efficient and a statement of the probability of the amplitudes is given. The amplitude of the single blades can be written as x=x0±ασ (3.2) where x0 is the average of all blades. Using the reduced DOF u in place of x economizes a lot of calculating time. It should be pointed out again that the equations of motion are decoupled and could be solved using scalar functions. Thismeans that for calculation of themaximumand minimum limit of the tolerance interval of the amplitude, Eq. (3.2) can be written as u=u0±ασ (3.3) considering the transformationfinally.Basedonthe results ofMignolet et al. (1999), it is assumed that the maximum amplitudes over all blades distributions at one frequency follow a Gaussian distribution. σ is the standard deviation andα the tolerance interval.α=2means that 95.45% of all bladeamplitudesarewithin the tolerance interval; forα=3, 99.73%of all bladeamplitudes are within the interval. The definition given in Eq. (3.2) specifies the upper and lower limit for a given interval of the vibration amplitudes. In the case ofN discrete data points, the standard deviation σ is defined as σ= √√√√ 1 N −1 N∑ k=1 (uk−u0)2 (3.4) Here, the amplitude of the single blades are used to calculate the standard deviation. It is assumed that the average amplitude of the vibration u0 is the amplitude of the nominal, tuned system. The amplitude of the blade k is denoted as uk. The parameters of interest are the mistuning factors δk introduced in thepreviousSection.The functionof theamplitudedepending on the variation factors can be approximated by a Taylor series uk(δk)=u0+ ∞∑ h=1 1 h! (∂u(δk) ∂δk )h ∣∣∣∣∣ δk=1 (δk−1)h (3.5) The derivation of the equation ofmotion inEq. (2.5) with respect to themistuning factors reads ∂ui(δk) ∂δk = −(1+d0i) ( ∂K̃i ∂δk )˜̂ F i [−M̃iΩ2+(1+d0i)K̃i(δk)]2 (3.6) High order sensitivity analysis of a mistuned blisk... 361 whereonly the stiffnessmatrixdependson themistuning factors. K̃i denotes the scalar entry K̃i,i and M̃i the scalar M̃i,i. To calculate the higher terms of the Taylor series, the deviation with the higher order is ∂hui(δk) ∂δh k =±h!   (1+d0i) ( ∂K̃i ∂δk ) [−M̃iΩ2+(1+d0i)K̃i(δk)]︸ ︷︷ ︸ =S   h ui(δk) (3.7) with “+” for even h and “−” for uneven h. Using the ratio test, it can easily be shown that the Taylor series converges if the supremum of S is less than 1. For this blisk, the maximum is δk =10 −3 like the maximum standard deviations of the single blades. As a consequence, the variance of the amplitude results in σi = √√√√√ 1 N−1 N∑ i=1 ( ∞∑ h=0 ±(S)hui(δk) )2 (3.8) In this equation, all variances are known from the system given in Eq. (2.5), except for the deviation of the stiffness matrix from Eq. (2.21) with respect to the mistuning factors ∂K̃ ∂δk =   [ diag ( ∂K̃b,ss ∂δk )] 0 0 0 0 0 0 0 0   (3.9) with [ diag (∂K̃b,ss ∂δ1 )] =   K̃b,ss 0 ... 0 0   (3.10) followed by [ diag ( diag ∂K̃b,ss ∂δ2 )] =   0 0 K̃b,ss ... 0 0   (3.11) and so on, until the last one [ diag (∂K̃b,ss ∂δN )] =   0 0 ... 0 K̃b,ss   (3.12) InFigs. 5a and5b, the frequency response function is shown for a randommistuningpattern. Thevariance of the single blades is 10−3, and three terms of theTaylor series are used. InFig. 5a, the first isolatedmode family is shown.Obviously, the FRFs of all blades are between the upper and the lower limit. The tolerance interval is set to α = 2, therefore, the amplitude can be higher by 4.55% of the cases. Due to the upper limit, the deviation of the maximum amplitude is given for every frequencywith the probabilitywhich is needed. InFig. 5b, the region of nearby eigenfrequencies is shown.The limits showagoodagreementwith the randomlymistunedblades. 362 L. Pohle et al. Fig. 5. Upper and lower limit of the FRF compared with a randommistuning for: (a) the first mode withEO=1 and (b) the second and third mode with EO=3 In all studies with the random mistuning, the upper limit fits well. In Fig. 6, a comparison of three sensitivity analyses with different numbers of terms of the Taylor series are given. The amplitudes are very close together but the behavior of the FRF ismore detailed thanwithmore terms. Calculating 1000MCS, the PDF of all amplitudes can be compared with the PDF of the sensitivity analysis. The function of the PDF of the Gaussian distribution is given by p(x)= 1 σ √ 2π exp [ − 1 2 (x−µ σ )2] (3.13) with σ derived from Eq. (3.8) and µ= u0 as the amplitude of the nominal system. In Figs. 7a and 7b, the probability distributions of theMCS and theGaussian distrubtion of the sensitivity analysis using Eq. (3.13) are shown. Fig. 6. Comparison of different numbers of terms of the Taylor series (withoutMCS) High order sensitivity analysis of a mistuned blisk... 363 Fig. 7. (a) Comparison between 1000Monte-Carlo simulations (MCS) and the sensitivity analysis (SA) with σ calculated at the first eigenfrequency at 41.15Hz forEO=1; (b) comparison between 1000MCS and the PDFwith σ calculated at the third eigenfrequency at 372.33Hz forEO=4 4. Case study: intentional mistuning Thesecondmodelhas30bladeswhichhave anoff-axis angle.Thefirstmode is theoneof interest. To show the generality of the reduction method, an intentional mistuning is introduced. This case study underlines the possibilities of such a method. A comparison between the reduced order model and the Finite-Element model is given in Hohl et al. (2011). With the sensitivity analysis a lot of parameter studies are feasible without consulting the calculating time. One of the most important ways to handle the increase of the amplitude is to use two different types of blades. For this purpose, an alternative pattern is used called ABmistuning. The benefits of this method are shown in Castanier and Pierre (2002), Han and Mignolet (2008), Mignolet et al. (2000) and Tatzko et al. (2013). An optimal intentional mistuning pattern is found for the frequency response including an additional randommistuning. In this way, costly statistics are used requiringa lotof computation time.With the sensitivity analysis shownhere, the intentional mistuning can be classified as a simple method. To calculate the maximum amplitude of the pattern, the upper limit has to be calculated just once. Thismethod is not restricted to just two different blade types or a geometric deviation. The secondblade is slightly thicker at its platform region, seeFig. 9.Therefore, it is necessary to rebuild the finite element mesh. Assuming that the contact nodes are at the same locations like the nodes of the disk, both types of blades are reduced using the Craig-Bampton with the same master nodes, as it is described in Eq. (2.17). With the same mesh at the contact nodes, the waves are the same for all blades. Only the reduced stiffnessmatrix of the blades is changed to [diag(K̃b,ss)]=   δ1K̃ A b,ss 0 · · · 0 0 δ2K̃ B b,ss · · · 0 ... ... ... ... 0 0 · · · δNK̃Bb,ss   (4.1) So, it is very simple to realize the intentional mistuning with geometric differences. The intro- duced sensitivity analysis is a suitable approach to estimate different patterns of the intentional mistuning using as many different blade types as required. The frequency response function 364 L. Pohle et al. Fig. 8. FEM of the secondmodel blisk Fig. 9. Two different blade types, including intentional mistuning has to be calculated only once to have a good estimation for the pattern. In this paper, three different patterns are evaluated. The first version is an AB mistuning, where the blades are set in alternation. In the second version, three blades of the same type are collected and sorted alternately. In the third version, all blades of typeAare arranged in a row, followed by all blades of type B arranged in a row (see Fig. 10). These patterns are examples to show the benefits of the introducedmethods. The results are unique for every blade and disk geometry and have to be reproduced for each example. Regarding the vibration amplitude without aerodynamical coupling, the tuned system with only one blade type has the smallest amplitude. Due to areodynamical coupling, the intentional mistuning due to flutter or friction damping can be very useful. Nevertheless, the influence of the inevitable random mistuning has to be taken into account. With respect to the standard deviation of random mistuning, the three versions show small differences. With a very small standard deviation between 10−8 and 10−6, version 3 is the best one.With a standard deviation higher than 10−5, version 3 is the worst one. The chosen engine order is 2. High order sensitivity analysis of a mistuned blisk... 365 Fig. 10. Three patterns of the two blade types Fig. 11. Maximum amplitude of the three patterns with respect to different standard deviations of the blades atEO=2 Figure 12 shows the normalized amplitude with different engine orders with a standard deviation of 10−6. Versions 1 and 2 are the most sensitive patterns at the first engine order. 5. Conclusions In this paper, a newanalyticalmethodhas been introduced to estimate themaximumamplitude of amistuned bladed disk. Its benefits are proved by a simplemodelwith characteristic dynamic behavior.After a short introduction of themodel, theused reductionmethodhasbeenpresented. Therewith, the frequency response function has been calculated in aminimumof time andwith good accuracy. Using the sensitivity analysis, the maximum amplitude for a given interval has been estimated. For this purpose, the force response function and themaximumamplitude have 366 L. Pohle et al. Fig. 12. Normalized amplitude with respect to theEO with a standard deviation of 10−6 to be calculated only once. This saves a significant amount of time for optimization tools or parameter variations. Intentional AB mistuning has been presented as a possible application. Two types of blades have been used to analyze different patterns. The sensitivity analysis is proved to be a good tool for estimating intentional mistuning as well. Acknowledgment The authors kindly thank German Research Foundation (DFG) for the financial support of Col- laborative Research Center (SFB) 871 “Regeneration of Complex Capital Goods” which provided the opportunity of their collaboration in the research project C3 “Influence of Regeneration-inducedMistu- ning on the Dynamics of Coupled Structures”. References 1. Beck J.A., Brown J.M., Cross C.J., Slater J.C., 2012, Probabilistic mistuning assessment using nominal and geometry based mistuning methods,Proceedings of ASME TURBO EXPO 2. Bhartiya Y., Sinha A., 2011, Reduced order model of a bladed rotor with geometric nistuning: comparison between modified modal domain analysis and frequency mistuning approach, Proce- edings of ASME TURBO EXPO 3. BladhR.,CastanierM.P.,PierreC., 2001a,Component-mode-based reducedordermodeling techniques formistuned bladed disks – part 1: Theoreticalmodels, Journal of Engineering for Gas Turbines and Power, 123, 89-98 4. BladhR.,CastanierM.P.,PierreC., 2001b,Component-mode-based reducedordermodeling techniques formistunedbladeddisks–part2:Application,Journal of Engineering forGasTurbines and Power, 123, 100-108 5. Bladh R., Pierre C., Castanier M.P., Kruse M.J., 2001, Dynamic response predictions for amistuned industrial turbomachinery rotor using reduced-ordermodeling, Journal of Engineering for Gas Turbines and Power, 123, 100-108 6. Castanier M.P., Pierre C., 2002, Using intentionalmistuning in the design of turbomachinery rotors,AIAA Journal, 40, 10, 2077-2086 High order sensitivity analysis of a mistuned blisk... 367 7. Castanier M.P., Pierre C., 2006, Modeling and analysis of mistuned bladed disk vibration: Status and emerging directions, Journal of Propulsion and Power, 22, 2, 384-396 8. C̆ermelj P., Pluymers B., Donders S., Desmet W., Boltez̆ar M., 2008, Basis functions and their sensitivity in the wave-based substructuring approach,Proceedings of ISMA 9. Craig R.R., 2000, Coupling of substructures for dynaimc analyses: An overview,American Insti- tute of Aeronautics and Astronautics, 1573, 3-14 10. CraigR.R., BamptonM.C.C., 1968,Coupling of substructures for dynamic analyses,American Institute of Aeronautics and Astronautics, 6, 1313-1317 11. Feiner D.M., Griffin J.H., 2002, A fundamental model of mistuning for a single family of modes, Journal of Turbomachinery, 124, 597-605 12. Feiner D.M.,Griffin J.H., 2004a,Mistuning identification of bladed disks using a fundamental mistuning model – part I: Theory, Journal of Turbomachinery, 126, 150-158 13. Feiner D.M.,Griffin J.H., 2004b,Mistuning identification of bladed disks using a fundamental mistuning model – part II: Application, Journal of Turbomachinery, 126, 159-165 14. Griffin J.H., HoosacT., 1984,Model development and statistical investigation of turbine blade mistuning, Journal of Vibration, Acoustics, Stress, and Reliability, 106, 204-210 15. HanY.,MignoletM.P., 2008,Optimization of intentionalmistuning patterns for themitigation of the effects of randommistuning,Proceedings of ASME TURBO EXPO 16. Hohl A., Kriegesmann B., Wallaschek J., Panning L., 2011, The influence of blade pro- perties on the forced response of mistuned bladed disks,Proceedings of ASME TURBO EXPO 17. MignoletM.P.,HuW., 1997,Direct predictionof the effects ofmistuning on the forced response of bladed disks, International Gas Turbine and Aeroengine Congress and Exhibition 18. Mignolet M.P., Hu W., Jadic I., 2000, On the forced response of harmonically and partially mistuned bladed disks. part I: Harmonicmistuning, International Journal of RotatingMachinery, 6, 1, 29-41 19. Mignolet M.P., Rivas-Guerra A., LaBorde B., 1999, Towards a comprehensive direct pre- diction strategyof the effects ofmistuningonthe forced responseof turbomachineryblades,Aircraft Engineering and Aerospace Technology, 71, 5, 462-469 20. MoyroudF., FranssonT., Jacquet-RichardetG., 2002,A comparison of twofinite element reduction techniques for mistuned bladed disks, Journal of Engineering for Gas Turbines and Power, 124, 942-952 21. Petrov E.P., 2011, Reduction of forced response levels for bladed disks by mistuning: Overview of the phenomenon, Journal of Engineering for Gas Turbines and Power, 133, 072501-1-072501-10 22. Sextro W., Panning L., Goetting F., Popp K., 2002, Fast calculation of the statistics of the forced response of mistuning bladed disk assemblies with friction contracts,Proceedings of ASME TURBO EXPO 23. Shuai W., Jianyao Y., 2010, Mistuning identification for integrally bladed disks based on the SNM technique,Proceedings of ASME TURBO EXPO 24. Siewert C., Stüer H., 2010, Forced response analysis ofmistuned turbine bladings,Proceedings of ASME TURBO EXPO 25. Sinha A., 1986, Calculating the statistics of forced response of a mistuned bladed disk assembly, AIAA Journal, 24, 11 26. Sinha A., 2009, Reduced-order model of a bladed rotor with geometric mistuning, Journal of Turbomachinery, 131, 031007-1-031007-7 27. Sinha A., Chen S., 1989, A higher order technique to compute the statistics of forced response of a mistuned bladed disk assembly, Journal of Sound and Vibration, 130, 2, 207-221 368 L. Pohle et al. 28. Tatzko S., von Scheidt L.P., Wallaschek J., Kayser A., Walz G., 2013, Investigation of alternatemistuned turbine blades non-linear coupled by underplatform gampers,Proceedings of ASME TURBO EXPO 29. Thomas D.L., 1979, Dynamics of rotationally periodic structures, International Journal for Nu- merical Methods in Engineering, 14, 81-102 30. Whitehead D.S., 1966, Effect of mistuning on the vibration of turbomachine blades induced by wakes, Journal Mechanical Engineering Science, 8, 1, 15-21 31. Yang M.-T., Griffin J.H., 2001, A reduced-ordermodel ofmistuning using a subset of nominal systemmodes, Journal of Engineering for Gas Turbines and Power, 123, 893-900 Manuscript received January 29, 2015; accepted for print September 13, 2016