1.Introduction

A vast amount of uncertainties exist in the practical aircraft design and application,which can cause fluctuations of aircraft performance.Therefore,it is important to take these uncertainties into account at the beginning of aircraft design.1,2Recently,many researches have concerned the topics.3–8The Uncertainty quantification(UQ)and uncertainty sensitivity analysis of aerodynamics are concerned in the paper.

Computational Fluid Dynamics(CFD)technology has been widely used to solve problems in fluid mechanics with the development of computer technology.Traditional CFD simulation is deterministic.However,a variety of uncertainties are inevitably introduced into CFD simulation due to the increasing complexity of the problems.This leads to the mismatch between the results of CFD simulation and the actual results.The UQ in CFD simulation has gained extensive attention in Ref.9.Sources and classifications of uncertainty in CFD were described in Ref.10.Several UQ strategies have been used in CFD,including Monte Carlo Simulation(MCS)method,moment method and Polynomial Chaos(PC)in Ref.11.MCS is a statistical method,which needs many samples to accurately quantify uncertainty.Moment methods are suitable to solve the problem of small parameter uncertainty or linear model.Recently,PC which is based on the spectral representation of the uncertainty has been adopted in UQ for fluid problems in Refs.12,13.PC methods can be divided into intrusive and non-intrusive ones according to the coupling ways with CFD solvers.In general,an intrusive approach is adopted to obtain unknown polynomial coefficients by projecting resulting equations into basis functions for different modes,and it requires the modification of CFD codes,which may be difficult and time-consuming for complex problems such as Navier–Stokes equations.To overcome the shortcomings of intrusive polynomial chaos,Non-Intrusive Polynomial Chaos(NIPC)has been developed.The CFD is regarded as a black box model without changing the CFD program codes in non-intrusive methods.There are two different sampling approaches to build NIPC:random sampling and deterministic sampling.Random sampling methods use MCS to evaluate the unknown coefficients,but their convergence rate is low.Deterministic sampling methods use the quadrature to evaluate the unknown coefficients.The quadrature-based methods are more efficient than random sampling methods for low-dimensional problems.However,they are inefficient for relatively high-dimensional problems because of the exponential rising of quadrature points with the increasing dimensions.The UQ based on PC and its applications in fluid mechanics were comprehensively reviewed in Refs.14,15.

Recently,the NIPC is sufficiently used for stochastic aerodynamic analysis with operational uncertainties.A subsonic aerodynamic analysis was conducted on a NACA0012 airfoil with an uncertain free stream velocity using a commercial flow solver in Ref.16.They proved that an uncertain free stream velocity led to the highest variation in pressure on the upper surface near the leading edge.Transonic stochastic response of two-dimensional airfoil to parameter uncertainty(Mach number Ma and angle of attack α)is focused using generalized Polynomial Chaos(gPC)in Ref.8.Two kinds of non-linearities are critical to transonic aerodynamics in their study:the shock characteristics and boundary-layer separation.A stochastic investigation of flows about NACA0012 airfoil was conducted at transonic speeds in Ref.17.A point-collocation NIPC method was used to quantify uncertainty of aerodynamic characteristics with uncertain variables Ma and α in the transonic-wing case in Ref.18.A stochastic fluid analysis on 3D wind blades considering the wind speed as an uncertain parameter was conducted in Ref.19.In their studies,when the flow separation appears,the separation vortex region corresponds to the maximum variation area which extends to the trailing edge and even to the whole suction side.However,the stochastic aerodynamic analysis considering geometric uncertainty was rarely involved.The geometric uncertainty on an aerodynamic surface resulting from manufacturing errors has significant effect on the aerodynamic performance.It is impractical to remove the impact of these geometric variations by improving the manufacturing tolerance due to the high cost of the precise surface manufacturing technique.In other words,the geometric uncertainty due to manufacturing errors is unavoidable.Therefore,it is important and necessary to conduct a stochastic aerodynamic analysis considering geometric uncertainty.Nevertheless,the description of the geometric uncertainty is difficult in a computing environment and many random variables are needed to represent aerodynamic surface fluctuations.Therefore,the quadrature-based NIPC method is inefficient in solving this high-dimensional stochastic problem.

To improve the computational efficiency of traditional quadrature-based NIPC method for high-dimensional problems,the sparse grid numerical integration is introduced to solve the coefficients of PC.The sparse grid technique extends one-dimensional formulae to higher dimensions by tensor product and then selects sampling points according to Smolyak theory in Ref.20.It has been widely used in numerical integration and interpolation21,22as well as data mining.23Recently,a new sparse grid-based method has been developed for UQ in Ref.24.From their research,it can be known that when the dimension of the uncertain variables is larger than 5,the computational cost required by the sparse grid method is much smaller than that required by tensor product method.A sparse grid interpolant was developed to solve the high-dimensional stochastic partial different equations in Ref.25.Sparse grid collocation schemes were applied to stochastic natural convection problems in Ref.26.Sparse grids-based stochastic approximations with applications to subsonic steady flow about a NACA0015 airfoil in the presence of geometrical and operational uncertainties with both simplified aerodynamics model and Reynolds-Averaged Navier–Stokes(RANS)simulation was presented in Ref.27.

In this paper,a Sparse Grid-based Polynomial Chaos(SGPC)method is constructed to UQ and sensitivity analysis for transonic aerodynamics considering airfoil geometric and operational uncertainties in detail.The paper is structured as follows:Section 2 introduces the mathematical formulations of the sparse grid technique;in Section 3,the SGPC is built;in Section 4,a stochastic aerodynamic analysis considering geometric and operational uncertainties is conducted in detail;Section 5 outlines several useful conclusions.

2.Sparse grid numerical integration

Sparse grid technique selects sampling points under Smolyak theory,which uses a weighted linear combination of special tensor products to reduce the grid size.The sparse grid has been successfully used for numerical integration.Locations and weights of the univariate quadrature points with a range of accuracy are determined in each dimension,and then the univariate quadrature point sets are extended to form a multi-dimensional grid using the sparse grid theory.The introduction of Sparse Grid Numerical Integration(SGNI)is as follows:

Fortheone-dimensional integralthe quadrature formula is defined as

whereanddenote the one-dimensional quadrature points and weights which can be obtained by Gaussian quadrature.The one-dimensional integral can be extended to a multivariate integral by the full tensor product:

where⊗denotes the tensor product rule.

区域协同发展的重要性不言而喻，因为现代经济发展的本质要求是开放、分工与合作。一组相互协调良好的区域经济，可以使地理位置、要素禀赋和产业结构不同的地区承担不同的经济功能，实现单个孤立市场无法达到的规模经济和集聚效应。

Quadrature points of multivariate integrals are all possible combinations of one-dimensional quadrature nodes(Nfull=nm1nm2···nmn).Thus,it is time-consuming for a problem with relatively high dimensions by the tensor product algorithm.

Since the full tensor product is inefficient for high dimensions,Smolyak theory is adopted because it can reduce the grid size with a weight linear combination of special tensor products.For the n-dimensional sampling points with p-level accuracy generated by the sparse grid technique,the tensor product rule based on the sparse grid is shown as

where|m|=m1+m2+ ···+mn.The bounded sum makes sure that tensor products exclude from full grids points that contribute less to the improvement of the required integration accuracy.The integration of multivariate nonlinear functionin terms of variable ξ by sparse grid can be computed by

where Psis the sum of all possible combinations of the multiple indices that satisfy q=p+n,n is the number of random dimensions,and ξ is the selected collocation points by sparse grid method.The weight alcorresponding to the lth collocation point is computed by

Table 1 shows the change of sparse grid collocation points(Nsg)from 2 to 10 dimensions with the 2-order accuracy Gaussian quadrature(Ntp).In a low-dimensional problem,with the same level p,the sparse grid approach needs more collocation points than the full grid approach.When n increases(e.g.n≥5),the number of collocation points of the sparse grid method is much smaller than that of the full grid.It can be observed that with the same polynomial exactness p the number of collocation points produced by the full grid scheme is(p+1)nwhich increases exponentially with dimensionality,whereas the sparse grid technique remarkably reduces the number of collocation points.

3.Sparse grid-based polynomial chaos

以产教融合培养模式改革为重点，积极争取把辽东学院—中兴通讯ICT产教融合创新基地建设成为“教育部—中兴通讯ICT产教融合创新基地”，将创新与创业教育融入教育教学的全过程，整合专业基础课和专业课，调整、修订各本科专业人才培养方案，探索校企合作、中外合作、创新创业、项目教学等人才培养模式。同时，按照教育行政主管部门对应用技术大学的专业评价指标，积极开展专业自评，加强电子信息工程、信息管理与信息系统、网络工程、计算机科学与技术等本科专业水平的各项指标建设，从而提升专业竞争力。

两组患者在接受治疗后,实验组的抗卵巢抗体、抗精子抗体和抗子宫内膜抗体的转阴率均高于对照组,P<0.05。

PC is a stochastic method based on the spectral representation of uncertainty.According to the spectral representation,the random function can be decomposed into deterministic and stochastic components.For example,a random variable(X)can be represented by

where αj(x)is the deterministic component and Ψj(ξ)is the random basis function corresponding to the jth mode.From Eq.(6),the random variable X is the function of the deterministic independent variable vector x and the n-dimensional standard random variable vector ξ =(ξ1，ξ2，···，ξn).PC expansion given by Eq.(6)should contain an in finite number of terms.In a practical computational context,PC is truncated in both order p and dimension n.The number of truncated terms is finite as follows:

The random basis function Ψjis chosen according to the type of the input random variable.For example,if the input uncertainty obeys Gauss distribution,the basis function is the multidimensional Hermite polynomial.If ξ is chosen to be uniform with the random variable,the basis function must be the Legendre polynomial.A complete description of gPC scheme is introduced in Ref.13.

The purpose of the PC method is to obtain unknown polynomial coefficients.Eq.(6)can be transformed to Eq.(8)by the inner product:

Table 1 Number of collocation points with different dimensions by 2-order accuracy of Gaussian quadrature(tensor product and sparse grid).

Dimension 2 3 4 5 6 7 8 9 10 Nsg13 25 41 61 85 113 145 181 221 Ntp9 27 81 243 729 2187 6561 19683 59049

〈·〉represents the inner product which can be expressed by

Because of orthogonality,Eq.(8)can be transformed to

现代高校在推广图书馆“互联网+”阅读教育时，需要对学生的学习动机进行改善，针对上述分析的问题，校方应当在课堂教育等途径当中，让学生了解到未来自身将面临怎样的压力，使学生了解到其中的利害关系，有助于激发学生自主学习的欲望，而为了深化改善此点，校方还可以推出一系列的有奖竞赛，鼓励学生自主学习，在奖励的引导之下可以使学生重视阅读学习。此外，在选择有奖竞赛中的奖励的时候，应当奖励对学生的吸引力，可以选择一些具有教育意义的奖励，例如助学基金、野炊等。

涵养静气（吴语） ...........................................................................................................................................4-45

And then it can be derived:

PC has been widely used in UQ,but the method is inefficient for high-dimensional problems.Therefore,the SGPC method is constructed to alleviate the computational burden for relatively high-dimensional problems.The method is introduced in detail in this section.

There are two ways to solve the coefficients:intrusive and non-intrusive methods.The intrusive method computes unknown polynomial coefficients by projecting the resulting equations into basis functions for different modes.It requires the modification of the deterministic codes,which may be difficult,expensive,and time-consuming for complex computational problems.The non-intrusive method treats the CFD as a black box without changing the program code when propagating uncertainty.Steps of the non-intrusive method are as follows:

Step 1.Adopting relevant sampling methods to produce sample

Step 2.For each sample ξj,evaluate the uncertainty parameter λj.

对于表面人工湿地的作用机理进行如下总结：对于悬浮物来说，植物与基质层具有不错的节流效果。对于淹没在水中的植物根、叶等，在其表面上容易形成生物膜，这对于水的净化，有机物及营养物的净化都能起到较好的效果。

Step 3.Use the selected N samples to determine the expansion coefficients by Galerkin projection.

Step 4.Given the computed coefficient αk,a polynomial approximation model can be built∑

Sampling approaches can be divided into random and deterministic sampling methods in Step 1.Random sampling methods use MCS to compute projection integrals,where the convergence rate is low.Deterministic sampling methods use the numerical quadrature to evaluate unknown coefficients.Using n-dimensional Gauss quadrature with q points in each dimension,the unknown coefficients can be computed by

One-dimensional integral is expanded into a highdimensional integral by tensor product in Eq.(12),the calculation times of which are qn(requiring(p+1)npoints for pth order chaos).For low-dimensional problems,the efficiency of the deterministic sampling method has been greatly improved compared with that of the random sampling method.However,the computational cost grows exponentially with the increasing dimensions.Thus,the NIPC method is inefficient in high-dimensional problems.Such shortcomings motivate us to improve NIPC method.If we use much less integral points than the conventional tensor product to solve the multi-dimensional integral Eq.(8),the computational cost of UQ can be reduced.It is well known that the SGNI can improve the computational efficiency to solve the multidimensional integral Eq.(8).From the description in Section 2,the sparse grid method is more efficient than the tensor product for high-dimensional integration(e.g.n≥5).Hence,we use the SGNI to replace the tensor product to solve Eq.(8)as follows:

据介绍，近三年瓮福农资公司将瓮福新品成功推广至神州大地，“DAP+”、磷酸二氢钾、聚磷酸铵等新型肥料在江苏、新疆、内蒙古等15个省（自治区），针对小麦、棉花、马铃薯、葵花、苹果等30多种作物开展了试验示范，试验总面积上万亩。秉着认真科学的态度，每个试验示范均进行了全过程跟踪。测产结果显示，瓮福新品在作物上的表现都超出预期，在广大农民朋友心目中建立了一流的口碑，同时也更加坚定了广大合作伙伴与瓮福一起做新品的信心。

Table 4 shows uncertainty analysis results of aerodynamic coefficients.The Cov of CDis relatively small for Case A and Case B while it is relatively large for Case C.The Cov of CLis small for all the three cases.This means that subsonic aerodynamic coefficients are not sensitive to the geometric uncertainty.The drag is sensitive to the geometric uncertainty,but the lift is not in the transonic region.

4.Stochastic aerodynamic analysis

A stochastic aerodynamic analysis is conducted considering airfoil geometric and operational uncertainties.The operational uncertainty generally contains two parameters:Ma and α.However,the relatively high-dimensional random parameters are needed to represent aerodynamic surface fluctuations.In this section,SGPC and MCS methods are applied to the stochastic transonic aerodynamics analysis around a RAE2822 airfoil.Firstly,the deterministic flow computations are briefly introduced.

4.1.Deterministic flow computations

The deterministic steady-state flow solutions are computed by the RANS equations or method combined with the Spalart–Allmaras turbulence model.The cell-centered finite volume method is used for spatial discretization and AUSM-up scheme is used to evaluate the numerical flux.The implicit dual-time stepping method is used for temporal discretization.The symmetric Gauss–Seidel iterative time-marching scheme is applied in the pseudo-time step,and the second order accurate full implicit scheme is used to solve the equation in the physical time step.The developed flow solver is also used for unsteady flows.28,29

The computational mesh surrounding a RAE2822 airfoil is based on a structured C-grid.The 2D mesh is composed of two blocks,the sizes of which are 300×150(C block surrounding the airfoil)and 200×150.The chord of the airfoil is c=1 m and the far- field boundary is placed at a distance d=20c from the airfoil.Fig.1 shows the computing grids of RAE2822 airfoil and the reliability verification of the CFD program,in which Cpis the pressure coefficient.The geometry of airfoil keeps changing in the process of UQ,and Radial Basis Function(RBF)is used to mesh deformation.30

Fig.1 Computing grids of RAE2822 airfoil and reliability verification of CFD program.

4.2.Geometric uncertainty

Fig.2 Eigenvalue with number of modes.

In aerospace Engineering,despite advances of manufacturing engineering techniques,airfoilsvery often exhibitsome deviation from their intended shape due to the noisy manufacturing process.Moreover,it is often impractical to remove the impact of these geometric variations by improving the manufacturing tolerance due to the high cost of precise surface manufacturing techniques.In other words,the geometric uncertainty resulting from manufacturing errors is unavoidable.How to describe this geometric uncertainty in computing environment should be concerned.

Figs.11 and 12 show the results of Sobol’s analysis of aerodynamic coefficients.It can be observed that α is the most important to lift coefficients.To drag characteristic,Ma is the most important;geometric modes of the upper surface and angle of attack also have obvious influences.

Main geometric variation modes are obtained by Principal Component Analysis(PCA)with a large amount of geometric statistical measurement data of an airfoil in Refs.31,32.The description of the airfoil geometric uncertainty by PCA is shown as

where gnis the nominal geometry;¯g is the average geometric variation;viis the geometric mode shape;nsis the number of mode shapes used to represent the variation in geometry.The geometric mode can be computed by PCA based on measurement samples.σiis the ith singular value of the measurement snapshot matrix,which represents the geometric variability attributable to the ith mode.ziis a random parameter which obeys the standard normal distribution,and thus the product σiziis the stochastic contribution of the ith mode.

14.Knio OM,Maitre OPL.Uncertainty propagation in CFD using polynomial chaos decomposition.Fluid Dyn Res 2006;38:616–40.

Fig.3 RAE2822 airfoil deformation of the first six modes.

It is difficult to describe this geometric variation in the computing environment.A Gaussian random process simulation is used to obtain the geometric data in Ref.33.Then PCA is used to obtain main geometric modes.In addition,a variety of parametric methods have been employed to describe the geometric variation in aerodynamic design so far,such as PARSEC-11 geometry parameterization,34Class-Shape function Transformation(CST)method,35Free-Form Deformation(FFD)method,36and Hicks-Henne bump functions.37,38By changing the parameter of these parametric methods,the geometric variation is realized.Generally,these parametric methods need a lot of parameters to represent the airfoil shape.To reduce the dimensions of the variables,the PCA technology combined with airfoil parameterization is used to describe the geometric variation39,40used in the paper.Firstly,a parametric method is used to generate a set of sample data,and then the PCA is used to obtain the main deformation modes based on the generated sample data.In this paper,we use a parameterized representation of the airfoil ARE2822 by CST method with 12 parameters.The data of the measurement points on the airfoil surface are obtained by random perturbation of CST parameters.In this way,the PCA based sample data are conducted.Fig.2 shows the eigenvalue with the number of modes, which represents the geometric variation attributable to each mode.The smaller the eigenvalue of one mode is,the smaller the proportion of the mode to geometric variation is.The first 6 modes obtained by PCA are showed in Fig.3.It can be observed that these modes are global geometric deformation modes and present some typical geometric deformation.specifically,Mode 1 and Mode 2 are the scale modes in the thickness direction;Mode 3 and Mode 4 are translation modes of the maximum thickness in the axial direction;Mode 5 and Mode 6 are the extrusion modes of the upper surface.

4.3.Stochastic transonic aerodynamics analysis

Fig.4 Range of geometric uncertainty.

In the paper,the uncertainties are specified by means of PDFs.The input uncertainties,which PDFs should obey,accord to the corresponding statistical information,which are found in the methods of statistical inference.In the current study,the uncertain geometric variables are zi(i=1,2,...,6)in Eq.(14)which obey the standard normal distribution according to Ref.26.The operational uncertainties(Ma and α)obey truncated Gaussian distribution.The Mach number has a 0.73 mean value and a±0.01 variability,and the angle of attack has a 2.5°mean value and a± 0.3°variability.The steady flow state is selected in a transonic region(Ma=0.73,α=2.5°,Re=3.0× 106),and the range of geometric uncertainty is showed in Fig.4.

4.3.1.Verification of SGPC

⑭ The European Union，“Consolidated Versions of The Treaty on European Union and The Treaty on The Functioning of The European Union”，Official Journal of the European Union，Volume 53，C 83，p．91．

Fig.5 Standard deviation distribution of pressure coefficient along chord.

Fig.6 Standard deviation distribution of skin-friction coefficient along chord.

Table 2 Results of uncertainty quantification of aerodynamic coefficients.

Aerodynamic coefficient Statistical moment SGPC MCS p=1 p=2 p=3 CLMean 0.7379 0.7366 0.7377 0.7373 StD 0.0132 0.0201 0.0201 0.0196 CDMean 0.0164 0.0164 0.0164 0.0164 StD 0.0019 0.0014 0.0014 0.0014

Table 3 Comparison of CFD calls among selected methods.

Method SGPC MCS p=1 p=2 p=3 p=1 p=2 p=3 CFD calls 17 145 849 256 6561 65536 5000 NIPC

Fig.14 shows error bars of pressure and friction coefficient distribution of RAE2822 airfoil in the three states.In case A,variations of Cpand Cfare the same,and the maximum fluctuations mainly lie in the upper surface near the leading edge.In Case B,we can also observe that the maximum fluctuations mainly display on the upper surface near the leading edge.Fluctuations in other regions are even smaller than those in Case A.This indicates that flow characteristics in the separation region are intensive to this geometric uncertainty.In Case C,variation regularities of Cpand Cfare different.For Cp,the variation mainly appears in the shock wave disturbance region,and for Cf,besides this region,it also exists in the boundary layer separation region.

Figs.5 and 6 show the standard deviations of the pressure coefficient Cpdistribution and skin-friction coefficient Cfdistribution along the chord and they also re flect the convergence of SGPC method.It can be seen that the standard deviation distributions of Cpand Cfcoincide well with the results of MCS when SGPC reaches the 3-order accuracy.Table 2 also gives the convergence of SGPC for lift coefficient CLand drag coefficient CD.It can be observed that when p reaches 2-order accuracy,the Standard Deviation(StD)of aerodynamic coefficients is convergent and coincides well with the results of MCS.The convergence study of SGPC verifies the accuracy and correctness of SGPC to the stochastic aerodynamic analysis.

The computational cost of SGPC should also be concerned for the stochastic transonic aerodynamic analysis.From Section 3,it can be learned that the computational cost of SGPC is related with parameters n and p.Table 3 shows computational costs of SGPC,NIPC and MCS for the 8-dimensional problem.It can be observed that SGPC is more efficient than the other two.It must be emphasized here that NIPC method is not used to the stochastic aerodynamic analysis because of the prohibitive computational cost.The CFD calls of NIPC is obtained according to the formula(p+1)n.

4.3.2.Uncertainty quantification of transonic aerodynamics

Fig.7 Fluctuation ranges of Cpand corresponding local PDF pro files at selected five chord locations.

Fig.8 Fluctuation ranges of Cfand corresponding local PDF pro files at selected five chord locations.

Now,the in fluence of input uncertainties on transonic flows is studied.Fig.7 shows the mean and fluctuations of Cpon the airfoil surface.It is well known that the shock wave exists in the transonic flow.The flow characteristics before and after the shock wave change dramatically with the input uncertainties.Therefore,it can be observed that a strong spurious effect appears near the shock wave foot.Fig.8 shows the mean and fluctuation of Cfon the airfoil surface.The skin-friction behavior displays discrepancies from the pressure coefficient distribution.The variation of Cfis similar to that of Cpbefore and in the shock disturbance region.However,the difference occurs in the downstream region of shock wave;the standard deviation has larger magnitudes below the shock disturbance region,and then it gradually decreases.This indicates that the flow characteristics not only near the shock wave but also in the boundary-layer separation are sensitive to geometric and operational uncertainties.

上世纪八十年代，邓小平曾经提出体制改革的设想；近年中央纪委研究室也曾指出，影响反腐败成效，体制障碍是最大的障碍，机制缺陷是根本的缺陷。那么涉及到体制改革，诸如民主与法治、土地产权、知识产权、权力的监督与限制等体制上的问题，现在仍然存在着种种不同的认识，影响相关领域改革的深化。

Moreover,Figs.7 and 8 show the corresponding local PDF pro files at the selected five chord locations.It is well known that once the mean position of the shock wave is constrained by the separated shear layer,the probability density function of the solution may exhibit a bifurcation corresponding to a jump in the solution.Therefore,the stochastic response shows a bimodal nature at x/c=0.56.

4.3.3.Global sensitivity analysis of transonic aerodynamics

Global Sensitivity Analysis(GSA)should be applied to study how uncertainty of the model output can be apportioned to different sources of uncertainty of the model input.A sensitivity analysis is performed by using the Sobol’s analysis,which analyzes the individual and coupled effects between the random parameters(a deep introduction of the Sobol’s analysis can be referred to in Ref.41).When a PC model is built,the Sobol’s analysis based on the PC can be directly calculated from the expansion coefficients.41In the paper,the aerodynamic load distribution and aerodynamic coefficients are included for Sobol’s analysis.

Fig.9 Results of Cpdistribution by Sobol’s analysis.

Fig.10 Results of Cfdistribution by Sobol’s analysis.

Figs.9 and 10 show the results of Sobol’s analysis of Cpand Cf.The partial standard deviation and standard deviation caused by the coupling effect for each uncertainty variable are given.It can be observed that Ma is important to the flow characteristics both in the vicinity of the shock wave and in the boundary-layer separation.Now,we concern the contribution of the geometric deformation modes.It can be observed that deformation modes of the upper surface are much important than those of the lower surface to aerodynamic load distribution.Specifically,the scale mode(Mode 1)and extrusion mode(Mode 5)are important to the flow characteristic of shock wave and boundary-layer separation,but the translation mode(Mode 3)is only important to the flow characteristic of shock wave.Moreover,it can also be observed that the coupling effect among geometric deformation modes is not negligible near the shock wave.

Fig.11 Results of Sobol’s analysis of lift coefficient.

Fig.12 Results of Sobol’s analysis of drag coefficient.

Synthesis and crystal structure of 10-phenyl-6,7,8,10-tetrahydro-cyclopenta[b]pyrrolo[3,2-f]quinolin

4.4.Stochastic aerodynamic investigation considering geometric uncertainty in different flow regimes

A further stochastic aerodynamic investigation is conducted in different flow regimes.Three flow states are selected:Case A,subsonic flow with a small angle of attack(Ma=0.5,α=2.5°);Case B,subsonic flow with a large angle of attack(Ma=0.5,α =20°);and Case C,transonic flow with a small angle of attack(Ma=0.73, α =2.5°).SGPC with 3-order accuracy is used in the stochastic aerodynamic analysis.At first,the three kinds of flow states have been employed in the steady flow computation,respectively.The streamline of the flow field is shown in Fig.13.It can be observed that the flow is smooth and there is no separation along the airfoil in Case A.There is a separation vortex above the upper surface of the airfoil in Case B.In Case C,the separation appears on the aft part of the airfoil because of the boundary layer separation.

A convergence study of SGPC has been performed by the stochastic transonic aerodynamic analysis.The polynomial order p is increased to enhance the accuracy of SGPC.Moreover,MCS is also used in this analysis,which aims to validate the accuracy and efficiency of SGPC.The number of CFD calls of MCS is 5000.

whereis the number of collection points of n-dimension and p-accuracy by SGNI and alis the lth coefficient corresponding to the lth collection point.Given the same polynomial accuracy p, sampling points can be reduced dramatically in high-dimensional problemsThe combination of sparse grid and PC retains the advantages of PC and reduces the com put ational cost for highdimensional problems.After computing the unknown coefficients,the approximate mathematical model is built,and then the model is used to compute the first two statistical moments and sensitivity in dices.Moreover,the Probability Density Functions(PDF)can be easily evaluated.

Fig.13 Streamlines of steady flow in three flow states.

Fig.14 Error bars of pressure coefficient and friction coefficient(99%).

Table 4 Uncertainty quantification of aerodynamic coefficients in three different flow states.

Aerodynamic coefficient Flow state Mean StD Cov CDCase A(Ma=0.5,α =2.5°) 0.0116 0.0002 0.0172 Case B(Ma=0.5,α =20°) 0.0642 0.0010 0.0156 Case C(Ma=0.73,α =2.5°) 0.0163 0.0008 0.0491 CLCase A(Ma=0.5,α =2.5°) 0.4993 0.0072 0.0144 Case B(Ma=0.5,α =20°) 1.0410 0.0159 0.0153 Case C(Ma=0.73,α =2.5°) 0.7192 0.0130 0.0181

5.Conclusions

In the paper,the sparse grid-based polynomial chaos expansion was used for UQ and GSA for transonic aerodynamics with geometric and operational uncertainties.The method is much more efficient than NIPC and MCS in dealing with relatively high-dimensional stochastic aerodynamic problems.The accuracy and efficiency of the SGPC are verified by the stochastic transonic aerodynamic analysis.It is proved that the method is more efficient than NIPC and MCS.It is observed that the flow characteristics in the vicinity of the shock wave and the boundary-layer separation region are sensitive to uncertainties because of the nonlinear flow characteristics of shock wave and boundary-layer separation.By uncertainty sensitivity analysis(Sobol’s analysis),it can be learnt that Ma is important to the flow characteristics both in the vicinity of the shock wave and in the boundary-layer separation;the geometric deformation modes of the upper surface are much important than those of the lower surface to aerodynamic load distribution.Specifically,the scale mode(Mode 1)and extrusion mode(Mode 5)are important to the flow characteristic of shock wave and boundary-layer separation,but the translation mode(Mode 3)is only important to the flow characteristic of shock wave.The coupling effect is not negligible in the vicinity of the shock wave foot.

Moreover,a stochastic aerodynamic investigation considering geometric uncertainty is conducted by the SGPC in different flow states.It can be concluded that fluctuations of subsonic aerodynamic characteristics are mainly concentrated in the airfoil head.When the flow is in the transonic condition,the affected region shifts from the upper surface near the leading edge to the shock wave and boundary-layer separation region because of the nonlinear characteristics of the shock wave and boundary-layer separation behind the shock wave.

当前，世界经济延续2017年全球同步复苏的良好势头，经济增长根基强健，但也面临诸多不稳定不确定因素。这为我国在新时代深入推进供给侧结构性改革、促进更高水平改革开放、实现经济高质量发展提供了机遇，也带来了挑战。我国应进一步加强研判，综合施策，趋利避害，防范风险。

Acknowledgements

This work was supported by the National Natural Science Foundation of China(No.11572252),the ‘111” Project of China(No.B17037),and the National Science Fund for Excellent Young Scholars(No.11622220).

References

1.Zang TA,Hemsch MJ,Hilburger MW,Kenny SP,Luckring JM,Maghami P,et al.Needs and opportunities for uncertainty-based multidisciplinary design methods for aerospace vehicles.Washington,D.C.:NASA;2002.Report No.:NASA/TM-2002-211462.

2.Yao W,Chen XQ,Luo WC,Tooren MV,Guo J.Review of uncertainty-based multidisciplinary design optimization methods for aerospace vehicles.Prog Aerosp sci 2011;47(6):450–79.

3.Song SF,Lu ZZ,Zhang WW,Ye ZY.Reliability and sensitivity analysis of transonic flutter using improved line sampling technique.Chin J Aeronaut 2009;22(5):513–9.

4.Tang J,Wu Z,Yang C.Epistemic uncertainty quantification in lf utter analysis using evidence theory.Chin J Aeronaut 2015;28(1):164–71.

5.Zhu H,Tian H,Cai GB,Bao WM.Uncertainty analysis and design optimization of hybrid rocket motor powered vehicle for suborbital flight.Chin J Aeronaut 2015;8(3):676–86.

6.Dai Y,Yang C.Methods and advances in the study of aeroelasticity with uncertainties.Chin J Aeronaut 2014;27(3):461–74.

7.Pagnacco E,Souza DCE,Sampaio R.Subspace inverse power method and polynomial chaos representation for the modal frequency responses of random mechanical systems.Comput Mech 2016;58:129–49.

8.Simon F,Guillen P,Sagautn P,Lucor D.A gPC-based approach to uncertain transonic aerodynamics.Comput Method Appl Mech Eng 2010;199:1091–9.

9.Roache PJ.Quanti fication of uncertainty in computational fluid dynamics.Annu Rev Fluid Mech 1997;29:123–60.

10.Pelletier D,Turgeon E,Lacasse D.Adaptivity,sensitivity,and uncertainty:Toward standards of good practice in computational fluid dynamics.AIAA J 2003;41(10):1925–32.

破解土地发展制约 打造乡村振兴战略（金钦帅） ...........................................................................................1-19

11.Walters RW,Huyse L.Uncertainty analysis for fluid mechanics with applications.Washington,D.C.:NASA;2002.Report No.:NACA/CR-2002-211449.

12.Mathelin L,Hussaini MY,Zang TZ.Stochastic approaches to uncertainty quantification in CFD simulations.Numer Algorithms 2005;38(1):209–36.

13.Xiu DB,Karniadakis GE.Modeling uncertainty in flow simulationsvia generalized polynomialchaos.J ComputPhys 2003;187:137–67.

那么，法院的行为是否经得起检验？家属的质疑又是否站得住脚？记者日前来到福建龙岩，试图解出事件背后的疑团。

15.Najm HN.UQ and polynomial chaos techniques in computational lf uid dynamics.Annu Rev Fluid Mech 2009;41:35–52.

16.Loeven GJA,Witteveen JAS,Bijl H.Probabilistic collocation:An efficient non-intrusive approach for arbitrarily distributed parametric uncertainties.45th AIAA aerospace sciences meeting and exhibit;2007 Jan 8-11;Reno,Nevada.Reston:AIAA;2007.

17.Chasseing JC,Lucor D.Stochastic investigation of flows about airfoils at transonic speeds.AIAA J 2010;48(5):918–49.

18.Hosder S,Walters RW,Balch M.Point-collocation nonintrusive polynomial chaos method for stochastic computational fluid dynamics.AIAA J 2012;48(12):2721–30.

19.Liu ZY,Wang XD,Shun K.Stochastic performance evaluation of horizontal axis wind turbine blades using non-deterministic CFD simulations.Energy 2014;73:126–36.

20.Bungartz HJ,GriebelM.Sparse grids.Acta Numerica 2004;13:147–269.

21.Thomas G,Michael G.Numerical integration using sparse grids.Numer Algorithms 1998;18:209–32.

22.Novak E,Ritter K.High dimensional integration of smooth functions over cubes.J Numer Math 1996;75:79–97.

23.Garchke J,Griebel M,Thess M.Data mining with sparse grids.Computing 2001;67:225–53.

24.Xiong FF,Greene S,Chen W.A new sparse grid based method for uncertainty propagation.Struct Multidisc Optim 2010;41:335–49.

25.Xiu D,Hesthaven J.Higher-order collocation method for differential equations with random inputs.SIAM J Sci Comput 2005;27(3):1118–39.

26.Baskar G,Nicholas Z.Sparse grid collocation schemes for stochastic natural convection problems.J Comput phys 2007;225:652–85.

27.Resmini A,Peter J,Lucor D.Sparse grids-based stochastic approximations with applications to aerodynamics sensitivity analysis.Int J Numer Meth Eng 2016;1(106):32–57.

28.Zhang WW,Li XT,Ye ZY,Jiang YW.Mechanism of frequency lock-in in vortex-induced vibrations at low Reynolds numbers.J Fluid Mech 2015;783:72–102.

29.Zhang WW,Gao CQ,Liu YL,Ye ZY,Jiang YW.The interaction between flutter and buffet in transonic flow.Nonlinear Dynam 2015;82(4):1851–65.

30.Jakobsson S,Amoignon O.Mesh deformation using radial basis functions for gradient-based aerodynamic shape optimization.Comput Fluids 2007;36:1119–36.

31.Garzon V,Darmofal D.Impact of geometric variability on axial compressor performance.J Turbomac 2003;125(4):692–703.

32.Thanh TB,Willcox K.Parametric reduced-order models for probabilistic analysis of unsteady aerodynamic applications.AIAA J 2008;46(10):2520–9.

33.Chen H,Wang QQ,Hu R,Paul C.Conditional sampling and experiment design for quantifying manufacturing error of transonic airfoil.49th AIAA aerospace sciences meeting including the new horizons forum and aerospace exposition;2011 Jan 4–7;Orlando,USA.Reston:AIAA;2011.

34.Dodson M,Parks GT.Robust aerodynamic design optimization using polynomial chaos.J Aircraft 2009;46(2):635–45.

35.Kulfan B,Bussoletti J. ‘Fundamental” parameteric geometry representations for aircraft component shapes.11th AIAA/ISSMO multidisciplinary analysis and optimization conference;2006 Sep 6–8;Portsmouth,USA;Reston:AIAA;2006.p.1–45.

36.Padulo M,Maginot J,Guenov M.Airfoil design under uncertainty with robust geometric parameterization.50th AIAA/ASME/ASCE/AHS/ASC structures,structural dynamics,and materials conference;2009 May 4–7;Palm Springs,USA.Reston:AIAA;2009.

37.Duan YH,Cai JS,Li YZ.Gappy proper orthogonal decomposition-based two-step optimization for airfoil design.AIAA J 2012;50(4):968–71.

38.Padulo M,Campobasso MS,Guenov MD.Novel uncertainty propagation method for robust aerodynamic design.AIAA J 2012;49(3):530–43.

39.Wu XJ,Zhang WW,Song SF.Uncertainty quantification and sensitivity analysis of transonic aerodynamics with geometric uncertainty.Int J Aerospace Eng 2017;8107190:1–16.

40.Wei PF,Lu ZZ,Song JW.Variable importance analysis:A comprehensive review.Reliab Eng Syst Saf 2015;42:399–432.

41.Bruno S.Global sensitivity analysis using polynomial chaos expansions.Reliab Eng Syst Saf 2008;93:964–79.