Abstract
This paper addresses the construction of a dynamical model for a thinwalled beam with circular crosssection in the framework of onedimensional higherorder beam theory. And a method for pattern recognition of circular thinwalled structures is proposed based on principal component analysis. Initially, a set of equal length linear segments are defined to discretize the midline of a circular section. Preliminary deformation modes of thinwalled structures, defined over the crosssection through kinematic concept, are parametrically derived through changing the discretization degree of the section. Next, the generalized eigenvectors are derived from the governing equations, and the characteristic deformation modes of circular sections with different discretization degrees are solved based on principal component analysis. In addition, a reduced higherorder model can be obtained by updating the initial governing equations with a selective set of crosssection deformation modes. The features include further reducing the number of degree of freedoms (DOFs) and significantly improving computational efficiency while ensuring accuracy. For illustrative purposes, the versatility of the procedure is validated through both numerical examples and comparisons with other theories.
Highlights
 A dynamic model of thinwalled beam with circle section is constructed base on 1D higherorder beam theory.
 The method relies on crosssection deformation data of thinwalled beams in free vibration.
 Further selection of deformation modes condenses DOFs of refined higherorder beam model.
 Examples verify the modeling accuracy and computation efficiency in dynamic analysis.
1. Introduction
Thinwalled structures are now widely used in many applications, such as civil engineering, automotive industry, aerospace and marine engineering, due to their low weight, high flexural capacity and easy fabrication [1][2]. Cylindrical thinwalled structures have higher compressive strength and better mechanical stability than traditional prismatic thinwalled structures. It is essential to develop a cylindrical thinwalled structural mechanics model with high accuracy and fast calculation speed.
In terms of simplifying calculations and improving efficiency, onedimensional (beam) models have more advantages than twodimensional (plate/shell) and threedimensional (solid) approaches because they fit geometric dimensions, consider the ease of design and time cost, and also possess advantages in interpreting the structural response from the viewpoint of crosssection property. Although the static and dynamic behavior of a thinwalled beam structure can be accurately evaluated using twodimensional shell elements, it is difficult to interpret its structural response from the viewpoint of the crosssection property [3]. To avoid unnecessary complexity in describing the beam response with precise threedimensional methods, different simplified onedimensional theories have been developed in the last decades [4]. Their list includes EulerBernoulli theory which is suitable for slender beams and Timoshenko theory by taking into account the effects of shear stress and rotary inertia. However, they all overlook significant section deformations of thinwalled beams, such as warping and distortion, and cannot accurately reflect the mechanical properties of thinwalled structures.
At present, a higherorder beam theory is used to appropriately reflect the effect of crosssection variation, by employing higherorder crosssection deformations as field variables. Threedimensional displacements at a general point of a beam are approximated using crosssection shape functions and onedimensional deformation field variables. By doing so, threedimensional elasticity formulation can be reduced to a onedimensional model. Since the early study by Vlasov on nonuniform warping is defined and used to refine the displacement field of thinwalled beams, many studies on higherorder beam theories have been carried out. For example, Kim and Kim [5] analytically derived the shape functions of the torsional and bending distortions for a rectangular cross section by assuming constant tangential displacement; Yoon et al. [6] present an efficient warping model for nonlinear elastoplastic torsional analysis of composite beam developed based on Benscoter warping theory; Shin et al. [7] newly derived section shape functions for composite thinwalled box beams for efficient structural analysis of composite thinwalled box beams; Nguyen et al. [8] proposed a beam frame modal approach based on higherorder beam theory for the analysis of thinwalled functionally graded straight and curved beams with general nonuniform polygonal crosssections; Habtemariam et al. [9][10] established dynamic models of thinwalled pipes under different boundary conditions by taking into account deformation modes such as bending, warping, and torsion based on generalized beam theory (GBT).
Over past years, scholars have conducted extensive works on GBT, which originates from the work of Schardt [11], and has been extended into almost all fields of structural analysis of thinwalled beams [12][15]. GBT is an extension of the classic Vlasov beam theory, being presently wellestablished as an efficient, versatile, accurate and insightful approach to assess the structural behavior of prismatic thinwalled beams. Through decades of continuous refinement, GBT has been an efficient tool to perform buckling [16], post buckling [17] and dynamic analysis [18] of thinwalled beams in elasticity. Accordingly, many attempts have been made to extend GBT to the design and calculation of cylindrical thinwalled structures over the years. For example, Silvestre [19] proposed the assumption of partially overcoming null transverse extension and membrane shear strain by considering axisymmetric and torsion deformation modes; Basaglia et al. [20] evaluated the buckling analysis of cylindrical shells under a combination of axial compression and external pressure based on GBT; Peres et al. [21] proposed an extension GBT that enables the calculation of globallocal bifurcation loads and associated buckling mode shapes, for thinwalled members with circular axis; de Miranda et al. [22], Gonçalves et al. [23], Muresan et al. [24] and Sahraei et al. [25] incorporated shear deformation effects into a GBT formulation for circular and prismatic thinwalled crosssection to improve displacement and stress field results.
In this paper, a dynamic model for a thinwalled structure considering circular crosssection characteristics is established based on onedimensional higherorder dynamics theory. And a crosssection analysis procedure is proposed for the polygonal approximations of curved geometries. The preliminary deformation modes based on the displacement continuity condition are established on the crosssection midline and linearly superimposed to construct the threedimensional displacement field. It is worth noting that a simple procedure is proposed to parameterize the degree of discretization when discretizing the midline of a circular section. The generalized matrix containing all modal information is derived from the governing equations, and then the main crosssection deformation modes of the onedimensional higherorder model are identified by principal component analysis. In this aspect, principal component analysis takes particular advantage since it can directly extract deformation patterns hiding in free vibration behaviors, achieving dimension reduction with minimum information loss. Next, the identified deformation modes are compared and analyzed to determine the range of discretization that can accurately describe the dynamical behavior of circular thinwalled structures. Then, to further improve the computational efficiency, a reduced higherorder model can be obtained by updating the initial governing equations with a selective set of crosssection deformation modes. The resulted refined model possesses advantages not only in a minimum number of DOFs, but also from clear physical interpretation of the deformation modes coming from real structural behaviors. In practice, it ﬁts the geometric dimensions of thinwalled beams, and takes into account the ease of design and analysis time cost.
2. Preliminary higherorder beam model
The preliminary higherorder beam model for cylindrical thinwalled structures is presented considering crosssection deformation. Specially, a set of equal length linear segments with the number of ${N}_{s}$ is defined to discretize the midline of the circular section. For the sake of generality, a circular thinwalled beam whose crosssection midline is discretized into 16 equallength straight segments shown in Fig. 1(a) and Fig. 1(c) is taken as an illustrative example, with the length of equallength straight segments and the crosssectional radius denoted by $a$ and $r$, respectively, and the parameters $l$ and $t$ respectively represents the overall length and wall thickness of the circular thinwalled beam. The global and local coordinate systems are also shown (Fig. 1(a) and Fig. 1(b)). Where $s$ being the coordinate along the midline, $n$ indicating the perpendicular direction to the midline of the wall.
Fig. 1Circular thinwalled structure: a) the global coordinate system, b) the local coordinate system, and c) the discretization of the crosssection centerline with a series of nodes
a)
b)
c)
Correspondingly, 16 nodes are discretized on the midline of the thinwalled beam crosssection shown in Fig. 1(c), and axial, tangential, normal and torsional unit displacements are applied at each node. At the same time, the adjacent nodes are constrained to have zero displacement, resulting in four fundamental deformation modes for each node. For a better presentation, Fig. 2 shows four elementary deformed shapes obtained with unit displacements separately imposed on node 3.
2.1. Displacement field
According to onedimensional higherorder theory, the displacement field d of the circular thinwalled structure crosssection is described by three components, namely $u(s,z)$, $v(s,z)$, $w(s,z)$, which are expressed as:
where ${N}_{a}$ is the number of deformation modes considered, ${\alpha}_{i}\left(s\right)$, ${\phi}_{i}\left(s\right)$, and ${\omega}_{i}\left(s\right)$ are the deformation mode displacement components, and ${\chi}_{i}\left(z\right)$ are their amplitude functions along the beam length.
Fig. 2The elementary deformed shapes on node 3: a) the axial unit displacement, b) the tangential unit displacement, c) the normal unit displacement, and d) the torsional unit displacement
a)
b)
c)
d)
The threedimensional displacement field of the thinwalled structure is expressed with three components ${U}_{1}(s,n,z)$, ${U}_{2}(s,n,z)$ and ${U}_{3}(s,n,z)$. By considering the membrane and flexural behaviors of thin walls, and employing Kirchhoff’s thinplate assumption, the displacement field $\mathbf{D}$ can be obtained as:
Substitute Eq. (1) into Eq. (2), and the threedimensional displacement field can be rewritten in a onedimensional form by involving a transformation matrix $\mathbf{H}$ as:
In the case of ignoring defects and material uncertainties under the premise of small stress and by employing the Saint VenantKirchhoff material law, the strain field $\mathbf{\epsilon}=\left[{\mathbf{\epsilon}}_{zz}\right(s,n,z),{\mathbf{\epsilon}}_{ss}(s,n,z),{\mathbf{\gamma}}_{zs}(s,n,z){]}^{{\rm T}}$ and stress field $\mathbf{\sigma}=\left[{\mathbf{\sigma}}_{zz}\right(s,n,z),{\mathbf{\sigma}}_{ss}(s,n,z),{\mathbf{\tau}}_{zs}(s,n,z){]}^{{\rm T}}$ can be obtained as:
where $\mathbf{C}$ and $\mathbf{E}$ are the compatibility operator and the constitutive matrix, respectively, $E$ and $v$ are the material Young’s modulus and Poisson’s ratio, respectively.
2.2. Governing equations
The energy of the thinwalled beam includes strain energy $U$, potential energy $W$ and kinetic energy $T$. In the absence of dissipative forces, the dynamical modeling of thinwalled structures involves the application of the Hamiltonian principle, reading:
where $H$ is Hamiltonian, ${t}_{1}$ and ${t}_{2}$ are boundary times. The strain energy $U$, potential energy $W$ and kinetic energy $T$ are given by:
where $V$ and $\rho $ are the beam volume and the material density, respectively, $\mathbf{p}$ is the load component (including axial, tangential and normal components). The governing equation for the thinwalled structure is obtained by substituting Eqs. (3)(5) and Eq. (7) into Eq. (6) as:
where $A$ and $L$ are the crosssection area and the beam length, respectively.
To facilitate the calculation, the finite element method is used to axially discretize $\mathbf{\chi}$, reading:
where $\mathbf{N}$ and $\mathbf{X}$ are the linear interpolation function and the node generalized displacement vector, respectively.
Substituting Eq. (9) into Eq. (8), the governing equation is reformulated as:
3. Identification of crosssection deformation modes
This section presents the concepts and procedures involved in the approach to perform crosssection deformation modes recognition, including the preprocessing of the data and the subsequent presentation of the recognition algorithms. In addition, a reduced higherorder model is proposed for practical applications.
3.1. Preparing crosssection deformation data
The eigenvectors of the higherorder model are the basis for recognizing deformation modes, which makes it necessary to extract and process the vibrational parameters prior to mode recognition. The generalized eigenvectors and generalized eigenvalues are obtained by solving Eq. (6) using the finite element method, and the generalized eigenvector matrix $\mathbf{\Gamma}$ contains all information about the deformation of the circular thinwalled structure. The data are preprocessed to obtain the characteristic deformation of thinwalled sections. By definition, $\mathbf{\Gamma}$ is given by:
where ${N}_{0}$ is the number of interpolation nodes along the axial direction of the thinwalled cylindrical beam, ${\mathbf{\Gamma}}_{k}$ is the $k$th order generalized eigenvector and each eigenvector is a combination of the amplitude functions. Thus, ${\mathbf{\Gamma}}_{k}$ can be given by:
where ${D}_{n}$ is the $n$th generalized eigenvalue.
The first ${N}_{md}$ pattern vectors are selected to compose a generalized feature vector matrix ${\stackrel{}{\mathbf{\Gamma}}}_{1}$ for pattern recognition, and matrix ${\stackrel{}{\mathbf{\Gamma}}}_{1}$ can be obtained by:
Reintegrate ${\stackrel{}{\mathbf{\Gamma}}}^{\left(n\right)}$ into the form of ${\stackrel{}{\mathbf{\Gamma}}}_{\mathrm{\Delta}}^{\left(n\right)}$, which can be given by:
Both outofplane and inplane deformation modes are considered at any node $j$ of the $n$th order modal vector. Thus, ${\mathbf{\Gamma}}_{j}^{\left(n\right)}\left(z\right)$ can be expressed as:
where ${\mathbf{\Gamma}}_{j\left(\mathrm{o}\mathrm{u}\mathrm{t}\right)}$ and ${\mathbf{\Gamma}}_{j\left(\mathrm{i}\mathrm{n}\right)}$ refer to the weight vectors corresponding to the outofplane and inplane basis functions of node $j$, respectively.
Converting the first ${N}_{md}$ mode vectors into a matrix $\stackrel{~}{\mathbf{\Gamma}}$. And each column of the matrix $\stackrel{~}{\mathbf{\Gamma}}$ represents a set of generalized displacements in the crosssection. By definition, $\stackrel{~}{\mathbf{\Gamma}}$ is given by:
where ${f}_{n}$ is the intrinsic frequency of the nth order mode.
In order to facilitate the elimination of the interference of the extracted deformation patterns for subsequent recognition, $\stackrel{~}{\mathbf{\Gamma}}$ is converted to the following form:
The effect of the extracted deformation pattern ${\mathbf{R}}_{i}$ on subsequent pattern recognition is eliminated by Schmitt orthogonalization and an updated matrix ${\mathbf{\Gamma}}_{\left(i\right)}$ is given by:
where $\mathrm{d}\mathrm{o}\mathrm{t}\left(\right)$ is the inner product of ${\mathbf{\Theta}}_{i}$ and ${\mathbf{R}}_{i}$, ${\mathbf{\Gamma}}_{\left(i\right)}$ denotes the matrix obtained after eliminating the main mode vector ${\mathbf{R}}_{i}$. This equation eliminates the interference of ${\mathbf{R}}_{i}$ with subsequent recognition, and ${\mathbf{R}}_{i}$ consists of six classical shape vectors (Corresponding to ${\alpha}_{1}~{\alpha}_{3}$ in Fig. 5 and ${\beta}_{1}~{\beta}_{3}$ in Fig. 6, respectively).
Next, the generalized eigenvector matrix $\mathbf{\Gamma}$ is to be processed to identify crosssection deformation modes using the principal component analysis.
3.2. Recognition of basic algorithms
In this part, the principal component analysis is used to extract principal deformation patterns in the form of vectors. The elements of these vectors are virtually the coefficients of basis shape functions, which can be further used to mathematically describe crosssection deformation modes. Specifically, the crosssection deformation patterns hidden in the matrix $\mathbf{\Gamma}$ can be recognized through the principal component analysis, with $\mathbf{\Gamma}$ transformed into a low dimensional space and crosssection deformation modes ranked in clear hierarchy. According to the higherorder beam theory, inplane and outofplane deformation modes are independently derived and described. Therefore, it is reasonable to decouple the amplitude matrix $\mathbf{\Gamma}$ into submatrix ${\mathbf{\Gamma}}_{\mathrm{o}\mathrm{u}\mathrm{t}}$ consisting of outofplane nodal displacement values and ${\mathbf{\Gamma}}_{\mathrm{i}\mathrm{n}}$ consisting of inplane nodal displacement values. Thus, $\mathbf{\Gamma}$ can be rewritten as:
The amplitude matrices of the inplane and outofplane basis functions are decentered separately to obtain the new matrices ${\mathbf{A}}_{\mathrm{o}\mathrm{u}\mathrm{t}}$ and ${\mathbf{A}}_{\mathrm{i}\mathrm{n}}$, reading:
where $\left(:,i\right)$ denotes all elements of column $i$ of the matrix, $j$ is the column element number, and $m$ denotes the number of matrix rows.
The covariance matrices ${\mathbf{C}}_{\mathrm{o}\mathrm{u}\mathrm{t}}$ and ${\mathbf{C}}_{\mathrm{i}\mathrm{n}}$ are denoted as:
The eigenvectors are the extracted combined weight vector of basis functions. In this paper, the eigenvalues, used to measure the contribution of the eigenvectors, are arranged from largest to smallest as:
${\sigma}_{in1}^{\left(k\right)}\ge {\sigma}_{in2}^{\left(k\right)}\ge \cdots \ge {\sigma}_{in{r}_{2}}^{\left(k\right)},{r}_{2}=rank\left({\mathit{\Gamma}}_{in}\right).$
3.3. Updating higherorder beam model
The number of crosssection characteristic deformations derived from pattern recognition is proportional to the number of discrete nodes in the crosssection. In engineering operations, too many section feature deformation modes will lead to model complexity and reduce calculational efficiency. According to the actual accuracy requirements of the model, crosssection characteristic deformation modes should be selected in order of priority to reduce the degrees of freedom and construct an improved onedimensional higherorder model. In engineering practice, the greater the accuracy requirement, the greater the number of inplane and outofplane characteristic deformations that need to be added. The original generalized coordinates are linearly superimposed and replaced by the selected crosssection feature deformations to form the new shape function vectors $\widehat{\mathit{\alpha}}$, $\widehat{\mathit{\varphi}}$ and $\widehat{\mathit{\omega}}$, which are given by:
where $g$ and $\eta $ denote the number of outofplane and inplane characteristic deformation modes in the new combination of section characteristic deformations, respectively. $\widehat{\mathit{\alpha}}$, $\widehat{\mathit{\varphi}}$ and $\widehat{\mathit{\omega}}$ are given by:
A new weight vector is introduced to describe the axial displacement variation of the thinwalled structure, $\widehat{\mathit{\chi}}$ is given by:
The finite element implementation adopts the linear Lagrange interpolation and leads to $\widehat{\mathit{\chi}}=\mathit{N}\widehat{\mathit{{\rm X}}}$. As a result, the updated governing equation can be obtained by substituting Eq. (27) into Eq. (10) as:
4. Numerical examples and discussion
The approach of crosssection analysis developed in this work is applied to circular thinwalled structures with different discretization degrees of the crosssection midline. Several illustrative examples are also provided to demonstrate the validity and accuracy of the improved higherorder models in this paper. They concern Convergence analysis, unconstrained structures and cantilevered structures.
4.1. Crosssection comparison analysis
In this section, pattern recognition is performed for circular thinwalled structures with different degrees of discretization in the midline of the crosssection. Specially, a set of equal length linear segments with the number of ${N}_{s}$ is defined to discretize the midline of the circular section. The results for different degrees of discretization of the crosssection midline are shown in Fig. 3. Firstly, ${N}_{s}=\text{8}$ is chosen and the geometric dispersion is gradually increased, and the abovementioned pattern recognition method is used to analyze and compare the section deformations. It can be proved that solutions for regular convex polygonal thinwalled structures gradually approaches those for circular thinwalled structure as the number of walls increases.
Fig. 3Results of different discretization degrees of crosssection midline
Fig. 4Elementary deformed shapes of thinwalled beam whose crosssection midline is discretized into 16 equallength straight segments
Prior to pattern recognition, the deformed shapes of the discrete cross section are described with basis shape functions ${\alpha}_{i}\left(s\right)$, ${\phi}_{i}\left(s\right)$, and ${\omega}_{i}\left(s\right)$. Take the crosssection discretized with ${N}_{\mathrm{s}}=\text{16}$ nodes as shown in Fig. 1(c) as an example. Referring to a previous paper of Zhang [26], by imposing unit displacements on these nodes, it leads to $4\times {N}_{s}=\text{64}$ elementary deformed shapes as shown in Fig. 4. All the deformed shapes are employed to form the basis function set $\mathbf{\alpha}$, $\mathbf{\varphi}$ and $\mathbf{\omega}$ for the preliminary higherorder beam model.
Pattern recognition of the first 60th order modal vectors of the model with different discretization degrees are performed using the method in this paper. The principal components with a cumulative contribution rate greater than 99.9 % are selected to identify the characteristic outofplane deformations of ${N}_{s}$ with different degrees of discretization as shown in Fig. 5 and inplane deformations as shown in Fig. 6. The crosssection deformation modes from ${\alpha}_{1}$ to ${\alpha}_{3}$ in Fig. 5 and ${\beta}_{1}$ to ${\beta}_{3}$ in Fig. 6 correspond to six rigidbody deformation modes in classical beam theory, respectively. Other modes are higherorder characteristic deformation modes of the crosssection, which demonstrate the mechanical properties of warping and distortion unique to thinwalled structures.
Fig. 5Outofplane characteristic deformation modes derived from Ns identification with different discretization degrees
Fig. 6Inplane characteristic deformation modes derived from Ns identification with different discretization degrees
The two graphs show that the number of deformation modes increases with ${N}_{s}$. It is observed that the deformation mode shapes are virtually identical, irrespective of the ${N}_{s}$ value adopted. The only visible difference is that the warping constant is significantly overestimated for coarse meshes but, as expected, a refined geometry discretization can markedly improve the accuracy, and the results tend asymptotically to the solutions for circular tubes.
However, the use of a refined discretization greatly increases the number of preliminary deformation modes and therefore reduces computational efficiency. To overcome this drawback, the characteristic deformations shown in Fig. 5 and Fig. 6 are selected for the construction of an improved onedimensional higherorder model. For lower accuracy requirements, it is feasible to reproduce the classic Timoshenko beam by using only six rigid body deformation modes. For occasions with high accuracy requirements, a certain amount of inplane and out of plane feature deformations can be added to achieve the required accuracy of the model.
4.2. Convergence analysis
In order to grasp the relationship between the accuracy and efficiency of the model, convergence analysis of the model is required to determine a reasonable number of discrete elements. A fixed constraint is applied to one end of the structure shown in Fig. 1 for convergence analysis. The beam model is discretized into different numbers of onedimensional higherorder elements along the axial direction, and as shown in Fig. 7, relative errors of the first ten natural frequencies varied with the number of elements employed. The convergence data are obtained with 120 finite elements. It can be seen that at least 60 elements need to be discretized along the axial direction for the results of the beam model to converge.
Fig. 7Convergence of the first ten natural frequencies of the cantilevered thinwalled structure, varying with the number of employed finite elements
4.3. Grid independence verification
Since the accuracy of ANSYS shell theory calculations is affected by several factors, meshing stands out as a significant determinant. Therefore, the results obtained from ANSYS shell theory calculations need to be verified for mesh independence as necessary before comparison with the model presented in this paper. It should be noted that the thinwalled beams studied in this paper do not exhibit complex structures and have good consistency in the axial direction. Hence, it is only essential to keep the boundary conditions and loads unchanged, and to analyze the relationship between the sparseness of the mesh and the calculation results by gradually refining the mesh. The initial number of grids is set to 120 based on the model shown in Fig. 1(a), and then the grid elements are incrementally increased to observe the trend of the numerical solution. As shown in Fig. 8, it can be seen that at least 3840 grid elements need to be set for the result of the modal natural frequency no longer changing significantly.
4.4. Discretized error analysis
In order to verify the accuracy of the improved model in this paper relative to the circular thinwalled structure, the onedimensional higherorder initial model and the reduceddimensional model are applied to the free vibration analysis of the thinwalled structure. And the two sets of natural frequencies calculated are compared with those of circular thinwalled beams calculated by ANSYS shell theory.
Fig. 8Modal natural frequency varies with the number of employed grids
The models with different degrees of crosssection discretization are calculated by the finite element method, and the linear interpolation method is used to discretize the axial direction into 80 onedimensional higherorder initial elements. The results are compared with the ANSYS shell model, which consists of 7680 shell elements, distributed as 120 along the length and 64 over the crosssection. Fig. 9 presents the relative errors about the natural frequencies of the first 20 modes. It should be pointed out that the relative errors are calculated based on the assumption that the results derived from ANSYS shell theory are accurate enough.
Fig. 9Comparison of the relative errors of the first 20 orders of natural frequencies of circular thinwalled structures with different degrees of discretization
It is worth noting that the models with crosssection discretization of ${N}_{s}=\text{12}$ and above in Fig. 9 can obtain an equally modeling accuracy as that of 7680 twodimensional shell units. The relative error is kept within 4 %. It is indicated that a regular convex polygonal thinwalled structure with crosssection discretization of more than 12 equallength walls would be able to characterize well the variation of vibration patterns of circular thinwalled structures.
In addition, the improved higherorder model has an overall improvement in accuracy compared to the initial onedimensional higherorder model. This is because the stiffness of the initial onedimensional higherorder model reduces as the number of degrees of freedom considered increases, resulting in a lower natural frequency of the model. It is observed that the improved onedimensional higherorder model not only further reduces the degrees of freedom, but also significantly improves computational efficiency. Accordingly, the calculational results of natural frequencies are more accurate as the stiffness of the model is increased to a certain extent. This data indicate that the improved onedimensional higherorder model of this paper has greater potential.
4.5. Analysis of cantilever thinwalled structure
In order to verify that the theory of this paper is also applicable in modeling a thinwalled structure with different displacement constraints, numerical example is carried out on the cantilevered structure proposed in Section 4.2. Table 1 presents the natural frequencies of the first ten modes derived from the present model and the ANSYS shell theory, and their relative differences. Similarly, the results of the present model are obtained with 80 elements uniformly distributed along the axial direction. The ANSYS model consists of 7680 shell elements, 120 distributed along the length and 64 divided along the cross section.
Table 1Comparison of the first ten natural frequencies of the cantilevered thinwalled structure
Mode number  Present model  ANSYS shell  Relative error 
${f}_{i}$_{}(Hz)  ${f}_{Ai}$_{}(Hz)  ${\delta}_{i}$ (%)  
1st  127.85  132.58  –3.57 
2nd  127.85  132.58  –3.57 
3rd  640.74  652.29  –1.77 
4th  655.68  652.57  0.48 
5th  655.68  652.57  0.48 
6th  667.72  675.80  –1.20 
7th  667.72  675.80  –1.20 
8th  725.91  720.57  0.74 
9th  725.91  720.57  0.74 
10th  959.45  979.60  –2.06 
As expected, the results in Table 1 show that the natural frequencies obtained with present model are very close to those from the ANSYS shell theory, with relative differences smaller than 4 % for the studied model. These facts prove that the present model could accurately reproduce threedimensional behaviors of thinwalled structure.
4.6. Analysis of unconstrained thinwalled structure
An unconstrained thinwalled beam with circular crosssection shown in Fig. 1 (c) is considered for dynamic analyses so as to verify the performance of the proposed higherorder models. Related parameters of this thinwalled beam include length $l=\text{12}$ m, width of each arm $a=\text{0.04}$ m, thickness $t=\text{0.01}$ m, density $\rho =\text{7850}$ kg/m^{3}, modulus of elasticity $E=$2×10^{11} Pa, Poisson’s ratio $v=\text{0.3}$.
The free vibration shapes of the thinwalled structures of the improved onedimensional higherorder model and ANSYS shell model are calculated, respectively. The comparisons of the 7th to 16th order free vibration shapes are shown in Fig. 10, and the results are divided into 10 pairs according to the order of vibration modes. In each pair, the left is derived from a modified onedimensional higherorder model while the right represents the ANSYS shell models. There is no significant difference between the 10 pairs of vibration modes, which proves the excellent prediction capability of the improved onedimensional higherorder model in this paper for the threedimensional vibration modes of circular thinwalled structures.
To further illustrate the applicability of the improved onedimensional higherorder model in this paper, the thinwalled structure shown in Fig. 1 is used for numerical analysis of circular thinwalled structures with different slenderness ratios. Fig. 11 presents the relative errors of natural frequencies for thinwalled structures with a slenderness ratio ranging from 3 to 10 based on the shell model. It is observed that the calculational accuracy of the natural frequencies of thinwalled structures improves as the slenderness ratio increases. The relative error of natural frequencies is kept within 2.8 % despite the slenderness ratio is 3. It indicates that the improved onedimensional higherorder model in this paper can be applied to the dynamic modeling of circular crosssection thinwalled structures with a slenderness ratio of more than 3, and has a wider application range than classical beam theories.
Fig. 10Comparison of free vibration modes between improved onedimensional higherorder model and ANSYS shell model
Fig. 11Comparison of the first 20 natural frequencies of thinwalled structures with different slenderness ratios between the improved onedimensional higherorder model and the ANSYS shell model
5. Conclusions
In this paper, the dynamics model of thinwalled structure with circular crosssection is constructed based on onedimensional higherorder beam theory, and the crosssection characteristic deformation modes are extracted on this basis using the principal component analysis. With a compact set of crosssection deformation modes employed, the preliminary onedimensional model was updated to the refined beam model of high precision and efficiency. A simple procedure is proposed to implement the parameterization of the degree of discretization and it is based on the discretization of a set of equallength linear segments in the crosssection midline, which avoids the difficulties related to GBT formulations for genuinely curved sections. The basic idea is to obtain the deformation modes by increasing the discretization degree to approximate the crosssection geometry. The following conclusions and suggestions have been drawn.
1) The characteristic deformation modes of circular thinwalled structures can be efficiently recognized and extracted by using the principal component analysis.
2) A refined discretization makes it possible to accurately describe the threedimensional dynamical properties of circular thinwalled structures.
3) To ensure the accuracy of the established circular thinwalled dynamics model, at least 60 onedimensional highorder elements should be utilized in order to achieve good convergence.
4) The refined higherorder beam model is valid for thinwalled beams with various boundary conditions and slenderness ratios, justifying the applicability of the proposed approach.
5) The refined higherorder beam model is able to accurately predict dynamic behaviors of thinwalled beams with much higher computation efficiency comparing with plate/shell theory.
References

M. RezaieePajand, A. R. Masoodi, and E. Arabi, “Improved shell element for geometrically nonlinear analysis of thinwalled structures,” Proceedings of the Institution of Civil Engineers – Structures and Buildings, Vol. 175, No. 4, pp. 347–356, Apr. 2022, https://doi.org/10.1680/jstbu.19.00130

Q. Wu, H. Gao, Y. Zhang, and L. Chen, “Dynamical analysis of a thinwalled rectangular plate with preload force,” Journal of Vibroengineering, Vol. 19, No. 8, pp. 5735–5745, Dec. 2017, https://doi.org/10.21595/jve.2017.18330

H. Kim and G.W. Jang, “Higherorder thinwalled beam analysis for axially varying generally shaped cross sections with straight crosssection edges,” Computers and Structures, Vol. 189, pp. 83–100, Sep. 2017, https://doi.org/10.1016/j.compstruc.2017.04.015

V. Adámek, “The limits of Timoshenko beam theory applied to impact problems of layered beams,” International Journal of Mechanical Sciences, Vol. 145, pp. 128–137, Sep. 2018, https://doi.org/10.1016/j.ijmecsci.2018.07.001

Y. Kim and Y. Y. Kim, “Analysis of thinwalled curved box beam under inplane flexure,” International Journal of Solids and Structures, Vol. 40, No. 22, pp. 6111–6123, Nov. 2003, https://doi.org/10.1016/s00207683(03)003676

K. Yoon, P.S. Lee, and D.N. Kim, “An efficient warping model for elastoplastic torsional analysis of composite beams,” Composite Structures, Vol. 178, pp. 37–49, Oct. 2017, https://doi.org/10.1016/j.compstruct.2017.07.041

D. Shin, S. Choi, G.W. Jang, and Y. Y. Kim, “Higherorder beam theory for static and vibration analysis of composite thinwalled box beam,” Composite Structures, Vol. 206, pp. 140–154, Dec. 2018, https://doi.org/10.1016/j.compstruct.2018.08.016

T.T. Nguyen, N.L. Nguyen, J. Lee, and Q.H. Nguyen, “Vibration analysis of thinwalled functionally graded sandwich beams with nonuniform polygonal crosssections,” Composite Structures, Vol. 278, p. 114723, Dec. 2021, https://doi.org/10.1016/j.compstruct.2021.114723

A. K. Habtemariam, C. Könke, V. Zabel, and M. J. Bianco, “Generalized beam theory formulation for thinwalled pipes with circular axis,” ThinWalled Structures, Vol. 159, p. 107243, Feb. 2021, https://doi.org/10.1016/j.tws.2020.107243

A. K. Habtemariam, F. Tartaglione, V. Zabel, C. Könke, and M. J. Bianco, “Vibration analysis of thinwalled pipes with circular axis using the generalized beam theory,” ThinWalled Structures, Vol. 163, p. 107628, Jun. 2021, https://doi.org/10.1016/j.tws.2021.107628

R. Schardt, “Lateral torsional and distortional buckling of channel – and hatsections,” Journal of Constructional Steel Research, Vol. 31, No. 23, pp. 243–265, Jan. 1994, https://doi.org/10.1016/0143974x(94)900124

J. M. Davies and P. Leach, “Firstorder generalised beam theory,” Journal of Constructional Steel Research, Vol. 31, No. 23, pp. 187–220, Jan. 1994, https://doi.org/10.1016/0143974x(94)900108

N. Silvestre, D. Camotim, and N. F. Silva, “Generalized beam theory revisited: from the kinematical assumptions to the deformation mode determination,” International Journal of Structural Stability and Dynamics, Vol. 11, No. 5, pp. 969–997, Nov. 2011, https://doi.org/10.1142/s0219455411004427

R. Gonçalves, M. RittoCorrêa, and D. Camotim, “A new approach to the calculation of crosssection deformation modes in the framework of generalized beam theory,” Computational Mechanics, Vol. 46, No. 5, pp. 759–781, Jul. 2010, https://doi.org/10.1007/s0046601005122

A. D. Martins, D. Camotim, R. Gonçalves, and P. B. Dinis, “Enhanced geometrically nonlinear generalized beam theory formulation: derivation, numerical implementation, and illustration,” Journal of Engineering Mechanics, Vol. 144, No. 6, p. 04018, Jun. 2018, https://doi.org/10.1061/(asce)em.19437889.0001457

R. Bebiano, C. Basaglia, D. Camotim, and R. Gonçalves, “GBT buckling analysis of generally loaded thinwalled members with arbitrary flatwalled crosssections,” ThinWalled Structures, Vol. 123, pp. 11–24, Feb. 2018, https://doi.org/10.1016/j.tws.2017.10.045

A. W. Ruggerini, A. Madeo, R. Gonçalves, D. Camotim, F. Ubertini, and S. de Miranda, “GBT postbuckling analysis based on the implicit corotational method,” International Journal of Solids and Structures, Vol. 163, pp. 40–60, May 2019, https://doi.org/10.1016/j.ijsolstr.2018.12.011

R. Bebiano, R. Calçada, D. Camotim, and N. Silvestre, “Dynamic analysis of highspeed railway bridge decks using generalised beam theory,” ThinWalled Structures, Vol. 114, pp. 22–31, May 2017, https://doi.org/10.1016/j.tws.2017.01.027

N. Silvestre, “Generalised beam theory to analyse the buckling behaviour of circular cylindrical shells and tubes,” ThinWalled Structures, Vol. 45, No. 2, pp. 185–198, Feb. 2007, https://doi.org/10.1016/j.tws.2007.02.001

C. Basaglia, D. Camotim, and N. Silvestre, “GBTbased buckling analysis of steel cylindrical shells under combinations of compression and external pressure,” ThinWalled Structures, Vol. 144, p. 106274, Nov. 2019, https://doi.org/10.1016/j.tws.2019.106274

N. Peres, R. Gonçalves, and D. Camotim, “Generalised beam theory formulation for the buckling analysis of thinwalled members with circular axis,” ThinWalled Structures, Vol. 176, p. 109322, Jul. 2022, https://doi.org/10.1016/j.tws.2022.109322

S. de Miranda, A. Gutiérrez, R. Miletta, and F. Ubertini, “A generalized beam theory with shear deformation,” ThinWalled Structures, Vol. 67, pp. 88–100, Jun. 2013, https://doi.org/10.1016/j.tws.2013.02.012

R. Gonçalves, R. Bebiano, and D. Camotim, “On the shear deformation modes in the framework of generalized beam theory,” ThinWalled Structures, Vol. 84, pp. 325–334, Nov. 2014, https://doi.org/10.1016/j.tws.2014.07.012

A.A. Muresan, M. Nedelcu, and R. Gonçalves, “GBTbased FE formulation to analyse the buckling behaviour of isotropic conical shells with circular crosssection,” ThinWalled Structures, Vol. 134, pp. 84–101, Jan. 2019, https://doi.org/10.1016/j.tws.2018.07.032

A. Sahraei, P. Pezeshky, S. Sasibut, F. Rong, and M. Mohareb, “Finite element formulation for the dynamic analysis of shear deformable thinwalled beams,” ThinWalled Structures, Vol. 173, p. 108989, Apr. 2022, https://doi.org/10.1016/j.tws.2022.108989

L. Zhang, W. Zhu, and A. Ji, “Application of pattern recognition to the identification of crosssection deformation modes of thinwalled structures,” IEEE Access, Vol. 7, pp. 169586–169598, Jan. 2019, https://doi.org/10.1109/access.2019.2954712
About this article
This work was supported by the National Key Research and Development Program (Grant No. 2022YFB4703401).
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.
Tao Zeng: writing–original draft preparation and visualization. Lei Zhang: methodology and writingreview and editing. Yuhang Zhu: formal analysis and investigation.
The authors declare that they have no conflict of interest.