COMPUTATIONAL MECHANICS New Trends and Applications S. Idelshon, E. Oate and E. Dvorkin (Eds.

) CIMNE, Barcelona, Spain 1998

ANALYSIS OF THE COLLAPSE OF STEEL TUBES UNDER EXTERNAL PRESSURE

Andrea P. Assanelli*, Rita G. Toscano** and Eduardo N. Dvorkin***
Center for Industrial Research, FUDETEC Av. Crdoba 320 1054 Buenos Aires, Argentina * e-mail: andrea@fudfem.org.ar ** e-mail: rita@fudfem.org.ar *** e-mail: dvk@fudfem.org.ar

Politcnica de.unl.edu.ar/gtmKey words: Finite elements, Collapse, Tubes, Elasto-plasticity. Abstract. The collapse of steel tubes under external pressure is described using finite element models. The models are validated comparing their predictions against the results of collapse tests performed in our laboratory.

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

INTRODUCTION

The external collapse pressure of very thin steel tubes is governed by classical elastic buckling formulas1,2; however, for thicker tubes more involved elasto-plastic considerations have to be taken into account. There are many factors that have some degree of influence on the external pressure that produce the collapse of a steel tube, among them3-9: 1. The relation (outside diameter/thickness) (D/t ratio). 2. The yield stress of the tube steel (y). 3. The work - hardening of the tube steel. 4. The shape of the tube sections (outside diameter shape and thickness distribution). 5. The residual stresses locked in the tube steel. 6. The localized imperfections introduced either in the tubes production, in the tubes handling or due to localized wear. In this paper we discuss the experimental techniques and the hierarchy of finite element models that we have developed to study the collapse behavior of seamless steel tubes. In Section 2 we comment on our facilities for performing external pressure collapse tests and we also present an imperfections measuring system that we have implemented at our laboratory based on the device described by Yeh et al10 and Arbocz et al11,12. In Section 3 we describe some simple 2D finite element models that we have developed as a first approach for the simulation of the external pressure collapse test. We compare the results that these 2D models provide with the experimental results obtained at our laboratory; the comparison demonstrates that a 2D geometrical characterization of the tubes does not contain enough information to assess on their collapse strength. The tube shape imperfections and the wall thickness normally change along the tube length; also localized imperfections can be found in the tubes. In Section 4 we present the 3D finite element models that we have developed to overcome the limitations of the simpler 2D models. The 3D model predictions have been qualified via experimental testing at our laboratory. The geometrical information on the tested samples has been acquired using our imperfections measuring system and has been used as input data for our 3D finite element models. 2 EXPERIMENTAL PROCEDURES In the present Section we describe our laboratory facilities for performing external pressure collapse tests and we also present the imperfections measuring system that we have developed based on the previous work of Yeh et al10 and Arbocz et al11,12. 2.1 External pressure collapse tests A schematic representation of our laboratory assembly for performing external pressure collapse tests is presented in Fig. 1.

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

Packer

Free to slide

Chamber length > 10 D

Figure 1. Collapse chamber

It is important to notice that: 1. The experimental set-up does not impose any axial restrain on the tubes. 2. The length of the sample being tested fulfills in all cases the relation: L/D > 10. 3. The collapse of the samples is detected by a drop in the pressure of the water used to pressurize them. 2.2 Imperfections measuring system In order to acquire information on the geometry of the tube samples that are going to be tested in the collapse chamber, the shape of the external surface is mapped and the wall thickness is measured at a number of points. 2.2.1. Mapping of the samples external surface To map the external surface of a sample we rotate it in a lathe and an LVDT (Linear Variable Displacement Transducer) placed on the lathe cross-slide (see Fig. 2) measures, at regular intervals of axial and angular position, the distance ri shown in Fig. 3.

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

We have developed the following algorithm to transform the data acquired with the LVDT into the Fourier series description of the samples external surface.

Figure 2. Mapping of the samples external surface

Figure 3. Algorithm to process the data acquired with the LVDT

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

Data ri: radial distance from the rotation axis to the external surface (see Fig. 3). qi: total turns corresponding to the i-th section, measured from an arbitrary defined zero. Algorithm 1. We can define:
i = 2 q i INT (q i )

Note: the data is acquired along a spiral path path (with a pitch less than half of the wall thickness); however, in subsequent analyses we will consider that the points corresponding to a turn are located on a planar section, at an axial distance x K from an arbitrary origin.
x K = x INT ( qi )

2. For the k-th section we can define a "best-fit circle", with radius Ro and with its center located at (x o yo) in a Cartesian system, contained in the section plane and with its origin at the section rotation center10. For determining Ro, x o and yo we solve the following minimization problem:

( Ro ,xo , yo ) = arg
i

[min E 2 (R , x , y )]
o o o

E 2 = ri g ( i , Ro ,xo , yo )

]2
2

2 g ( i , Ro , xo , yo ) = [ xo cos i + yo sin i ] + Ro ( xo sin i yo cos i )

To solve the above nonlinear minimization problem we apply the Levemberg-Marquard method13, using as first trial a simplified solution in which the term in (x o yo) inside the square root in the expression for g, is neglected14. 3. Once the center of the "best-fit circle" is determined we reduce the acquired data to it,
$ xi = ri cos i xo $ yi = ri sin i yo $ ri = $ $ ( xi )2 + ( yi )2

y $ $ i = tan 1 i $ xi

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

$ We expand r ( ) using a discrete Fourier transform $ r ( ) = Ro +

j =1

$ [ a$ cos ( j$ )+ b$ sin ( j) ]
j j

where,
$ aj = 1
k =1 M

$ $ r$k cos ( j k ) k
M

] ]

1 $ bj =

k =1

$ $ r$k sin ( jk ) k

In Figs. 4 we present the Fourier decomposition obtained using the above described algorithm in the case of a tube with a nominal outside diameter of 9 5/8" and a metric weight of 47 lb/ft. The length of each sample is approx. 2.5 mts, the pitch of the cross-slide is 3mm/turn and the angular distance between the acquired points is one degree.

Figure 4. Fourier decomposition of the outside radius

2.2.2. Wall thickness measurement We measure the wall thickness at a number of points evenly distributed on the sample external surface, using manual ultrasonic gages; the obtained thickness map is represented in Fig. 5.

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

Thickness [mm]

A...I:: axial locations 1,...8 : angular locations Figure 5. Thickness map

3. TWO DIMENSIONAL FINITE ELEMENT MODELS In this Section we discuss the 2D finite element models that we have implemented to simulate the behavior of ideal "long specimens" in the external pressure collapse test. We also compare experimental results with the predictions of these 2D models: the comparison demonstrates that a 2D geometrical characterization of the tubes does not contain enough information to assess on their collapse strength. However, the 2D models can provide useful information on the trends of the external collapse pressure value when some parameters -such as residual stresses, shape imperfections, etc.- are changed. 3.1. Formulation of the 2D models We develop the 2D finite element models using a total Lagrangian formulation15 that incorporates: 1. Geometrical nonlinearities due to large displacements/rotations (infinitesimal strains assumption); 2. Material nonlinearity: an elasto-plastic constitutive relation is used for modeling the steel mechanical behavior (Von Mises associated plasticity16).

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

We carry out the finite element analyses using a special version of the general purpose finite element code ADINA that incorporates the quadrilateral QMITC element17-19. For modeling the external hydrostatic pressure we use follower loads2, and we introduce in our models the residual stresses via temperature gradients through the thickness. In order to validate our procedure for simulating the residual stresses, in Fig. 6 we present the result of the 2D finite element simulation of a slit-ring test; which is the test that we use at our laboratory to measure the residual stresses in the tube samples. The model results, that we present in Fig. 6, agree with the experimental observations; hence,we can consider that our residual stresses simulation is realistic enough.

Figure 6. Two dimensional simulation of the slit-ring test

It is important to recognize that the actual collapse test is not modeled exactly neither by plane strain nor by plane stress models: 1. The absence of longitudinal restrains imposes a plane stress situation at the sample edges. 2. The length of the samples (L/D > 10) approximates a plane strain situation at its center. In this Section, in order to explore the limitations of the 2D models, we analyze the collapse test using both, plane stress and plane strain finite element models. 3.2. Two dimensional finite element results vs. experimental results The 2D finite element models have been developed considering an elastic-perfectly plastic

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

material constitutive relation (we will show in Section 3.3 that disregarding the steel work-hardening introduces only a negligible error in the calculated collapse pressures). For standard samples in which the tube geometry is not analyzed using the above described imperfections measuring system, the laboratory keeps records of the following geometrical parameters: 1. Average, maximum and minimum outside diameter at three sections (the central section and the two end sections) 2. Thicknesses at the two end sections (eight points per section). We construct the geometry of the 2D finite element models using: 1. The ovality [Ov = (Dmax - Dmin)/Daverage] and the average outside diameter of the central section. 2. The eccentricity ([ = [(tmax -tmin)/taverage] and t is the thickness]) obtained by averaging the eccentricities of the two end sections and the average thickness obtained by averaging the sixteen thickness determinations. The residual stresses for each sample are measured using a slit-ring test and its values are imposed on the model as discussed above. The actual transversal yield stress of the sample material is measured and its value is used for the elastic-perfectly plastic material constitutive relation. To determine the collapse loads of the sample models we calculate the nonlinear load/displacement path and seek for its horizontal tangent. To perform the numerical analyses we use, for half the section, a 2D mesh with 720 QMITC elements and 1572 d.o.f.(see Fig. 7.a). To assess on the quality of this mesh we analyze the plane strain collapse of an infinite tube and we compare our numerical results against the analytical results obtained using the formulas in Timoshenko et al1. Average OD Average thickness D/t Ov y pcr as per [1] / pcr FEA 245.42 mm 12.61 mm 19.47 0.18% 9079 kg/cm2 0.992

Table 1. Qualification of the 2D finite element model

From the results in Table 1 we conclude that the proposed 2D mesh of QMITC elements is accurate enough to represent the collapse of very long specimens. From the analyses of different cases we have identified two basic types of load/displacement paths: 1. In Fig. 7.b we present the load/displacement path of an eccentric tube with (D/t =22.13). 2. In Fig. 7.c we present the load/displacement path of another eccentric tube but with (D/t = 17.67). This second case presents an "inverse collapse behavior" that was previously

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

described for the case of collapse under external pressure and bending20,21.

a) Finite element mesh (720 QMITC elements - 1572 d.o.f.)

Displacement [cm] b) Eccentric tube: direct collapse behavior

10

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

Displacement [cm] c) Eccentric tube: inverse collapse behavior Figure 7. 2D models of the external pressure test

A total of 32 collapse tests, for tubes 9 5/8"x47lb/ft and 7"x26lb/ft both Grade 95, have been analyzed using plane stress and plane strain models. The comparisons between the numerical and experimental results are plotted in Figs. 8.a (plane stress models) and Fig. 8.b (plane strain models). It is important to observe that the collapse pressures determined using the 2D models present a significant error. Also, for most cases the numerical determined collapse pressure values are lower than the experimental ones. Some reasons for this behavior are: 1. The middle section ovality is not fully representative of the sample geometry. The numerical/experimental technique to be presented in Section 4, that combines our imperfections measuring system and 3D finite element models, will remove this source of disagreement between numerical and experimental results. 2. In developing the 2D models the measured ovality is entirely assigned to the first elastic buckling mode and this conservative approach partially accounts for the fact that the numerical values are lower than the actual ones. Again, the 3D finite element models will remove this source of disagreement between numerical and experimental results because they will incorporate the actual tube geometry. 3. The experimental set-up (see Fig. 1) imposes on the samples unilateral radial restraints at both ends, these restraints are not described by the 2D models; the difference between the numerical and actual boundary conditions also partially accounts for the fact that the numerical values are in general lower than the actual ones. The 3D models will also remove this disagreement incorporating those end restraints to the external pressure collapse test simulation.

11

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

FEA collapse pressure/experimental collapse pressure

a) 9 5/8" 47lbs/ft and 7" 26lbs/ft Grade 95 - Plane stress

FEA collapse pressure/experimental collapse pressure

b) 9 5/8" 47lbs/ft and 7" 26lbs/ft Grade 95 - Plane strain Figure 8. 2D models vs. experimental results

It is obvious, from the 2D results in Fig. 8, that the geometrical characterization of only one section of a tube is not enough for assessing on its collapse performance. However, the 2D models are very useful for performing parametric studies on the relative weight of different factors that affect the external collapse pressure of tubes.

12

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

3.3. Strain hardening effect We have previously assessed that the strain hardening of the tube material does not play an important role in the determination of its external collapse pressure. In this Section we examine the above assessment using 2D finite element models. We analyze two 9 5/8" tubes (a thin and a thick one) using a bilinear material model; in order to explore different hardenings we consider three values for the constant tangential modulus: ET = 0.0 (perfect plasticity); 0.057E and 0.10 E. where E is the Young modulus. We also consider two ovality (Ov) values but we do not include in this analysis neither the tube eccentricities nor its residual stresses. We summarize the numerical results of our analyses in Table 2: D/t 17.66 17.66 24.37 24.37 Ov 0.75% 0.35% 0.75% 0.35% ET = 0.0 611 684 272 296 ET = 0.057E 614 686 272 297 ET = 0.10E 616 688 272 297

Table 2. Analysis of the work-hardening effect on the external collapse pressure

As it can be seen from the above table, for all the analyzed cases, the strain hardening has a negligible effect on the external collapse pressure value. It is important to recognize that we have considered y to be independent of ET. This is not the case if as "yield stress" we adopt the one corresponding to a relatively large permanent offset. 3.4. Effect of ovality, eccentricity and residual stresses Using 2D finite element models we develop in this Section a parametric study aimed at the analysis of the effect, on the tube collapse pressure, of the following parameters: 1. Ovality (Ov); 2. Eccentricity () 3. Residual stresses (R) In the present analyses the ovality is considered to be concentrated in the shape corresponding to the first elastic buckling mode and the eccentricity is modeled considering non-coincident OD and ID centers. In Fig. 9 we plot the results of our parametric study, normalized with the collapse pressure calculated according to API standards. It is obvious from these results that the main influence on the external collapse pressure comes from the ovality and from the residual stresses; however, the effect of the residual stresses diminishes when the ratio (D/t) evolves from the plastic collapse range to the elastic collapse range. The eccentricity effect, in the case of the external collapse test with neither

13

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

axial nor bending loads, is minor.

Figure 9. Parametric study of the effect of out-of-roundness, eccentricity and residual stresses on the casing collapse pressure. Performed using 2D finite element models

4. THREE DIMENSIONAL FINITE ELEMENT MODELS In this Section we discuss the 3D finite element models that we have implemented to simulate the external pressure collapse test. We use the 3D models with the geometrical data acquired with our imperfections measuring system, described in Section 2, and we compare the numerical collapse pressure predictions with the experimental determinations carried out at our laboratory.

14

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

4.1. Formulation of the 3D models The (radius/thickness) relation of the tubes to be analyzed ranges from 7.61 to 12.08, hence a shell model that incorporates shear deformations is a suitable one. Following what we did for the 2D models we use a total Lagrangian formulation15 that incorporates the geometrical nonlinearities coming from the large displacements/rotations and the material nonlinearity coming from the elastoplastic constitutive relation. We carry out the finite element analyses using the code ADINA and the MITC4 shell element22-24. As in the previous Section, follower loads are used to model the hydrostatic external pressure and the residual stresses are modeled via thermal gradients through the thickness. To check the capability of our 3D finite element models to simulate different residual stress patterns we consider a 9 5/8" x 47 lb/ft tube with y = 77.3 kg/mm2 and R = 0.2 y. We calculate the value of the samples opening in a slit-ring-test: 1. Using the analytical relation between residual stresses and sample openings that we presente in Table 3.
Sample L=3D (approx. plane strain) R

at E 4 R 2 1 v2

L=1" (approx. plane stress)

a t E 4 R 2

Table 3. Post-processing formulas for the slit-ring test

2. Using the finite element model shown in Fig. 10, where the residual stresses are simulated via thermal gradients through the thickness and the slit of the cylindrical sample is simulated by removing a row of elements. The results are:
Sample length 1" 3D

FEA predicted opening Analytical predicted opening

1.02 0.99

Table 4. Validation of the residual stresses simulation using 3D finite element models

Hence, our 3D finite element procedure for representing the residual stresses can be considered as realistic enough. 4.2. Three dimensional simulation of the external pressure collapse test In order to validate our 3D finite element models of the external pressure collapse test, we

15

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

analyze three tubular samples. The meshes are generated using only the first 12 modes (see Fig. 11) in the external surface Fourier decomposition.

a) Lateral view

b) Front view Figure 10. Three dimensional finite element simulation of the slit-ring test

16

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

a) Lateral view Imperfections amplified 50X

b) Front view Figure 11. Finite element mesh developed considering the first 12 modes (sample #1)

The actual transversal yield stress of the samples and the residual stresses measured using "long" slit-ring samples are shown in Table 5: Sample 1 y [Kg/mm2] 86.15 R/y 0.17

17

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

2 3

82.24 84.7

0.38 0.27

Table 5. Mechanical properties of the three analyzed samples

The comparison between the FEA determined collapse pressures and the experimentally determined ones is presented in Table 6: Sample 1 2 3 pcr FEA / pcr exp. 1.005 1.025 0.984

Table 6. Comparison between 3D FEA and experimentally determined collapse pressures

From the above results it is evident that the developed 3D finite element models simulate very accurately the external pressure collapse test. 4.3. Effect of low and high shape modes The numerically calculated values of external collapse pressure that we report in Table 6 have been calculated considering 12 shape modes in the external surface Fourier decomposition (see Fig. 4). In order to investigate the effect of the different shape modes we re-calculate the collapse pressures but considering only the first 4 modes. The obtained numerical results indicate that for the 3 analyzed cases:
4 12 p crmod es = p cr mod es

Hence, we can assess that the external pressure collapse load is determined only by the first modes of the external surface Fourier decomposition. 5. CONCLUSIONS Experimental/numerical procedures for analyzing the behavior of tube samples in the external pressure collapse test were discussed. It was shown that 2D finite element models do not contain enough information on the samples geometry to provide accurate information on the tubes collapse pressure. However, the 2D models are very useful for developing parametric studies: using 2D finite element models it was shown that the tube ovality and residual stresses have a strong influence on the value of the external collapse pressure. Three dimensional finite element models, developed using shell elements, provide a very accurate prediction of the value of the external collapse pressure, when the input data contains a detailed description of the tube geometry and residual stresses.

18

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

An imperfection measuring system was developed to acquire detailed information on the tubular samples geometry. Using 3D finite element models it was shown that the high frequency modes of the external surface Fourier decomposition do not have a significant effect on the value of the external collapse pressure. Acknowledgement We acknowledge the experimental data provided by D.H. Johnson from the Center for Industrial Research laboratory. We acknowledge the financial support for this research of SIDERCA (Campana, Argentina) and TAMSA (Veracruz, Mexico). REFERENCES [1] S.P. Timoshenko and J.M. Gere, Theory of Elastic Stability, Mc Graw Hill (1961). [2] D.O. Brush and B.O. Almroth, Buckling of Bars, Plates and Shells, Mc Graw Hill, New York (1975). [3] O. Heise and E.P. Esztergar, "Elastoplastic collapse under external pressure", ASME, J. Engng. for Industry, 92, 735-742 (1970). [4] W.O. Clinedinst, "Analysis of collapse test data and development of new collapse resistance formulas", Report to the API Task Group on Performance Properties (October 1977). [5] Y. Tamita and A. Shindo, "On the bifurcation and post bifurcation behavior of thick circular elastic-plastic tubes under lateral pressure", Comput. Methods Appl. Mech. Engrg., 35, 207219 (1982). [6] J.R. Fowler, E.F. Klementich and J.F. Chappell, "Analysis and testing of factors affecting collapse performance of casing", ASME, J. Energy Resources Technology, 105, 574-579 (1983). [7] G. Krug, "Testing of casing under extreme loads", Institut of Petroleum Engineering, Technische Universitt Clausthal (1983). [8] M. Kanda, Y. Yazaki, K. Yamamoto, H. Higashiyama, T. Sato, T. Inoue, T. Murata, H. Morikawa and S. Yanagimoto, "Development of NT-series oil-country tubular good", Nippon Steel Technical Report, 21, 247-262 (1983). [9] H. Mimura, T. Tamano and T. Mimaki, "Finite element analysis of collapse strengh of casing", Nippon Steel Technical Report, 34, 62-69 (1987). [10] M.K. Yeh and S. Kyriakides, "Collapse of deepwater pipelines", ASME, J. Energy Res. Tech., 110, 1-11 (1988). [11] J. Arbocz and C.D. Babcock, "The effect of general imperfections on the buckling of cylindrical shell", ASME, J. Appl. Mechs., 36, 28-38 (1969). [12] J. Arbocz and J.G. Williams, "Imperfection surveys of a 10-ft diameter shell structure", AIAA Journal, 15, 949-956 (1977).

19

Andrea P. Assanelli, Rita G. Toscano, and Eduardo N. Dvorkin.

[13] W.H. Press, B.P. Flannery, S.A. Teukolsky and W.T. Veherling, Numerical Recipes, Cambridge University Press, Cambridge (1986). [14] M.S. Shunmugam, "Criteria for computer-aided form evaluation", ASME, J. Engng. for Industry, 113, 233-238 (1991). [15] K.J. Bathe, Finite Element Procedure, Prentice Hall, Upper Saddle River, New Jersey (1996). [16] R. Hill, The Mathematical Theory of Plasticity, Oxford University Press, New York (1971). [17] E.N. Dvorkin and S.I. Vassolo, "A quadrilateral 2-D finite element based on mixed interpolation of tensorial components", Engng. Comput., 6, 217-224 (1989). [18] E.N. Dvorkin and A.P. Assanelli, "Elasto-plastic analysis using a quadrilateral 2-D element based on mixed interpolation of tensorial components", Proc. Second International Conference on Computational Plasticity (COMPLAS II), Pineridge Press, Swansea (1989). [19] E.N. Dvorkin, A.P. Assanelli and R.G. Toscano, "Performance of the QMITC element in twodimensional elasto-plastic analyses", Computers & Structures, 58, 1099-1129 (1996). [20] E. Corona and S. Kyriakides, "An unusual mode of collapse of tubes under combined bending and pressure", ASME, J. Pressure Vessel Technol., 109, 302-304 (1987). [21] J.R. Fowler, B. Hormberg and A. Katsounas, "Large scale collapse testing", SES Report, prepared for the Offshore Supervisory Committee, American Gas Association (June 1990). [22] E.N. Dvorkin and K.J. Bathe, "A continuum mechanics based four-node shell element for general nonlinear analysis", Engng. Comput., 1, 77-88 (1984). [23] K.J. Bathe and E.N. Dvorkin, "A four-node plate bending element based on Mindlin/Reissner plate theory and a mixed interpolation", Int. J. Numerical Methods in Engng., 22, 367-383 (1985). [24] K.J. Bathe and E.N. Dvorkin, "A formulation of general shell elements - The use of mixed interpolation of tensorial components", Int. J. Numerical Methods in Engng., 22, 697-722 (1986).

20