NOTE: THE INSTITUTE OF APPLIED MATHEMATICS AND STATISTICS (TAMS) SHALL BE NOW KNQWN AS: THE INSTITUTE OF APPLIED MATHEMATICS LECTURE NOTES ONTHE LINEAR THEORY OF THIN ELASTIC SHELLS OF REVOLUTION F.Y.M. Wan Technical Report No. 84-9 August, 1984 Cover photo courtesy of the U.B.C. Museum of Anthropology: Haida totem pofe; main figure, possibly bear, holding wolf between Legs, grog in mouth, wolf between ears.LECTURE NOTES ON THE LINEAR THEORY OF THIN ELASTICS SHELLS OF revotutron Fev.M. wan(?) Institute of Applied Mathematics The University of British Columbia Vancouver, B.C. V6T 1Y4 Canada August, 1984 (1) The research is supported in part by Canadian N.S.E.R.C. Operating Grant A9259 and by U.S.-NSF Grant No. MCS-8306592. (2) Currently on leave as Professor of Applied Mathematics and Mathematics at the University of Washington, Seattle, WA 98195.TABLE OF CONTENTS FOREWORD PART I - THE LINEAR ELASTOSTATICS OF SHELLS OF REVOLUTION Introduction 1) The Scope of This Volume 2) Geometry of Surfaces of Revolution.. Boundary Value Problems of Linear Shell Theory 1) Equilibrium and Compatibility Equations for Shells of Revolution. 2) Displacement Components and Stress Functions. 3) Boundary Conditions ...sssecceeeeevcsceeee see PART II - AXISYMMETRIC STRESS DISTRIBUTIONS Intrinsic Formulation with Axisymmetry 1) Axisymmetric Bending and Stretching 2) Axisymmetric Torsion and Twistin 3) First Integrals and Overall Equilibrium 4) Determination of Displacement Fields. 1) Shallow Shell Approximation 27 2) Reduction of the Axisymmetri 29 3) The Classical Theory for the Bending Problem. 32 4) A Shallow Spherical Cap Under Its Own Wetght. 35 5) Regular Perturbation Solution..........++ 6) WKB Solution, Langer's Method and Turning Point Problems. 7) Reduction of Axisymmetric Torsion Problem. Reduction for Axisymmetric Problems of Nonshallow Shells ) Reduction of the Axisymmetric Bending Problem ) Circular Cylindrical Shells...s..seeeee ) A Second Order Complex Equation for Spherical, Conical and ) Torodal Shells. Finite Difference S Utions e eseReferences. PART III - LATTERALLY LOADED SHELLS OF REVOLUTION Sinusoidal Stress Distribution in Shells of Revolution 1) Intrinsic Formulation of Shell Problems 2) First Integrals....ssceeeene 3) Non-periodic Displacement Fields. 4) Reduction to Two Simultaneous Equations Laterally Loaded Shallow Shells of Revolution 1) Governing Differential EquationSssssacsseecsceeeees 2) Reduction to Two Simultaneous Differential Equations 3) Shallow Spherical Cap in a Face-Side Position. 4) The Side Force and Tilting Moment Problem for a PART IV - SHELLS OF REVOLUTION UNDER ARBITRARY LOADING Shallow Shells of Revolution 1) Shallow Shell Equations... 2) Reduction of Equations for the Class 3) Shallow Spherical Shel1S.sssceseeeeeeseveeees Spherical Shells 1) Governing Equations for Spherical Shells 2) Reduction for the Classical Theory... Circular Cylindrical and Conical Shells 1) Equations for Higher Harmonics of the Shell Response. 2) The Classical Theory of Circular Cylindrical Shells.. 3) Conical She'ls and Generalized Hypergeometric Functions... nee ceaceeeecccneeenscecessesenseeeeees ee lZeForeword Over the years, this writer has taught courses on the theory of thin elastic shells at different institutions, mainly to students in applied mathematics. A set of lecture notes have been compiled from and for these courses. Portions of these notes are now published as technical reports of the Institute of Applied Mathematics (and Statistics, at one time) at the University of British Columbis. These technical reports have limited distributions but are deposited in the UBC Archive for future reference. Because of time constraints, these shell courses usually focussed on the mathematical structure of the various special theories such as the linear theory, shallow shell theory, theory of shells of revolution, etc., and on the solution techniques applicable to the relevant boundary value problems. Some discussion of the foundations of shell theory, i.e., the adequacy of shell theory as an approximate solution of a three-dimensional elasticity problem, is usually included to make the subject meaningful for engineering applications. ‘or this part of the course, it is necessary for the audience to have some background on the three-dimensional linear theory of elasticity. To accommodate students in applied mathematics without the requisite background, the material in elasticity theory essential to the discussion of foundations of shell theory is outlined in reference [1] at the end of this Foreword (not to be confused with references of the main body of this report given at the end of the report). The foundation problem for shells is, at this writing, not completely under- stood and is, in any event, far too complicated for students exposed to the sub- ject for the first time, The writer has found it more effective to discuss the-ii - corresponding problem for linear plate theory as an illustration of the nature of the foundation problems and the types of results attainable. For these reasons, the results of Friedrichs & Dressler and of E, Reissner are presented in refer- ence [2]. It has always been the writer's intention to supplement [2] with a summary of his joint work with R. D. Gregory of the University of Manchester such as [3] and [4] to indicate an attractive and practical alternative solution of the foundation problem for plates. In principle, the same method of approach is also feasible for the corresponding shell problem though its implementation re- quires some fresh ideas. The compilation of the material on the foundation of plates, possibly including other results such as [5], will be the writer's next writing project. To allow for a separate discussion of the structure and solution techniques of shell theory without any reference to three-dimensional elasticity theory, (and the foundation problem), the writer usually follows the approach of W. Gunther, E. Reissner and H. Schaefer, and develops shell theory as the mechanics of a two-dimensional deformable continuum. The constitutive relations will be adopted with little or no justifications whenever there is not sufficient time to discuss the foundations of the theory. The fundamentals of linear shell theory is presented in this way in [6], though the relationship between shell theory and three-dimensional elasticity theory is briefly discussed by way of a class of transversely rigid thin elastic shells. The theory developed there is for gen- eral orthogonal surface coordinates. A tensorial treatment to allow for oblique surface coordinates is avoided to gain time for shell theory proper. Most shell designs encountered by this writer can in fact be analyzed by working with or~ thogonal surface coordinates. In any event, it is not needed for the present of lecture notes.- iii - The developments of two particular classes of shell problems have had sig- nificant impact on the theory and applications of shell structures. For their importance in engineering applications alone shallow shells and shells of revolu- tion deserve special attention. In this set of lecture notes, we summarize the salient features of the linear elastostatics of shells of revolution. While an exhaustive discussion of the very rich subject of shells of revolution is not possible in the space of a technical report, our summary does present the main results of the subject which completes the solution process for an infinitesimal deformation theory under general loading. The report will be short on specific problems; there will be just enough of these to illustrate the main theoretical developments for our class of problems. A similar summary for both linear and nonlinear shallow shells is presented in [7]. As it is pointed out in [7], the very important subject of shell buckling will be treated in a separate report. Through many years of association, the writer has undoubtedly absorbed many ideas from his Ph.D. thesis supervisor and research collaborator, Professor E. Reissner, It would be fair to say that a greater part of these lecture notes is an outgrowth of information assembled over the years from Professor Reissner's class lectures, his writings, and his conversations with this writer. In fact, whatever merits these lecture notes may have, the writer owes it all to his teacher and good friend, Eric.-ive References 1. F.Y.M Wan, “Lecture Notes on Problems in Elasticity: I. Fundamentals of Linear Theory of Elasticity," I.A.M.S. Tech. Rep. No. 82-17, Univ. of British Columbia, Vancouver, Canada, November, 1982. 2. » "Lecture Notes on Problems in Elasticity: II. Linear Plate Theory,” I.A.M.S. Tech. Rept, No. 83-15, Univ. of British Columbia, Vancouver, Canada, September, 1983 3. RD. Gregory and F.Y.M. Wan, "Decaying States of Plane Strain in a Semi- infinite Strip and Boundary Conditions for Plate Theory," J. Elasticity ‘14, 1984, 27-64. 4. » “On Plate Theories and Saint-Venant's Prin- ciple,” TANS. Tech, Rep. No. 83-12, Univ. of British Columbia, Vancouver, Canada, August, 1983, and to appear in Intern'l. J. Solids and Structures, 1985. 5. J.G. Simmonds, "An Improved Estimate for the Error in the Classical, Linear Theory of Plate Bending," Quart, Appl. Math. 29, 1971, 439-447, 6. F.Y.M. Wan, “Lecture Notes on Theory of Thin Elastic Shells," I.A.M.S. Tech, Rep. No. 83-11, Univ. of British Columbia, Vancouver, Canada, July, 1983. 7. “Lecture Notes on Shallow Shell Theory," I.A.M.* Tech. Rep. No. 84-8, Univ. of British Columbia, Vancouver, Canada, July, 1984. *The Institute of Applied Mathematics and Statistics of the Univ rsity of British Columbia was formally re-named as the Institute of \pplied Mathe: tics in March, 1984,Part I - The Linear Elastostatics of Shells of Revolution Jthe_Linear Elastostatics of shells of Revolucion 1. The Scope of This Volume Even with the built-in simplifications associated with its inherent static-geometric duality, the general linear theory of thin elastic shells still poses rather difficult mathematical problems in partial differential equations. Qualitative understanding of the theory and tractable quantitative solutions for specific applications are possible only through additional math- ematical simplifications associated with particular classes of problems. For shells of revolution, certain simplifications naturally suggest themselves. In cylindrical coordinates (r,é,z) the coefficients of the linear partial dif- ferential equations governing the linear elastostatics of shells of revolution (whose middle surface is given parametrically by z = 2(6,) and r = r(é, )) are, as we shall soon see, independent of the polar angle 6. It follows that the loading and response of circumferentially complete shells of revolution (in the form of a dome or a frustum) have Fourier series representation in the form F(,48) = Een(é,)e0s(ne) + sp(=,)sin(ns)) Linearity then allows for a separate analysis of each Fourier coefficient, cn(é,) or Sp(6,)+ 1 = 0, 1, 25 «++ Not So obvious however, are the different addi- tional simplifications associated with the different harmonics. Also, not so easily seen are the simplifications and reduction associated with important classes of problems involving circumferentially incomplete shells of revolu- tion including the special case of a shell of revolution slit along a meri- dian, It is our intention to summarize in this volume the main theoretical results of the rich body of knowledge accumulated from various simplifications-2- and reductions for shells of revolution. These results are essential not only to a true understanding of the theory of shells of revolution but also to an effective solution for specific problems. Historically, the first result in this direction was for the axisymmetric bending (and stretching) of spherical shells [1]. With all stress and strain distributions independent of the polar angle 8, the linear partial differen- tial equations of force and moment equilibrium reduce to a system of first order ordinary differential equations in a meridionat (independent) variable §,+ The requirement of overall axial force equilibrium gives a first integral of these ODEs. The existence of this first integral makes it possible to transform the only relevant differential equation of moment equilibrium into a second order ODE for a stress function and an angular displacement variable. Not having the benefit of the static-geometric duality at that time, it was considered a stroke of genius that Hans Reissner succeeded in obtaining the appropriate compatibility equation Bp ee for the same two un- knowns. The two simultaneous second order equations were further reduced to a single second order equation for a complex function. This second order equa- tion in turn shows that the general solution for the axisymmetric bending of spherical shells can be expressed in terms of associated Legendre functions. Shortly after the pioneering work for spherical shells, E. Meissner [2] used a similar approach to reduce the axisymmetric bending problem for general shells of revolution also to two simultaneous second order ODE's for a stress function and a strain function. His results are esthetically unsatisfactory. They do not preserve the static-geometric duality inherent in the shell equations (though not known at the time). Also, there is some difficulty in taking his results to the flat plate limit. Both of these shortcomings were removed later by a different (and more natural) choice of the stress functionintroduced by E. Reissner [3]. More recently, a similar reduction was accom~ plished for the more general Gunther-Reissner-Schaeffer type shell theory as well as for the axisymmetric torsion (and twisting) problem [4]. In both cases, allowance was made for shells in the form of a segment of a frustum (and therefore circunferentially incomplete) without destroying the elegant mathematical structure of the Reissner-Meissner-Reissner type formulation. As we shall see, this mathematical structure enables us to bring many exact, asymptotic and numerical solution techniques to bear on specific axisymmetric problens for shells of refolution which would not be tractable otherwise. References to other results of intermediate generality can be in [4]. ¥ Problems associated with the first harmonics in the Fourier decomposition in the polar angle © are also amenable to simplifications and reduction simi- lar to the axisymmetric case. However, simplifications and reduction for these so-called lateral (or wind) load problems were only recognized and im- plemented for the first time in 1959 by V. S. Chernina [5]. First integrals of the equilibrium and compatibility equations were derived for lateral load problems in [5] and the system of shell equations was also reduced in [6] to two simultaneous second order ODE's for a stress function and a strain func- tion. Similar to Meissner's results for axisymmetric bending, Chernina's re- duction also does not preserve the static-geometric duality inherent in shell theory and the resulting two ODE's take on an indeterminate form in the flat plate limit, These shortcomings were removed in [6,7]. To the degree of ac- curacy of shell theory, the two new second order simultaneous equations are remarkably similar to the Reissner-Meissner-Reissner equations for axisynme- tric bending problems. The new reduction of [6,7] also allows for nonpertodic displacement fields which correspond to strain fields proportional to sind and cos8; thus the results obtained apply to laterally loaded ring shell sectors as well.a4. For more general loading leading to shell response with higher harmonics in the Fourier decomposition in the polar angle 6, it has been known for some time that the equations of the classical linear (and nonlinear) shallow shell theory can be reduced to two simultaneous fourth order PDE's for the trans- verse midsurface displacement and a stress function [8]. A similar reduction is also possible for spherical shells [9,10]. The reduction to two simultan- eous equations has again led to exact and asymptotic solutions for specific problems which would not have been recognized otherwise. The success attained jn the case of shallow shells and spherical shells naturally stimulated effort toward similar reduction for other classes of shells. This effort was further encouraged when the desired reduction was accomplished for circular cylindri- cal shells, at least within the error inherent in shell theory [11]. However, less encouraging results began to emerge shortly afterwards [12,13]. In the last part of these notes, we will describe the reductions for shallow shells, spherical shells and circular cylindrical shells. For the circular cylindri- cal shell case, we will accomplish the reduction using the procedure proposed in [14]. The same procedure will then be applied to conical shells [15] to show for the first time that a two simultaneous fourth order ordinary differ- ential equations formulation is always possible for the nth harmonics Ca(é,) and Sq(é,) of the shell response (n > 2). It is seen from the development for ‘a conical shell that the same reduction to two simultaneous fourth order differential equations can also be carried out for general shells of revolu- tion and that, consistent with [12], a reduction to two simultaneous fourth order PDE's similar to situation of w and F of the shallow shell is not always possible.-5- 2. Geometry of Surfaces of Revolution The position vector for a point on a surface of revolution may be written as P= (Edin + 26, )iz wnere Tp and 7, are the unit vectors in the radial and axial direction, respectively, for the cylindrical coordinate system (r,6,z), with Fp = costly + sinly. The third unit vector fg of the orthogonal triad Grodarts is given in terms of the cartesian unit vectors ix and iy by Yq = -sinoly + cosody with d(7p)/d0 = ip and d(¥g)/de = -fp. From ry we can compute the unit tangent vectors €, and ¢, along the coordinate curves on the surface, the arc length coefficients (i.e.; the coefficients of the first fundamental form of the surface), the unit normal vector and the radii of curvature (or the coefficients of the second fundamental form). We take €, = © and indicate partial differentiation with respect to 6, and © by ()' and ()", respectively. We get from Tope rtip + 2'tz and Fr= rfg so that t, = coséip + Singiz, t,= ig where~6- The quantity € is evidently the angle made by the tangent of a surface meri- dian with the base plane (x,y-plane). The surface coordinates are orthogonal as t+ t, = 05 it follows that n=t) xt, = - singip + cossiz and from the Weingarten formulas of differential geometry of surfaces (see Chapter 4, section (4) of Reference [16]), 1... Rr 1 __ sing 2 22 r ra? 1 .J.-9 Riz” Rar The surface coordinates (£,8) are evidently lines of curvature coordinates. The quantities a 1 7% &, and Ris are functions of ay only. That they do not depend on ¢, = ais a feature of surfaces of revolution and will be useful for the analysis of shells of revolution. In the subsequent development, the subscript 2 in all stress, strain and displacement measures for shells of revolution will be replaced by 6 to reduce the amount of bookkeeping. To the extent that ¢, will always be a quantity which varies along a surface meridian (including the possibility of &, = &) the subscript 1 will be replaced by & whether or not we have £, = £. We note that it is not always possible to choose & = 8s given that & is a constant for a conical surface of revolution.2 - Boundary Value Problems of Linear Shell Theory 1. Equilibrium and Compatibility Equations for Shells of Revolution With aj and Ryj as given in the previous section, we may now specialize the scalar differential equations of equilibrium for orthogonal surface coordinates (given in chapter 4, section (5) of [16]) to the special case of shells of revolution. These equilibrium equations are recorded here for references and/or further specializations in later sections. (FNge)' + (aNge)” - aCOSENgs - ré'Qz + rape = 0 (rNeg)' + (aNgg)” + acosENge - aSineQs + rapg = 0 (rz)! + (aQg)" + PE'Nee + aSINENgg + Papy = 0 (rMge)' + (aMge)’ - acoséMgg - raQe - asingP, + rage = 0 (rMzg)' + (aMgg)” ocoSEMge - raQs + ré'Pe + radg = 0 (rPc)' + (aP9)” + ra(Neg - Noe) - PE'Meg + aSiNEMge + ragn = 0 where ( )' = 9( )/2, and ( y= a( )/28, respectively, and £,, is generally different from the meridional slope angle &. The compatibility equations for the strain measures of shells of revolution can be obtained immediately from these equilibrium equations with the application of the static-geometric duality of linear shell theory (see Chapter 6 of [16]). These compatibility equations are identical in form to the above equilibrium equations with all their stress measures replaced by the dual strain measures according to the following table:Nee Nos Neo Nog Qe Qe Pe Pe Pr -koe -Keg og kee Mee Meg Meo Moe Pe Pa de Gen feo fee fee «=e YO YEO 0 0 The equilibrium and compatibility equations may be supplemented by a set of stress strain relations to form a complete system of governing equations for the elastostatics of shells. Unless indicated otherwise, these stress strain relations will be taken in the form cee = A(Nee - vsNoo)» eae = A(Neo - YsNez)» eco = ec = Ag(Neo + Nog)» Yo = Age» ve = AQhe Meo = D(kao + Ypxee)» Mee = D(kee + Ypkeg)» Meg » Dg(kge + Kee)» Pe = pres Pg = Dpdg where Ag = A(l+vg)/2 and Dg = D(1-vp)/2. We note that the first half of these relations are the static-geometric duals of the second half if we observe the following dual relations among the elastic parameters for the shell:-9- For the conventional first approximation theory of isotropic (and linearly elastic) shells which are homogeneous across the shell thickness, we have 3 i Eh asi, o-—h _, zvpzv, Dp = Ag= 0 ih 12(1-v2) Ys = Yb = ¥ peg where E, is Young's modulus, v is Poisson's ratio and h is the shell thick- ness. In the subsequent development, we consider only shells for which these parameters do not vary in the circumferential direction. To the extent that only quantities which appear in the stress strain relations are involved in the present formulation of the elastostatic problem of shells of revolution, it is customary to refer to this formulation as an intrinsic theory of shells.10- 2. Displacement Components and Stress Functions With € = © the strain-displacement relations for shells of revolution may be obtained from the corresponding relations for general orthogonal surface coordinates given in [16] (Chapter 4, section (5)). Ne record them here for future references and/or further specializations: egg = E(ug-6'w)," cee = 4(ug + costug - singw) eco = UB ales coy Ege = pve - COSEUg) + w liye ‘ 1 ; Ye = ge +Z(w' + Etue), Yo = 0 + p(we + sinéug) kee = 2 68 Kog = Elo + COSES,) a3 a a0 ~ Ti% & Keg = U(oh + Eu) = 16: + cos singw) 8 ~ alo Kae ~ PE Bbq - sing Lut - £'9,) Ag = (wr + Singd,) a 5° bq 67 $. where r, a and € are functions of &, only. From the above strain displacement relations, we can immediately obtain a stress function solution of the equilibrium equations by the static- geometric duality of linear shell theory. The stress function representa- tions of the stress resultants and couples are identical in form to the above strain-displacement relations with all strain measures replaced by their dual stress measures according to the rules given in the last section and with the displacement components replaced by their dual stress functions accord to the following table-11- The strain-displacement relations and the stress function representations along with a set of stress strain relations also form a complete system of governing differential equations for the elastostatics of shells. To the ex- tent that displacement components and stress functions are not quantities in- volved in the stress strain relations, we shall refer to the present formula- tion of shell problems as an extrinsic formulation.-12- 3. Boundary Conditions A complete set of shell equations is to be supplemented by an appropriate set of boundary conditions to form a well-posed boundary value problem for the elastostatics of the shells. The appropriate set of boundary conditions is of course dictated by the particular physical problem. That the resulting boundary value problem is well-posed may be confirmed by variational considera - tions [16]. For problems with a prescribed displacement field along an &, = constant edge, we have from [16] the following six scalar displacement boundary conditions Aug = AUg = AW = Age = Adg = Aw = 0 where af = f(é7,0) - f*(9) and ( * is a prescribed quantity. For classical (ise. the conventional first approximation) shell theory, the condition Pz = 0 eliminates aw = 0 as an Euler boundary condition of the variational principle for stresses, strains and displacements [16], while the condition yg = 0 and the strain displacement relation Yo = 5 + Hw + singug) a 8a Fr 6 do not allow $g to be prescribed independently. Hence, we have for the classical theory the following four displacement boundary conditions along * ei:-13- where Age = O may be replaced by aw’ = 0 in view of the condition of vanishing transverse shearing strain yz = 0. Appropriate displacement boundary condi- tions for “lower order" shell theories (such as membrane or inextensional bending theory) can also be obtained by specializing the corresponding condi - tions in [16] to shells of revolution with &, = 6, For problems with a prescribed stress resultant field and a stress couple field along an £, = &1 edge, we have from [16] the following six stress bound- ary conditions: £ = Ef: ANge = ANeg = OQe = OMee = OMeg = APE = 0. For classical shell theory, we must nave Pz(@) = 0 to be consistent with 0, while the requirement of no transverse shearing strain & Ye = 9 = 0, leads to the following four Kirchhoff-Bassett contracted stress boundary conditions along an £, = 6; edge [16]: * Ey) = Ej: aNg, = ANE e 12 Nee = ANeg = GUE = aMee = 0 ee where e | e ly? Neo" Neo * Roe? SE 7 tT Meo (Note that 1/R,, = 0 so that Neg = Nee + Meg/Reg = Ne,-) Appropriate con- tracted boundary stress boundary conditons for lower order shell theories may be found in [16] and will not be discussed here. It is of considerable interest to note that displacement boundary condi - tions may be alternatively formulated in terms of strain measures. This is particularly important for an intrinsic formulation as it would be rather-14- cumbersome at best to integrate the strain-displacement relations before the general solution of the intrinsic theory can be specialized by the actual displacement boundary conditions. When all six displacement components are prescribed along an &) = es edge, we may use the following six conditions on the strain measures for an intrinsic formulation of shells of revolution: 1 68 eet: _ = = E, = EY! Okgg = Akge = Ag = Ae The corresponding conditions for the classical shell theory are: = art = 5 96 = Skee 7 bAg = S€gq = 0 where Se and 2€ are the static-geometric duals of Neg and Oe respectively. Both sets of strain conditions are obtained by specializing the corresponding conditions in [16] for shells of revolution with s, = @. It is clear from the strain displacement relations of section (3) that Rees K hee Mo Shas he and y% can be determined once we have UE UG Ws OE 6% and u* as only differentiation with respect to 6 is involved in the relevant strain displace- ment relations. By the static-geometric duality, we may also formulate the stress bound- ary conditions in terms of stress functions if we wish to do so. The result- ing boundary conditions for the stress functions are just the static-geometric duals of the displacement boundary conditions for displacement boundary value problems. It is also clear that similar boundary conditions along a 6 = 6 edge can also be written down. As these conditions are similar to those for &) = con- stant edge, they will not be recorded here.-15- Mixed boundary conditions with some of the conditions given in terms of stress resultant(s) and/or couple(s) and the remaining conditions in terms of displacements discussed in [16] for them to lead to a well-posed boundary value problem. The most convenient (admissable) form for a set of mixed conditions in a given formulation of the shell problem (intrinsic, extrinsic or others ) will have to be investigated on an individual basis. In general, a set of mixed conditions given in terms of stress resultants, stress couples and displacement components may be left in its original form in an extrinsic formulation.-16- Part IL - Axisymmetric Stress Distributions, 3 - Intrinsic Formulation with Axisymmetry ng and Stretchi For stress and strain measures which are independent of the polar angles 8, the equations of equilibrium and compatibility for shells of revolution given in Chapter 2 of these lecture notes naturally divide themselves into two uncoupled groups. The first group consists of the three equilibrium equations ° (PNee)' = @ COSE Nog - PE'Qe + Pope = (PQ)! + PE'Nee + a SINE Nog + Papy = 0 (PMze)' - « COSE Mgg - PaQe - a Sing Py + rage = 0 and the three dual compatibility equations. The stress and strain measures involved in this group of equations are evidently associated with the stretching and bending action of the shell. Six of the twelve stress strain relations for the shells of interest in these lecture notes given in Chapter 2 also contain only the same twelve stress and strain measures: e997 A(Ngg - vsNeg)» ege = A(Ngg - vsNeg)> YF Agde Mee = D(keg + vpKee)> Mee = D(koe + YbKEE) > Py = Dpdg The three equilibrium equations, three dual compatibility equations and these six stress relations form a sixth order system of twelve (differential ) equations for Nez, Ngg> Qe» Mees Mag» Pg. and their dual strain measures.<1 Supplemented by suitable boundary conditions at the edge(s) of the shell, we have a well posed boundary value problem in ordinary differential equations. Of the six stress boundary conditions along and &, = a edge, three involve stress measures which appear in the axisymmetric bending (and stretching) problem*. They are: 5) 8T: ANeg = AQ = AMee = 0 As we shall see in a later section the quantities Nee and & cannot be complet- ly arbitrary at one of the edges of the shell, as the shell structure must be in overall equilibrium, Note also that the three stress boundary conditions remain unchanged for classical shell theory, keeping in mind that we have ade = a(Q, + rl 9) = AQe because of the stipulation of axisymmetric stress measures. If displacement boundary conditions are prescribed along &, = ae they are most conveniently taken in the form of strain boundary conditions. Among the six appropriate strain boundary conditions given in Chapter 2, three involve the strain measures associated with the axisymmetric bending problem. They are BT: Akg = bg = degg = 0 Keeping in mind a§ = A(A9- r-less) = MAg because of axisymmetry these three conditions remain unchanged for the classical shell theory. By the static geometric duality, we expect that Kee and 4 cannot be prescribed arbitrarily at one of the shell edges because of the requirement of overall compability. (*) It is customary in the shell literature to omit the reference to stretching for this class of shell problems.-18- 2. Axisymmetric Torsion and Twisting The second group of equilibrium and compatibility equations for axisymmetric stress and strain distributions consists of the three remaining equilibrium equations (PNe9)' + acosENge - asin&Qs + raps = 0 (TMeg)' + ocoseMge - raQe + ré'Pe + raqg = 0 (rP¢)' + ra(Neg - Nog) - PE'Meg + aSingMge + ragn = 0 and the three dual compatibility equations. They contain only the stress measures Neg, Nog, Qe» Megs Mge and Pg and the dual strain measures. The six remaining stress strain relations not used in the axisymmetric bending problem are 9 = cee = As(Neg + Noe)» Ya = AgQe Mog = Mee = Ds(kog + Keo)> Pe = Dpre with As = A(1+vs)/2 and Ds = D(1-vp)/2. They also involve only the same six stress measures and their dual strain measures. The sixth order system of three equilibrium equations, three compatibility equations and six stress strain relations for the twelve stress strain measures evidently governs the axisymmetric shearing and twisting actions of shells of revolution. Because of the types of physical problems involved, this general class of problems is commonly known as axisymmetric torsion (and twisting) of shells of revolution. The sixth order system is supplemented by three boundary conditions at each edge of the shell. The three stress boundary conditions along an & = ET edge are: gap ANeg = AMeg = APe = 0-19- For the classical theory, AP, = 0 should be trivially satisfied for consistency (in particular, we must have PE = 0) wile y, 20 leads to the Kirchhoff- Bassett contracted stress boundary condition : eo. 6, = 8: aNee =O sep ye 7 . eG with N2g = Neg + Meo/Rog = Neg - Sine Meg/r This Is the only stress boundary for an £, = constant edge for this class of problems. If displacement conditions are prescribed, they are most conveniently taken in terms of strain measures for our intrinsic formulation of the problem. The appropirate strain boundary conditions along an &, = gq edge are: *, = . = Bt: beg = Begg = AY, = 0 For the class theory, dy, = 0 should be trivially satisfied (in particular, we must have yf = 6% + r7'L(w#)"+ sincug] = 0, (Note that we have not required that all displacement components to be axisymmetric as long as they lead to axisymmetric strain measures. We will return to this point in section (4) of this chapter.) At the same time, the condition Pe = 0 leads to the contracted strain condition athe : with, = Kee &ge/Rog = Kog + SINE(Ege/F)- Similar to the axisymmetric bending problem, overall equilibrium and apd * * * * compatibility, NE, and Mf, for the stress case and x§_ and eG, for the displacement case cannot be arbitrarily prescribed at one shell edge. We will return to this point in the next section.~20- 3. First Integrals and Overall Equilibrium For the axisymmetric bending problem, we may form the following combination of the two force equilibrium equations, sing[(rNe¢)' - acoséNgg - Pé'Qe + rape] + coséL(rQc)' + FE'Neg + aSiNENog + rapa] = 0 to get [r(Negsing + Qecosé)]' + rapy = 0, py = PeSiNE + Pacosé. Upon integration, we get, for =} < & < fo» r(Ne-Sing + Q_cosé) = rV(E,), r= - in py(t)r(t)a(t)dt where F, is a constant of integration. The existence of the above first integral of the equilibrium equation is expected from the requirement that the shell be in overall force equilibrium. The overall equilibrium of an arbitrary shell frustrum requires 1°75 Ceaiigy + (allg)’ + r(tyaCt)b(t.0)}6tao = 6, 3 i which in component form is of the form + ore ; ix [ [> Cfp(é,)eoss - sine fe(é,) Ide de o € an &. + iy ff [fple,)sine + fo(e, )coselde de 0 € > én €. + + iz f [> Lir(Qecose + Neesing)}' + rapy]de,de = 0 é-21- (It will not be necessary to write out the expressions for f,(é,), etc.) We get from this only one nontrivial condition of overall equilibrium in the direction of the axis of revolution, i.e. the axial direction: 2n[r(Qecose + NeeSing) + ft py(t)r(t)o(t)dt] = Fz 5 For the case of no surface load so that py = 0, the constant of integration Fz is the resultant axial force experienced by the shell which must be the same throughout. If stress conditions are prescribed at both circular edges of a shell frustum, &) = €, and €, = &,, then we must have i i P(8, EQe(, eos (5(5,)) + Nge(E, )sin(6(6,))I = P(E )EQe(E,)cos(E(E,)) + Nee(E,)sin(E(E,))] so that Q° and NE cannot be arbitrarily prescribed at both edges. The situation is modified in a trivial way if py # 0. In that case Fz is the resultant axial force at the "lower" edge —& = €,; at the “upper" edge 1 i &. 7 & (> |)» the resultant axial force is now F, net the resultant force from py. By the static geometric duality, we have also the following first integral of the compatibility equations for the axisymmetric bending problem P(kggSing - Agcosé) = - Cg where Cg is a constant of integration, the static geometric dual of (rV) (or (Fz/2 if py = 0).=22- For axisymmetric torsion and twisting, we may form the combination Sine[(rMcg)' + COSEaMge - raQy + ré'Pe + rage] - cosé[(rPz)' + ra(Neg - Noc) - Pé'Meg + SinEGMpe + rag] = rE(rNe9)! + coseaNg - SingaQ, + rapg] = 0 of the three scalar equilibrium equations of the group to get Sing yy, SOSE 2 r'Ngg = p Meg top Pe) = Te with I. ey Po z + ! (rpg + CosEqn ~ Singq,)ra dé, i where Tz is constant of integration. Physically, this first integral is a consequence of the requirement overall moment equilibrium. From ae ae 5 oe > ff} ((rMe)! + (aMlg)” +r! x (rg) + 7 x (ally) + rag 0 E +P x C(riie)! + (aNlo)” + rap]} dg dg = 0. we get as in the case of force equilibrium only one nontrivial condition associated with the resultant axial torque (turning about the axis of revolution) : é -2n[r(Megsing = Pecosé - PNeg) + { ' (qgSing - qncose - rpg)radé] = Tz 5 . where the left hand side is the resultant moment turning about the positive iz direction.-23- By the static geometric duality, we have also the following first integral of the compatibility equations for the axisymmetric shearing and twisting problem: r(egesing + YgCosé + rege) = Cr where Cr is a constant of integration, the static geometric dual of T (or T,/2n if aq = 49 = Po = 0)+ The constants Cg and Cy in the two first integrals of the compatibility equations have their own physical significance which can be readily seen from the discussion in the next section on certain nonsymmetric displacement fields associated with axisymmetric strain fields.=24= 4. Determination of Displacement Fields Once we have the strain components from the solution of the boundary value problems for axisymmetric bending and stretching of section (1) and for axisymmetric torsion and twisting of section (2), we may then integrate the twelve strain displacement relations of section (2) of Chapter 2 to get the displacement components o;, $55 Ws Ug» Up and w. The fact that the differen- tial equations of compatibility have already been satisfied ensures that the twelve relations do not overdetermine the six displacement components. In the process of determining the displacement fields $ and u, it is important to realize that the displacement components themselves need not be axisymmetric although they must give rise to axisymmetric strain components. To see what possible forms of unsymmetric displacement components may be, we note that et > > > ake = o' = [-(coséeg + Singw)'ip + (cosgw - singds)'iz + etal re, = o'= [loi= (cosse, + singu)}i, + (w'cose - 92 sing)iz = (oe + 0895 + singu")ipd] If we write with no loss in generality f = f(c) + f(€.,8), it follows from the g 1 above relations that $,(6,.8) = $,(0), cosed - sings, = (0), coseg, + singa = $,(0) $50) = 5,00) #96), or coséa - sings (9) = Cyd, $,(8) = C,cose + C,sino + C,, $,(8) = - C\sina + C,cose + C,~25- C, are constants of integration still to be determined. The where Cy, Cis + terms associated with C and C, correspond to a rigid body rotation and may be deleted while the terms associated with C, and C, may be absorbed in $,(6,) and or (E,) F cosd,(€,) + sinea(é,), respectively, leaving us with b, 24, = 0 and $, = Cyo, or og #0 $4 = - Cposing, & = Cyscose, and therewith (E), 5 = $5(E,) - Cposing, w= G(E,) + Cgecose A similar analysis on the strain resultant vectors gives a, = 0, iy = Cor and Cy8, or G, = Cyesing, Cyer, W = Cyécosé, where Cy is another constant of integration, and therewith uz = G,(6,) + Cyosing, Gg(é,) + Cgor, w= WE.) + Cyocose Three of the corresponding components of strain resultant vectors and three of the corresponding components of strain couple vectors are just those given in section (2) of Chapter 2 with u,, Ws Ugs ¢-s %, and w replaced by the corresponding ues wW, etc. The remaining components of strain resultants and couples now take the form e Ty) = Heoseil, - sine) + 0g, eye = A(Cysine - coseli,y + W-26- 1 a lec A at FlCycos + U,sing), Kgg = pl-Cgsing + $,cosey c,. = -4(3,cose + dsing) + a, = Ucgcose + 3,sine) 8 TY 6 S BP 9 = pee gets With cgg and A, both expressed in terms of 3,, we have immediately the following linear relation between the two strain couple components r(xggsing - Agcoss) = -C B Similarly, the expressions for 24,4 yg and xg, may be combined to give a linear relation among the three quantities and Cy: rleg sing + ygc0sé + ge] = Cy These two relations are identical to the two first integrals of the compatibility equations obtained in the last section. From the nonperiodic displacement fields 3. Cae, G Cpre, @ we see that Cg2n is a relative rotation about the axis of revolution between two sides of a radial slit, 80 and 6=21, while Cy2r is a relative displacement in the direction of the axis of revolution between the two side of a radial slit. They characterize the only possible global incompatibilities or dislocations associated with axisymmetric strain measures.-27- 4. Axisymmetric Stress Distributions for Shallow Shells of Revolution 1. Shallow Shell Approximations The slope dz/dr of the meridional profile of a shell of revolution is related to the slope angle ¢ (between the tangent at a point along the meridian and the radial line emanating from the same point parallel to the base plane, ive., the (x,y) plane) by tone = A shell of revolution is shallow if (dz/dr)? << 1 so that 1 + (dz/dr)? = 1, tang = sing = & = dz/dr and cose ~ 1. It is evident that in-plane displacement components (in directions tangent to the shell's middle surface) are usually small compared to the out-of-plane displacement component (in direction normal to the shell's middle surface) for shell structures in general. A second feature characterizing a shallow shell stipulates that the in-plane displacement components are an order of magnitude smaller (by a factor z' for shells of revolution): Ug, Ug = O(EW), w = 0(6¢2,6¢9) It follows from these magnitude relations that there is no distinction between the axial displacement component uz and the normal component w as uz = cosgw + singue = w[1+0(z'2)] = w, etc. As we shall see, they also simplify the strain displacement relations in a qualitatively significant way. In the remainder of this Chapter, we take €, = r so that a? = (r')? + (z')? #1 for shallow shells. The strain-displacement relations are then simplified to read € - ZW, yy, =o, +W' ug = 2'W) + Cg, et te zl a0 = rh ie = pl2'te + Cg) 6? re~28- for axisymmetric bending, where now ( )' = d( )/dr, and 1 ' ' a fg = HCpz 7 uy) + ow, fg 7 Ug 7 Ys Ye = ¢9 + por eo, 2 - 4 K ot 8, re 50 ~ %6 for axisymmetric torsion and twisting. To have a virtual work principle for a shallow shell theory, the equili- brium equations should be taken in the form (rNee)' - Nee + Pe = 0, (rQc)' + rz"Neg + Z'Ngg + PPp = 0 (rMgz)' - Meg - rQe - 2'Pp + rq = 0 for axisymmetric bending, and rNeg)' + Neg + rg = 05 rMzg)' + Mog + rz"Pe - PQ + rag = 0 é é 3 e a) (rPc)' + r(Neg - Nog) + ran = 0 for axisymmetric torsion and twisting. It is not difficult to verify that there are six differential equations of compatibility which are the static geometric duals of these equilibrium equations. Also, the stress boundary conditions along an r = r™ edge consistent with the virtual work principle are aNeg = AQe = OMe = 0 — (Axisymmetric Bending) ANe9 = OMzg = APe = O (Axisymmetric Torsion & Twisting) The three conditions for the bending problem remains unchanged for the clas- sical shell theory. The other three for the torsion problem (so that Cy = 0) reduce to a single condition ANe, = 0 (not aney = 0) for the classical theory as the contribution from ug to y9 is neglected in the shallow shell theory. The appropriate displacement boundary conditions in terms of strain measures may be inferred from these stress boundary conditions by way of the static geometric duality.-29- 2. Reduction of the Axisymmetric Bending Problem A first integral of the force equilibrium equations may be obtained either directly from the combination Z'L(PNeg)' - Noo + re] + E(rQe)' + r2"Nee + Z'Ngo + Pn] = 0 or by specializing the first integral of the nonshallow shell case. By either approach, we have for a shell which spans the interval ry Q, = (PV - 2°48) wl _. 4! ve a0" F $B Seg 7 7 fp Ag = > (2 oq + CQ) The remaining stress and strain measures are obtained from the stress strain relations = All - 23. ee Faq = Altg - Fe og + PRY) See = Al-¥5%% + 7 Mm > YshPH) as 1. Mee = 7 O(tg + 7 tg)» Mgg = - Dap + tp) The radial and axial displacement components are calculated from Up = r(egg -Cg) = ac = vgbg + r2py) - Cpr r r Uz =W=Wo - J de(t)dt = Wo + f gy (t)at % i In writing down the above equations for the clasical theory of shallow shells, we have adopted the conventional practice of replacing pe by py and interpreting g/r as the radial resultant.-33- The seemingly trivial simplifications attained by specialization to the classical shell theory actually have some important consequences. If we multiply the second by i¥A/D, and then add the resulting equation to the first, we effectively combine the two ODEs for og and #g into a single complex ODE of the same order for a complex function x = 9g + i YA/Dog: Lally. az a re r/DK bg C = dy trae = (ov - 1/B troy + (2hyQ)Py + Ge) There is considerable advantage working with a single second order ODE for x instead of a fourth order system for og and g. For example, the homogenous equation for y has solutions in terms of Bessel functions if z' = &9(r/ro)" (where n may not be as integer); standard techniques such as WKB and Langer's method may be applied whenever |z'/r/DA| >> 1, etc. In fact, for n # -1, we have with x = r/to, Z x= CylViy) + Cyy(4y), y= rac! where 87 = Egrg/YDA = O(Eoro/h) >> 1 and y = 2/(n+1). If y is an integer, then J_y Should be replaced by Yy or Ky. dp(VTy) is more conveniently written in terms of Kelvin functions, Jp(viy) = berp(y) - ibeip(y). with PPberpei(y) = Lbers(y) - bei i(y)] - B Cberp(y) - beip(y) Poet peyly) = [bers(y) + beit(y)] - F fberp(y) + betp(y)3 and (for y >> 1)-34- ey/v2 ; _y 0, pr {costp, sintp}, tp = Bt {berp(y), beip(y)} ~ along with similar formulas for the Kelvin functions of the second kind kerp Z 4 P iked and keip with Kp(y/yT)/1 = kerp(y) - ikeip(y)+ For n= -1, the equation for x is equidimensional and we have for the complementary solution: va Y2 x= Cx 1 + Cx where y, and y, are the two roots of y> = 1 ~- Eofoi/YDA. We will illustrate the solution process for specific problems in the next few sections.-36- 4, A Shallow Spherical Cap Under Its Own Weight Consider a shallow spherical cap with z' = é9(r/rg) = Ex which is closed at the apex so that 0 ap eghe 3 *Readers should verify that there would be an extra term of order &°/8” on the right hand side of the ODE for X if pz had been used instead of py-36- with 8° = Egro/YDA. An exact particular solution of the inhomogeneous equation is ~ omare Je % 7 1 ED YD * With the exact complementary solution given in terms of Kelvin functions, we have as the exact solution for x. indong x = C,[ber (ex) - ibef (8x)] + C,[ker, (Bx) - ikei,(8x)] +5 /s where C, and C, are two complex constants of integration. for the stresses and displacements of the shell to be finite at the apex, we must have leaving us with roe [x x= (C, + iC, )[ber, (8x) - ibet,(sx)] + rp * 1C;)Eber, : D 2 where we have set C, = Cp + iCj, or tg = Cyber, (ex) + Cybei, (8x) i . A Toehg Jar = Cyber, (Bx) - C.bei, (8x) + /y —Ze- x The two real constants Cj and Cp are determined by the clamped edge conditions rary: 3 = Pega = 0 where we have made use of the relation up = regg to get the condition of no radial displacement at the edge in terms of the stress function og. These two conditions give the following two simultaneous equations for Cp and Cj-37- C,ber, (8) + Cjbei, (8) = 0 r3ehg C; [aber} (8) - veber, (8)] - CCabei ' (8) - vgbei (8)] =e js Eq (1-v,) The solution of this system is 2 ohg(1-v.) fh FoenSt tvs (C, Cj} = Sa (bei (8), - ber, (8)} A= ber,(s)bert(s) + bei‘ (a)bei;(8)] - 73 [bers(s) + bei2(s)] where a dot indicates differentiation with respect to the argument of the function, We may integrate 4g to get w if we wish, with w= Oat r= ro. With B >> 1 for the cap to behave as a shell, we see from the asymptotic behavior of the Kelvin functions that the complementary solution terms ber, (8x) and bei, (ex) are significant only near the edge r = ro and that their contribution is 0(8-!) in og even at (and near) the edge. Thus, the stress resultants Nog and Neg are both of the order phgro/2e, throughout the shell with the associated direct stress magnitude given by N N egr, = 80 = oy = O(-B2, EE) = 2) On the other hand, the stress couples Mrz and Mag are both of order VDA rgohg/Eg near the edge r = rg and are insignificantly small away from the edge. The associated bending stress magnitude is given by MOM egr, = oes, = o(2da op = OSE. -8) = 9) near the edge and again negligibly small away from r = rg. It follows that the bending stresses are significant only near the shell edge while the direct stresses are significant throughout the shell.-38- It is of interest to note that the particular solution of the ODE for y may also be obtained by setting D = 0 in (4.10). Such a solution is called the membrane solution as it corresponds to the limiting case of a shell having no bending stiffness. It is fortuitous that for the present problem, the membrane solution turns out to be an exact particular solution of the shell problem.=39- 5. Regular Perturbation Solution For meridional profiles which do not admit an exact solution for the ODE for x, we may take advantage of the fact that 8° = rgto//DA (where Eq = 2'(rg)) is large compared to unity and seek an asymptotic solution for the boundary value problem. To obtain an asymptotic solution, we set x = r/rg and write the ODE for x (with qe = 0 for simplicity) as lod a2 = -—=> + iB x xX oye t(x)x = - oot (xy 4 aft oy = EGE tla) + EE Ayo)» Ba where a dot indicates differentiation with respect to x, and where 2a, 2 rz = - B°e(x) Seve , Py = Bygy(x)s Py = Bygy (x) raPy : &y =p eH OPH Fylx) = (x94)" + (24¥,) oy - ip fy(x) = - 4 fg (t)tat with py and py chosen so that |gy(x)| < 1 and |gy(x)| <1. With 8? >> 1, we may seek a parametric series solution in powers of g-2: x= J amlx)3-2 = yp(x) m=0 If ey, ey and Cg/éq are all O(1) in magnitude, then we have from the ODE for x~40- 1 <8 C xo = stay Cie Ful) = ap x1 = qytlrod - Ath tHce xael = sup Uae] (n= 1,2,004) where the linear differential operator L is defined by wd treter-30) It follows that - £8 1D teuty + eyurfty + oree* #8 acta)” wer (OntH + eveE ED + oe 25 C BlOeN EF veers eet [eee Ole & EQXt f ca 8 xt Bat Note that successive terms in the perturbation series are determined algebrai- cally and that the solution process gives no constant of integration. Given that the structure of the ODE for x is of the singular perturbation type, the paramet- ric series solution obtained above constitutes only a particular solution of the ODE (or only the outer solution in the method of matched asymptotic expansions). The general solution for x will be obtained by seeking a WKB type solution for the homogeneous ODE for x in the next section.-41- Before leaving the topic of a regular perturbation solution, it should be noted that the leading term for 9g and %g obtained above correspond to the limit- ing case of a membrane and inextensional bending solution of the shell problem. As in the case of the shallow spherical cap under its own dead weight (in the face-up position), the leading term solution for ¢g can be obtained by setting D = 0 (corresponding to a shell with no bending stiffness and hence a membrane) in the ODE obtained from moment equilibrium (with qz = 0). On the other hand, the leading term solution for og can be obtained by setting A = 0 (corresponding to a shell with no stretching compliance 2nd hence inextensible) in the ODE asso- ciated with strain compatibility. In either case, we have 2 = YDA/Egrg = 0 so that the parametric series solution for X reduces only to its leading term. The two limiting solutions corresponding to membrane and inextensional bending ap- proximation play an important role in simple approximate solutions for many shel problems. For example, the membrane solution in the case of a spherical cap under its own dead weight constitutes an accurate approximation of the exact solution except in a narrow region adjacent to the shell edge. As we shall see later, similar situations arise in many other problems including problems involv- ing finite deformation.42. 6. WKB Solution, Langer's Method and Turning Point Problems For a shell frustum with rj > 0, we may seek the complementary solution of the ODE for X by the WKB method. It is customary to first transform the ODE into Liouville normal form for such a solution. For the shallow shell problem, a change of the dependent variable X = ¥xX transform the dimensionless homogeneous ODE for X in section (5) into neen ee e XT = SX + 1p2t(x)X = 0. 4x2 ue We seek an asymptotic (complementary) solution of the above ODE in the form X ~ e80(X) J Xq(xya-n n=0 c Upon substituting the series solution into the ODE and collecting terms of the same powers of 8, we get BE(c')? + itPey + BEL(G')? + ic}K, + 26'%) - <2. Xo} + O(e-) = 0. Pn cet ra + ALCS)" + it + 20K] - XY ae For this equation to be satisfied identically in 8, we must have (s')? + it = 0 or 2c'X = 0 p Fi 3C 2c'X} - xe = 20 te 2 Bxtg" etc. For each of the two roots of c', we have a different solution for x-43- We take a linear combination of these complementary solutions for X = X/¥x in the form: 8n(%)¢¢ cosen + Cysingn] + e°™(*)Ee,cosan - Cysingn] L fe) c cosan - ¢,sinan] + e®")c,cosan - Cysinan]} with x n(x) = f ¥t(x)72 dt where C., Cs C, and C, are four real constants of integration. They are to be determined by the boundary conditions at r = ry and r = ro applied to X ~ Xp + X¢ For a shell closed at the apex, we have ry = 0 and the WKB solution obtained above would not be appropriate near x = 0, The singularity at x = 0 of the ODE for x makes a nonnegligible contribution to the solution (at least in a neighborhood of the apex). Its effect must be included in the leading term solution by Langer's method. To do so, we change both the dependent and independent variable and transform the homogeneous ODE for x into @X 1 dk pl a aie we VG ue ~ Gly)]« + ig” X= 0, where x(y) = ¥(x)x(x) 1 v8 (x) 8) = stay CO + HT eS 1] is O(1/y) as y » 0 if x(x) is bounded away from zero, by choosing = 4x) (x) = Yx¥(X)/y(x)-44- with y(0) = 1. The term multiplied by G(y) is small compared to i8°x away from y = Oand is small compared to y-2 near y = 0 and hence may be neglected in a lead- ing term asymptotic solution. It follows that the leading term asymptotic solution uniformly valid for 0< x <1 is Xe x = Cidi(¥TBy) + CoY,(YTay) provided that 1(x) does not vanish throughout 0 < x < 1, The two complex constants of integration are to be determined by the appropriate boundary conditions for the problem applied to x ~ xp + xc+ If t(x) vanishes in the interval xj < x <1 we have a turning point problem in ODE and a different type of asymptotic solution is appropriate. For example, we have 2'(r) = 9(x-xq) (xj > 1, we again change both the independent and dependent variable in the ODE for x. Since x > 0 for the present problem, we take x = yx and y = y(x) but now with yy = t and y® = xy so that the (homogeneous) ODE for x becomes Ae d 2, a * S— + [is’y + G =0 az * Cig'y + G(y)Ix wher F VOLO” + xe isi It can be shown that G(y) is a bounded for xq < x <1 so that the term involving & may be neglected for a leading term asymptotic solution. In that case, we have KX 2 Ca 8/2y) + cB (1 2/2y)-45~ where C, and C, are two complex constants of integration and Aj and By are the Airy function of the first and second kind, respectively. The two complex constants (or four real constants) are determined by appropriate boundary conditions for the problem applied to the general (asymptotic) solution X 7 Xe + Xpe-46- 7. Reduction of Axisymmetric Torsion Problem As in the nonshallow shell case, we expect a first integral for the equilibrium equations. We may get this by specializing the nonshallow shell result or obtaining directly from the shallow shell equilibrium equations. In either case, we get r PPe + Neg = T(r) = Ty = f (r3p6\+ raq)dr Li where T, is a constant of integration, the resultant axial torque at the inner edge r= ry. By static geometric duality, we have the following first integral of the compatibility equations: ryg + rxg¢ = Cr With the two first integrals, the reduction of equations for axisymmetric torsion is particularly simple for the classical theory. For that case, we have Pe 20, $0 that the first integrals give Ur of Nee = 1, Kee a: The differential equation of force equilibrium and its dual then give Nog = -PPe = (rNeg)' = = rpg - (2)! kee = -(rkoe)' = = (et)!-47- The stress strain relations then give Meg = Mog and eco = cor Meg = Moe = $ D(1-v) (keg + Kee)> fe9 = fee = 4 A(1+¥5) (Neg + Noc) The remaining stress measure Q, and its dual Az are given by the moment equilibrium equation in the t, direction and its dual compatibility equation: 1 1 2 ' PQ, = (PMeg)' + Moe + Pay = BE(reM,g)' + r°qQ] + 1,2 aire ee) It remains to determine the displacement components. From the strain displace- ment relations, we have immediately rod z )o = Flees + Ege) -O Pg oo Mog = Re It should be noted that the stress and strain measures are all determined by algebraic relations (not differential equations). The only stress boundary condition at an edge in the classical theory must be satisfied by the only con- stant integration from the first integral which gives the axial torque. It is evident that we cannot have two arbitrarily prescribed conditions on Neg at two different edge as the shell must be in overall equilibrium, Also, the constant Cy is associated with nonperiodic circumferential displacement and should set equal to zero if the shell is complete in the circumferential direction.-48- For the general theory with Dp + 0 and Ag # 0, we let Teg = oy Pege = $y and, as before, we have immediately Nog = > oy - Mae Seo = > oT From the first integrals, we have also Pees ert i Yet - ot ici The stress strain relations then give Meg = Moe = 2 D(1-vy) (Keg + Kye) = - a DiI) (or - 2 $7) c= 1 - li ieee | co" en 72 A(L+y.) (Neg + Noe) z-3 A( 1+.) (87 Fort To) ¢ % ight or Mee Belts +r) The moment equilibrium equation (tMg9)' + Mg + r2"Pe - PQ + rqg = 0 and the dual compatibility equation (reer)! + egg = rz"yg -PAg = 0 then give two simultaneous second order ODEs for and 1. With the constant of integration in T(r), the solution of the two ODEs can satisfy the three stress boundary conditions AP; = ANeg 4Meg = 0 at one edge but only two more-49- stress conditions at a second edge as the shell must be in overall equilibrium. The sixth condition for a shell with two edges can be prescribed in terms of the circumferential displacement Aug = 0 with no restriction.-50- 5. Reduction for Axisymmetric Problems of Nonshallow Shells 1. Reduction of the Axisymmetric Bending Problem The reduction is similar to the shallow shell case. We begin with the first integral expressing the axial stress resultant mF NepSinE + Q,c0SE = Ary ei cg. Qo Pv(t)r(t)a(t)dt) "a and designate the radial resultant by og/r so that a eee N, cose ‘+ Q-sing 2 a, The corresponding dual relations are ; Ce ~KggSing + Agcose eT wkggCOSE B AQSINE = T og The four relations may be inverted to give Nee = (rV¥)sing + ogcosé, Qc = (rV)cose - egsing =rkgg = Cpsing + 9gcose, The = Cgcosé - ogsing The differential equation of force equilibrium in the radial direction then gives Nye = 8g + roe, (ye-51- and the dual compatibility equation gives *8 -« ain ee” Next, we use the stress strain relations to express the remaining stress and strain measures in terms of $g and ¢g: . = ptt gt 420 i } Mee = D(x + Yyk&gq) = DES og + z-(coséd, + singC,)} Mog = D(xop + Yq) = -DEZ of + L(coseon + singCg)} 00 0 * “boo? = “PIG oR + plcoséog B P, = Dpdg = = o(ogsing - Cycose) v = A(Nog = %eNgg) = A CL og + roy - 2Sfeossog + sing (rv) D “ee sie lane i 7 ALS % - verpy + plcoséeg + sing(rgv)]} The moment equilibrium equation which has not been used up to this point may now be written in terms of og and og 2 (vpDr'/ay' z' HY - iaeray lhe ~ qa "s (vp0z"/a)'rt(rV) = rag —wr7ay Tray-52- The dual compatibility equation may also be written in terms of ég and g SAC ayy, 2 Bray J° * Thr7ay 9B vgAz"(PV)/a]! Cpe! 2p)! ae ed = ay a rn ~ Dsr'apy + oe = (rv) For shells with uniform thickness and material properties, these two equations simplify slightly to read rtfay' a +s “tray le * Tar7ay 8 ON . oe = (rv) SE - 3h + vy feed +a - beset opy + faerie i : The fourth order system of ODEs and one additional constant of integration in the expression for rV allow us to satisfy at most five stress boundary condi- tions for any two constant €, edges as the shell must be in overall equilibrium,~53- One of the six appropriate boundary conditions for the shell frustum must be prescribed in terms of the axial displacement component. With ul, = (singu, + coséw)' = ( = Dee + PY, + Pop where Eee and vz are known in terms of % and ops we have u(r) = Uo + f84 (2tege rly, + r'og)déy 5 The appropriate form of the stress and displacement boundary conditions along a given edge £, = &% can be found in section (4) of Chapter 2 of these notes. Finally, the constant Cg is associated with a circumferentially nonperio- dic displacement field (see Section (4) of Chapter (3)). It should be set to zero for shells complete in the circumferential direction.~54- 2. Circular Cylindrical Shells For a circular cylindrical shell with a midsurface radius ro, the two simultaneous ODEs for og and og simplifies considerably. For a shell of uniform thickness and material properties, they take the form 24 fog . Fond J ro 1 200 +a 4B = %5CPQN)' - PoP where we &) = Z/fg So that a= fg, sing = 1 and cose = 0. The corresponding auxiliary equations for stress and strain measures become roNee = (rgV), r -4 al Nee Be Neo = Fp *B t ToPH 0%, The two ODEs for +, and % are of constant coefficients; the complementary solutions are in terms of exponential functions. To motivate subsequent de- velopments, we observe that the two ODEs may be combined into a single complex ODE of the same order 0, x= Ge = a 2x = Tore ALvg(rg¥)' = ry] where 4 is a root of the quadratic equation r Al D eee a ya +r 055+ and x = 5 + SE With ro2/DA = O(rg2/h2) >> 1, A is usually a complex number In general, it is easier to work with a single second order complex ODE than with a fourth order system, especially in the case of equations with variable coeffi- cients. For the classical theory with Ag = Dp = 0, we have 4 = + YA/D i and Cvsrapyt 3PH with X= 9, + iA a, and 8° = rg/VDA. The complementary solution for x may be taken in the form 7 3 3 = UF BE 3 x, = Lebei//2 (0) cos FFL. cy gin MLy + 0861/72 t0, cos 4 - c, sin M15] 2 2 2 2 a. 3 3 —— + LeBE/¥2 {c, cos “2 - Cy sin Hedy + eBE1/2 fc, cos HL 4 5 sin ed B vB Bi Fy, Zz xe is significant only near an edge of the shell if B2/rg = a/tone?y 2B» 1. We now consider three specific problems within the framework of the classical theory: I. Axial End Loads: The shell is subject to no surface loads so that qz = py = py = 0. At the ends z = 0 and z = 2, the edges experience no transverse shear resultants and no bending moments so that Mez = Qe = 0. The two ends are subject to equal and opposite axial stress resultants of magnitude No. For this problem we have _ fz 2 (rv) Fz (Y) = yee Neg * Thy 7 Zito-56- The ODE for x is homogeneous; so are the four boundary conditions for $, and , (as we have Cy = 0 for a complete shell of revolution): Mee = Mog = Qe = Nea = 0 (0 <6 < 2/r) The boundary conditions on Neg at the two ends are satisfied by taking Fy = 2nrgNo .9 that Neg(&,) = No (0 < & < t/Po) Having determined all stress and strain measures for the classical theory, we may then calculate the displacement components from the strain-displacement relations if we wish. For a circular cylindrical shell, the strain displacement relations for the axisymmetric bending problem take the form Up = Pot ear oe = > OB fy uz = Uo +f) (roces)dé, (= Ue) 2 where Ug is a constant of integration. We see from these relations that the BVP for 4, and % continues to admit only a trivial solution if the moment free condition Mee = 0 is replaced by the condition of no rotation - y= Oat one or both edges. On the other hand, if the condition of no transverse shear resultant, Qe = 0, is replaced by the condition of no radial displacement at one or both edges, ¢, and #, wi11 no longer vanish identically asil. is an inhomogeneous boundary condition. Uniform Radial Surface Load Distribution: For this case, we have Py = 4c = O and py = po throughout the shell. If both shell edges are free of traction, we have also Nee = (rV)/ro = Fz/2nrg = Oat z= 0 and z = & so that Fy = 0. The boundary value problem for 5 and * is again homogeneous; it admits only the trivial solution 3 8 = 0 for 0< &, < £/fo). Correspondingly, we have 1 Neg = Qe = Mee = Moo = 0, Noo = PoPo throughout the shell. Again %3 and % would not vanish identically if the condition Q; = 0 at one or both edge is replaced by up = 0. For 21/2 2 = Ba7/ry” = 22/(DArG) >> 1, the contribution from 9, and @, in this case is in the nature of a boundary layer phenomenon. A Semi-infinite Shell with Edge Moment and Transverse Edge Load: With Pr = PH = Ge = Oand Nee = 0 at the edges, we have (rV) = Fz/2n = 0. For the stress and displacement components to remain bounded at infinity, we take c = C, = 0 in the complementary solution X¢ so that tg = eB/72 (C3 cos 3/5 eBE1//2 cc, cos-58- It remains to choose C, and C, to satisfy the edge conditions . - D 5 = 0: Mee = 5 These two conditions requires that C, and C, be the solution of elem, /Boy = -r40, - A . PoMoy 2. ‘ C, = - [40s «= 28 + [Rr The expressions for ¢, and 6, then becone > Be v2 Be os we 7 P5,/Y2 CVE My + 7 '4Q9)cos a + = pe be //e_ 1 Bey 1 i eg = Bem PQocos on (72 My + 3 Fy %y)sin Near the edge £; = 0 of the shell, we have for the direct stress %D and bending stress og the following order of magnitude estimates: oR 3 3M = 078) = o(Ftaza, 2 Me) Moo Doe Mo roQ = OS) = OB) = 0, =) rh 2” Bh2 It follows that Sw 082) « (Sy = (1) B To YA ro so that the direct and bending stresses near the edge are of comparable magnitude.-59- 3. A Second Order Complex Equation for Spherical, Conical and Torodal Shells For a spherical shell with midsurface radius a, we take r = asiné, and Zz acosé, so that a = a and & = mE). For shells with uniform thickness and material properties, the two ODEs for 4, and 6, become " ol D0 og + cotey9, - {esc°£, - [1-v, - pP]bog + H eg = FelE1) A + cobesas - {esc7e, - {ltv, - 92D, - k og = Fo (é1) where D f(E1) = - Cy cote (1-v, - QP : picoter (rv) - age} A Fe(é1) = = (rv) cote, (itv, - 99) - vg(rvy' + 4 Cycotes -[(2+¥5)a2cos PH + a’sing pi] The two second order ODEs may be combined into a single second order ODE for a complex function x = 4g + Ag of the form x" + cotsyx' + [u(uel) - ese2E, Ix = + dfe = F(Ei53A) , - A,Ag _ 9 « 2a DA AQ 2 As al = p> Ys) + A fr ga “yy - 4)? _ Ys-¥y Dp AQ. a pA Ag a won) = (1+ SSP) 1 A Be GR Py =e) where For the classical theory with Dp = Ag = Oand vs = vp = v, we have-60- With the second order ODE for x, it is now not difficult to see that the complementary solution for ¢, and 6, 15 a linear conbination of associated Legendre functions with Xe = tg + Mtg = CPu(c08e,) + C20}(cose, where C, and C, are two complex constants of integration. The parameters 1 and A take on different values depending on the particular theory employed for the analysis. Applications of this result to specific problems, including the problem of a complete spherical shell with equal and opposite point load at the two poles, can be found in [16] and elsewhere. For a conical shell with a constant slope angle &, we take r = Le,cosé and z = 25 sing - 2 for 6 < & < & with zq = 2ejsing so that a= 2. For shells of uniform thickness and material properties, the two ODEs for 4, and yin the classical theory (Aq = Dp = 0, % = vs = v) take the form: Lyd gtané Fy fe 2 8B” Ue, fe(éi) Lyd atane . _ ete fe zp He tg & £1) cf where tang 2 fe(E1) = =z Cg - pel(r¥) - 26,92] e(51) eSB De, tt ) 19e 1 Te, 8 BL Fe(E,) = 2828 (ev) + vetane (" y a1 s En ~ [(2#vs)pH + §,Ph]2° cose The two ODEs can be combined into a single ODE of the same order for x= cn + ivA7D %% in the form-61- 1 1 ; Btane [RK xo te xt Se x t 1 SS x = fe(E1) + iff fe(é1) o i eR 8 D The complementary solution of the above ODE is easily seen to be in terms of Bessel functions: = Ayd2(8VTEq) + AgY2(8vTE]) = [Cyber ,(v@]) + Cbein(sv@;) + Caker2(BvE]) + Cykei2(87%;)] + i [Cober,(av&;) - Cybei2(BvE;) + Cykero(BVEj) - Cskei2(8¥E])] where 8° = 4ztang/¥DA >> 1 for shell behavior. Consider next a toroidal shell with rsa + bsing, 2a +s, z= - beosé, = - bc, (0 < &, < 2n) so that a=b With A = b/a <1, the two ODEs for 9g and 4g become 1 ac ‘ n2c2 n252 ds ee Sy ey Teas, “B” “(i+as;)@ ~ D (1#As;)2 —D (1+as,)2-'8 bas, fel) -— 1 4 = A(l+#as;) 8 &"t 2e2 252 “ A, a2e2 Ag st As} 3 - (ey —t. +. —— 9 8° Ths, 8 Gas *® Gos,e *’s Tins) bas) ~ A(T#AS])~62- where 12s y¢ D i re S11 : , F,(61) = ¢ = = 8) = yy ot - PAE) _ bq 3 (Ae Bi (1+as1)2 bibs; DLAs; Aa ds ye) Fa(Er) = (r¥) ‘Gas? qa sre, AS) begrcy -v, (rv) + “Tas; S l+as A(1+As)) 2 ae: ~ [(2#v,)b7¢, py + ab(1+A8) )py J For the classical theory with Ag = DD = 0 and Aes for $, and ¢_ may be combined to get a single ODE for the complex function B B wy er Mey? ; > v20A 4S) x" + Ko oti x Tisy (Tis3) WK b2 T+asy = fel6,) + vfe(E,) =v, the two equations X= og + udp: where s, = sing,, c, = cosé, and An exact solution of the homogeneous equation for x in terms of known Function does not seem possible. Furthermore, the second order ODE has a simple turning point at €, = 0 and £, = 7 where sing, = 0, Applications Langer's method for an asymptotic solution for x is appropriate whenever we have 82 = Ab/YDA = b2/VDAaZ >> 1, Such an asymptotic solution and the use of this solution for a number of practical problems can be found in [17].-63- 4. Finite Difference Solutions The reduction of the shell equations for axisymmetric bending (and stretching) problems to the canonical form of two simultaneous second order ODEs for ¢, and #, ts evidently a very important step in our effort to under- stand the shell behavior for this class of problems. For example, the canoni- cal form has enabled us to deduce the exact solution of certain problems in terms of known functions. It has also enabled us to bring out the singular perturbation structure of (axisymmetric bending) shell problems which is not at all obvious from the original form of the shell equations. This special struc- ture of the canonical ODEs in turn allows us to bring a wide range of asympto- tic techniques such as the WKB method, Langer's method, etc., to bear on our problems for accurate approximate (asymptotic) solutions*. In addition to all these significant features, the canonical form is an attractive arrangement for efficient numerical solutions of the relevant boundary value problems. For our discussion of an efficient numerical solution process here, it suffices to consider shells with uniform thickness and material properties and to take the two canonical ODEs in their Liouville normal form, o° - [o(x) + vpn(x)] - 82t(x)@ = fy (x @ ~ [o(x) ~ vgn(x)] + 82x(x)@ = f(x) where o(x), n(x), t(x)s f,(x) and f,(x) are known functions of the dimension- less independent variable x, ( )" indicates differentiation with respect to x 2 and 8° is the Reissner number for the shell. is large compared to unity if the shell is not plate-like. The two ODEs are supplemented by two appropriate * The asymptotic solution process for shallow shells of Chapter 4 (Section (6)), can be extended to the non-shallow shell case. See also [17,18]-64- boundary conditions at each of the shell edges, x=xj and x=X9. We take these in the form x = xq apie’ + gid = Cot, agie” + bei? = Cp X= Xo! azo¢" + Bgot = Cgo age” + boo? = Cao where ag}, etc. are known constants. The form of these four boundary condi- tions includes the appropriate stress and/or displacement boundary conditions as special cases. A more general set of boundary conditions would be needed for an elastically supported edge or for a curved tube with a closed cross sec- tion. For the purpose of a finite difference solution, we introduce a mesh of equal spacings {x,, x, +++s xn) for the interval [x4,x,] with x, = x4, XN = Xpo Xk = xq + KA and & = (xq-x4)/(NeL). For each of the interior mesh points x) 5 » XN-1} we have a finite difference analogue of the ODEs con- structed by replacing the second derivative of a function y(x) at a point x = xj by the finite difference expression [vj41 - 24j + ¥j-1]/42 where vj = v(Xj)+ We write the two finite difference equations as a single equation for the vector unknown Yj = [4j, ¢j]": Yoel + Bh Yn + Yntd = Dns (N= 2435 waey Nel) (5.1) where -2- A? (oK+ vp nk) -37 A cK ath By = ’ On = 87a 2 = a? (op=vgny) agp At each of the two end points, we have also a finite difference analogue of the two boundary conditions. There are a number of different ways for constructing-65- the analogues. For our purpose, it suffices to take the simplest approach (which may not be optimal) by using a forward difference formula for the derivatives at x = x; and a backward difference formula for the derivatives at X = Xo. In that case, the two boundary conditions at x = xj may be written a single equation for Y, of the form BLY, + CY, = D, (5.2) while the two boundary conditions at x = xo may be written as ANYy-1 + BNYN = DN (5.3) where the actual form of the coefficients 8,, C,, D,, By Ay and Dy depends on the difference formulas used. In fact, more elaborate formulas for the first derivatives may bring in additional unknowns, Y,, Y,, se. for (542) and Yy.2 Yyegs «e+ for (5.3). Equations (5.1) - (5.3) constitute a system of 2N linear equations for the 2N unknowns {¢,, +++ gy} and (@,, 242, ey}. The block tridiagonal form of the system enables us to solve it very efficiently by forward elimination and back substitution in a way that takes advantage of the special kind of sparse structure of the coefficient matrix. Equation (5.2) may be first solved for Y to get ic, and 0, = BiD,. The result may then be used to eliminate Y with Cc from equation (5.1) with n=2 to get Be ap yet 5 eae s where C, = -(C,+B,)-!C, and D, = (€,+8,)-1(D,-D,)~66- By repeating the same process, we get Yn = Cy Ynea + Dn (n=1,2, 265 Nel) (5.4 where Cy = -(Cp-1+ By)~!Cq and Dq = (Cheat Bp)-*(Dp-Dp_y)+ Finally, we have from (5.3) Yw = Dy = (ANCN-1+ BN)“2(DN = ANON-1) which gives the solution for 9(x,) and 2(x,). We back-substitute through (5.4 sees Y,« The solution for n = N-1, N-2, «e+, to get successively Y 1 New? "Nr? process involves only the inversion of N (2x2) matrices and the multiplication of two (2x2) matrices 2N times. The operation counts is O(N) which is to be compared with the O(N*) count if the coefficient matrix of the linear system is not sparse (which would be the case if we had sought a numerical solution of the original system of shell equations. In addition, the storage requirement is also an order of magnitude less than that for the shell equations before the reduction to canonical form. The above finite difference solution process is only the simplest (but crude) method for a numerical solution. In addition to the possibility of a more refined difference analogue based on finite difference formulas of higher order accuracy or some spline approximation, we may improve the efficiency of the numerical scheme by using a variable mesh with more mesh points (of smaller spacing) inside the boundary layer region of pronounced effect of edge bending. Here, the availability of some exact solutions makes it possible for us to estimate the boundary layer width and hence the appropriate mesh size for our problem, etc. The exact solutions also offer a bench mark for the accuracy of the numerical solutions. Finally, it should be mentioned that the reduction of-67- operation counts by two orders of magnitude is even more important for problems involving finite deformations. Numerical solution scheme for nonlinear BVP in ODE generally require an iterative solution of a repeated modified linear sys- tem. As we shall see later, axisymmetric bending problems involving finite de- deformations may also reduce to a canonical form with nearly the same structure.~68- Part III - Laterally Loaded Shells of Revolution 6 - Sinusoidal Stress Distributions in shells _of Revolution 1. Intrinsic Formulation of Shell Problems In many applications, the surface Toad components pes Pe and py for shells of revolution vary in the circumferential direction in the form of sin and/or cos. One example is a dome-shaped shell of revolution in its "face- side” position so that the gravity vector is perpendicular to the shell's axis of revolution. With no loss in generality, we may let iy be in the direction of the gravity vector. In that case, the surface load vector p is given by p = pghty where p is the mass density of the shell material, g is the gravitational acceleration and h is the shell thickness. It follows that Pe = Prte = pghcosécos#, Pg = Prt, = - pghsing Pp = pen = - pghsingcose , where € is the meridional slope angle, i.e., the angle between the meridional tangent and the horizon, cosé = r'/a and sing = z'/a with a2 = (r')2 + (2')2. Ina linear theory of shell, the stress and strain distributions of a shell of revolution are proportional to sine and/or cos@ whenever the surface load components are proportional to sino and/or cose. With {(PePyt = {P_(E1)>B,(E1)}cose, Py = By(1)sine, (6.1) it can be verified that the stress resultants and stress couples are of the form (ees Nogs Qe Mees Nggs Pg} = {Ne (Ex)s seer Bol Ei) }e08e ge (6.2) (Neg Noge Qgr Megs Moes Pz} = (NeglEr)s ears Be(E1)}sing-69- Correspondingly, we have { € » ¥_(51)}cose Keer “oor or Seer Soar Ved = ReelEn)s (eg Kage go Eger Sag? Yo! = CEeg(Ei)> ve Fo(Er)?sino for the strain measures. Other surface loads which are of the form (6.1) include the conventional approximate expressions for wind loads. Shells of revolution without surface loads but subject to edge loads may also develop stress and strain distributions with a sing/cos8 type of circumferentia variations. Examples of such edge loads include the case of a side force and/or a tilting moment applied to a rigid plug with the edge of the shell welded to the plug. Shells of revolution having stress and strain measures which vary like sine/cose in the circumferential direction are called laterally loaded shells of revolution. For laterally loaded shells of revolution with stress, strain and load measures of the form (6.1) - (6.3), the equilibrium equations are effectively six linear ordinary differential equations for the a dependent portions of these shell quantities. For classical shell theory with Pz = Pe» =0, and dg = 49 = qn = 0, we have (lige)! + aig - acos ENg, E rota, + Pai E (rite)! - agg + acoséNo, - 2's + rap, = 0 (rQ,)' +a, + rele + 2'Ng, + rab, = 0 (rige)' + alge - acoseMgg - raQe = 0 (rMg9)' - aMgg + acoséMge - raQy = 0 ra(Neg - Nec) ~ 2'Meg + E'rMge = 0-70- These six equilibrium equations along with the six dual compatibility equations and the stress strain relations 4 2 on + < 7 1 oo. Hig = Moe = 5 D(1-) (Rep + Roe) 8a” 8 Nee)» &e Neg + Ys Eco = Bog = ¢ A(I-s)(Hieo + Hine) form a system of twenty equations for the twenty unknowns, Nez» Nog, Nees Neos des Qos Mees Moos Megs Mge and their dual strain measures. Unlike the axisym- metric stress distribution case, they do not uncouple into two separate groups. As such, the problem of laterally loaded shells of revolution is substantially more complicated than the axisymmetric load problem.eile 2. First Integrals Just as the axisymmetric problem, the equilibrium equations for laterally loaded shells of revolution have two first integrals. First, we form the combination 0 = cosé{(rNec)' +... + ade] - [(rligg)' + «2. + rae] - siné[ (rae)! ... + rain] of the three force equilibrium equations to get [r(coséNee - sinéQe - Neg)]' + [r(Becose - By - Basing)]' = 0 or rleoseWeg = singQe - Neg] = 4 Py(E1) (6.6a) with nf’! (Becosé - Bo - Bysing)arde, ze 5 where Po is a constant of integration. The existence of this first integral is expected from the requirement that the shell be in overall force equilibrium: an ts re + J f° [(rNg)'+ (aNg) + rapjdede = 0 7a i As aNg is periodic in 9 with period 27, it follows that ar, 8 ore, + Cf rNede] +f f rapde de = 0 0 g O86, i i The left side has only one nonvanishing component in the ix direction; thus overall force equilibrium for a laterally loaded shell of revolution leads to only one nontrivial condition (6.6) of vanishing resultant side force in that direction.-72- Similarly, overall moment equilibrium for a laterally loaded shell of revolution gives only one nontrivial condition on the stress resultants and couples. For classical shell theory with Pe = Pg = 0 (and yg = Ye = 0), this condition takes the form r[Meg - COSEMeg - rcoséQe -rsingNee] (6.7a) with Ary = J51 [(rcoss + zsing)by + (rsing - zcosé)Be + zBg]ardé, (6.7) Evidently, Ty is a resultant tilting moment turning about the y-axis. By the static geometric duality, we now know without a separate derivation the existence of two first integrals of the differential equations of compati- bility. For classical shell theory with ye = yg = 0, they take the form -r[cosézgg + Sinéhe + Roe] = 2 ay (6.8) rlEg9 + coséege - rcoséig + rsingigg] = - 2 a, + . Uy (6.9) In these, the quantities a, and Uy are constants of integration. As we shall see in the next section, these quantities have the interpretation of relative edge rotation and displacement, i.e., dislocations of the Volterra type for shells. With the four first integrals (6.6), (6.7), (6.8) and (6.9), the system of shell equations for classical shell theory is effectively reduced from an eighth order to a fourth order system. We will in section (4) make use of these first integrals to reduce this system to two coupled second order equa- tions of a form which is very similar to that of the Reissner-Meissner-Reissner system for axisymmetric bending (and stretching) of shells of revlution,-73- 3. Non-periodic Displacement Fields With a view toward an interpretation of the constants of integration 2 and Uy in (6.8) and (6.9), we recall from plate bending and generalized plane stress problems that strain measures of a laterally loaded flat plate in bending and extension with a sine and/or cos dependence on the polar angle 6 may correspond to a nonperiodic displacement with a coso and/or dosing dependence. This suggests that we consider nonperiodic displacement fields of the form {ues Ws ,} eh = ti,, W, F_tcos0 + {i,, W, $,}esine {Ups 92 0) = {Ug, 3g, B)sing + flig, $5, Ghacose where terms of the form f and f are functions of € only. Upon substituting the above expressions into the linear strain-displacement relations for a shel of revolution and requiring the resulting strain measures be periodic in 6, we get coséuy + (zcosé - rsing) oy, w= -singuy ~ (rcosé ~ zsiné) dy, lo * 2% where $9 and ug are arbitrary constants.-74- The corresponding strain measures are . ele . ee a na erate te & Zoe = pig + cossii, - singiv + 269 + Ug] Boe = Hid + costii - ri - costug - (zcost - rsing)eo] ine r (Gg - Ug) - (cose + 2 singso) From these follow the relations COSER gg + Zoe + Singlg = 2 4 ee + Kee 8 =F t0 r(Sinékgq - coséig) + Ege + cossey, = 2 (ug + 240) A comparison of (6.8) and (6.9) with the above relations leads to the identifica- tions fo=- Fee Yo = Fe ly (6.11) which allow us to interpret 2, and Uy as Volterra type dislocations. Nonperiodic displacement fields are not permissable for shells of revolution which are com- plete in the circumferential direction; hence, the constants % and Uy must-15- be set to zero for such a shell. In contrast, nonperiodic displacement fields are needed in the solution of a slit shell. A physically meaningful problem of this type is the asymmetric twisting of a ring shell segment formulated in [6]. Once the strain and curvature measures of the shell are known, the reduced strain displacement relations (6.10) may be solved as a system of first order ODE's for Uz, Up, W, 3g, dg and S, The system is not overdetermined as the strain and curvature change measures satisfy the compatibility conditions (6.8) and (6.9). The solution for the £,-dependent portion of the displacement components for the classical theory (with ye = vg = 0) may be given in the form 3. = fel ge= J Reeodén, Go = egg - cose(3e - oe 2) B= rhe - sine(3_ - oe %) Ug = rgg - Z3q ~ Ze(Uy - 2x) + [5 (2ege - cosseec)ader (6.12) = rhe = Egg -rsingde - cose(dig - 4 uy) + (rsing - zcoss) 2 ay = r2igq - (zsing + reose) (3, - 4.2%). + sing[régg - 4 uy + [51 (zéee - coset ee)ade,-16- 4. Reduction to Two Simultaneous Equations We re-write (6.6a) and (6.8) as coséNice - singQe = oS P(E) oh igo and cosékgg + singig = - Fay - dy respectively. The reduction procedure for the axisymmetric bending problem sug- gests that we take cosélics -singQ, and coszg9 + single as the two primary dependent variables. We adopt the equivalent choices by setting (6.13) Neg =p og = With (6.13), we now consider the sixth equilbrium equation, the stress strain relation for Mcg = Mg and their static geometric duals as four linear algebraic 5 8 equations for Meg = Mec, Ege = Egg, Nog and Reg in terms of 6 and . We solve these equations to get - - A(1+vs) 1 = lee 1 D(1-vb) > Fee = Fog = Tapeay [@ - g eD(1-v) 0], Ne= ao 7*- 4 re r(1te2) Te? 1-22 D(1-¥) 1 - ep 1, , Alleys) Meg = —— + % pA(1tv.)o], =—O2 26+ £6 So” F(te2) Bas CNET. “so " Treg 7 re? where p = (1/Rg) - (1/Re) and «2 = 4 DA(1-v2)p2. With the help of (6.14) and the remaining four stress strain relations, the four first integrals, the second equilibrium equation (with Qo eliminated by way of the fifth equilibrium equation) and the dual compatibility equations may be treated as six linear algebraic equations for the six unknowns Qc, Mzc, Meg and-17- their duals in terms of @, #', ¢ and 4’ and load terms. We emphasize here that no higher (than first) derivatives of @ and § appear in the solution of this system for the three stress and three strain measures. Also, Nez. Neg» Keg and eggs obtained from the inverted stress strain relations i vse eo 5. oot stag N ay » Nog = eS eee , etc. 6s AI? 98 * “A(-v2) and Qj and ig, obtained from the fifth equilibrium and compatibility, are also known in terms of ¢, ¢', ®, @' and load terms alone. The only two remaining equations in the whole system of shell equations for laterally loaded shells of revolution are the fourth equilibrium equation and the dual compatibility equation. Upon expressing all stress and strain measures in these two first order differential equations in terms of 4, $', , 6! and load terms by way of the results obtained above, we get two simultaneous second order ODE's for » and @ in the form AGS) ow ag on Dv) i og (6.15) where 1/Rp = - 2'/ra =~ Sing/r, <2 = O(DA/R2) with R = min(Rg,Rz} and where only o', , ¢', 6 and load terms appear in f, and f,. The exact equations (6.15) for 6 and » are rather complicated. We limit ourselves here to the case DA = O(h2) and vg = vp, and to shell properties and loadings varying significantly only over a distance of the order 2 >> h. To make use of these restrictions to simplify the ODE's for 4 and 6, we first solve (6.15) for 9" and 9" to get yo" = fy - DUbve) (r+ ey? es f, - Duley) ¢ 1" Reet)?”-78- where e2 = O(DA/RE) We then omit all h/R and h2/22 terms from these equations to get the following two differential equations for 4 and 6: 2 ‘yay! » 4 {Or/a)" 4s r {(1-vp)Dr'/a} 2'y2 z a + Gos GR) = arze +20) Wt - ora 1 Par czy, Ty ety 2 MD BK eg ey YY agg! y Sx (ryp)a core SS Dt rl opp Ge Gr te) - Se + (6.16) a 2 ((Itvs)Ar'/a}! 12 ‘ an Gebel” or = ta Ey?» Ss Fy + dase ket yay |W rte | 2 SR Pe a Ty cat Pe (ev)? Sonepat Dt mea Cpa Gere! + 22") 2 + Ar2Be)! aeeay' (6.17) “aaah + (1 = vs)ar'Be. when z = 0, these equations reduce to the exact equations for a flat plate. When r = constant, they differ from the exact equations for a homogeneous circular cylindrical shell of constant thickness only by terms of order h/Ri compared to other terms in the same equc-ions. When z #0 and r is not a constant, it is pos- sible to effect a further reduction to a single second order equation for a com- plex stress function for spherical, conical and toroidal shells (with constant Re). The method of reduction is the same as that for the corresponding axisym- metric problems. For simplicity, we have omitted the details leading up to (6.16) and (6.17) (which are remarkably like the Reissner-Meissner-Reissner equations for axisymmetric bending of shells of revolution) as well as the auxiliary equations for the stress and strain measures in terms of 9 and ¢, These details can be found in [6,7] where applications of the results are also given. A simpler but more complete reduction of the corresponding shallow shell problem and some appli- cations of the results will be discussed in the next chapter.-79- 7 - laterally Loaded Shallow Shells of Revolution 1. Governing Differential Equations With the same characterization of shallow shells of revolution given previously for axisymmetric stress distribution problems, the strain dis- placement relations for laterally loaded shells of revolution in the classical theory simplify to en a Fg = WL - 5, He 3, tw - lee oe - vu 1 Egg pL, +O, - ra- et (z-rz') 4] (7.2) where we have taken ¢ =r so that ( )' = d( )/dr in this chapter. Consistent with the principle of virtual work, the corresponding differential equations of force and moment equilibrium for the r-dependent portion of the stress resultants and couples are (PNee) + Nog ~ Nog * 78 Ne, - Noe = 0 (7.4)-80- for the case of Pj = 0 and no surface moment load so that Ge = Ge = Gy = 0. With 7, = Yq = 0 for the classical theory, the r-dependent portions of the strain measures satisfy six compatibility equations which are the static-geometric duals of the equilibrium equations. The stress and strain measures are related by the stress strain relations Mee = D(Kee + VyRyg)s — Mgg = DCE oe * “bFee) (7.5) 1 ao hog = Moe = 3 DUI) (Reg + Rye) yy = A(Nog - vgNer)» eee = A(Ne + VgNoo) along with ye = yg = 0 and Pg = 0 for the classical shell theory. The four appropriate stress boundary conditions at an r = r* edge now take the form = AQ = aM. = 0 (7.7) where e = Qc + Meo/r. Because of the overall equilibrium requirement , boundary conditions at both edges of a shell frustum may not all be prescribed in terms of stress measures. The four appropriate displacement conditions at an r = r* edge are =0 (7.8) They imply four strain boundary conditions which are the static-geometric duals of the four stress boundary conditions. For overall compatibility, all boundary conditions at the two edges of shell frustum may not all be prescribed~81- in terms of the strain measures. Hence, a combination of stress and strain (or displacement) measures are to be prescribed at each edge of the shell frustum. The four first integral will in general determine the admissibility of a particular set of mixed edge conditions.-82- 2. Reduction to Two Simultaneous Differential Equations As in the case of nonshallow shells, overall equilibrium consideration leads to two first integrals of the differential equations of equilibrium: r amep Nee = Neg = ap P,(r), (7.9) B eB - wy ek Mee - Meg - r(Q, +z Nee) ahar [Ty - 2] (7.10) with P. a f° (5, - 2'By - By)rar = P. {" (8, - B,)rdr x 7 TA AP, = 2)Pn - Poirdr = Po - ty (Pe - Pair i i (7.11) TyeTy-* { [(z-r2')B, = 2By - (zz' + r)B, Jer where Pg and Ty are two constants of integration. The underlined terms should be omitted for lateral load problems to be consistent with the definition of a shallow shell, since with pz = 0, we have Dp = -z'pp and pe = Pp for shallow shells. By the static-geometric duality, we have also two first integrals of the compatibility equations of the form. oe 68 ee 7 ap xe (7.12) fgg + Sg 7 (Ag - 2Ka9) = 4. tuy = nx] (7.13) where 2 and Uy are two constants of integration whose geometrical interpreta- tion can be found in chapter (6) section (3). The relations (7.12) and (7.13) can be verified by direct substitution of the expressions for the strain meas- ures in terms of the displacement components. We will make use of these first integrals to reduce the system of equations for laterally loaded shallow shells of revolution to a canonical form similar to that for the axisymmetric bending problem.-83- Analogous to the axisymmetric bending case, we set N =2 . af Negtp Foe t (7.14) so that we have from the first integrals (7.9) and (7.12) ii 1 Px z 1 7 s(o+2), = Ry te (Ot) (7.15) Next, we get from the sixth equilibrium equation and the dual compatibility equation N ao x ale ec * (7.16) and from the second equilibrium equation and its dual compatibility equation ie otal oA ed Ngg =e +40 + rpg, eer ete (7.17) The stress strain relations then give - ot. é ye $ Beg t Ege = AlIt¥g) Bs Meg = Mog = D(1-yy) F (7.18) - ; Oy itz, = - DLo'+ (1+y,) tev, 4, (7.19) Mo, = - OLy, 4 (4) $+ 24 ge = ~ Lye b) rt ar é = ALS! + (1ev,) 2- vy, 2X + wig] 88 s) F 7 Ys ar * TPe A 1 eo, Px = = AL-vcot # (Ie) 2+ % - veriel (7.20)-84- The first integrals (7.10) and (7.13) then give - a, rd, = - Oe +24 ys Px as ' py 2x _ ove Ke rb] - 2'9 + (z2'r) B- ra, = - ALS Ss TF ain while Qo and X¢ are given in terms of 4 and @ by the fifth equilibrium equation and the dual compatibility equation: 5 Q. rd, = Oo +29 +, (7.22) : P. a = Ate +26 4% - verb] We now substitute the expressions for Mec, Mag, Maz and Qe into the fourth equilibrium equation to get zoe Mx | (rztez) Px a Ty #- Fy (yy) Sp + Mee (7.23) From the dual compatibility equation, we get a second equation for 4 and 9: walg 4 Zine tee et iee . Px rz'-z) % 1 Uy 5 3 - (ny) 3 - EEA) vsPe - (Bg) Supplemented by appropriate boundary conditions, the two second order differen- (7.24 tial equations (7.23) and (7.24) determine o(r) and g(r). The stress and strain measures of the shell can then be computed by way of equations (7.14) = (7.22). The displacement components can also be obtained by integrating (7.1) and (7.2).=85- 3. Shallow Spherical Cap in a Face-Side Position With the gravity vector in the direction of the x-axis and the axis of revolution of the shell in the direction of the z-axis, we have for a shallow spherical shell of uniform mass density, p, and thickness, h, p = oghiy so that P, = eghcossa, Py = - eghsine, P, = ~eghz'cose (7.25) where z‘ = £9(r/rg) and go and rg are known constants. For a spherical cap extending over the region 0 By = PoBgp + ToBaT (7.32) where ro 1 [ t(rop)'s (roz)'} = dr i {B B BP? 6T 7 (AL=ve) + (levee Op + 2a] + 2 (nop) 3dr Boy = of CtAL-vgot + (Lev) 0,1 + 2 (ro) Dar are independent of Py and Tg. For a given shell, the flexibility coefficients Bop, Bet» Bgp and Bgt may be determined once and for all. Given Po and To, the flexibility relations (7.32) determine sy and By. Conversely, given 6x and fy, they serve to determine the force Po and moment Tp needed to produce them. For sufficiently thin shells so that their Reissner's number 8 is large compared to unity, it can be shown that the actual integration may be avoided in the evaluation of the flexibility coefficients (see [7]).-90- Part IV - Shells of Revolution Under Arbitrary Loading ———EEeEE—E—S—EEe—— 8. Shallow Shells of Revolution Shell Equations Under the order of magnitude relations of the shallow shell theory adopted in chapter 4, the strain displacement relations for shells of revolution given in section (2) of chapter 2 simplify considerably. With &) =r , we have yet tt Ye 7% =a = 6! =i - = 1@: Oerpel’ TEECEE Lc cfeel® CHAM INR oc Wad ci mnt came aan OORT Siete fT Loot gon aicatene ag EW" + 2'8,) sul etw = lt eu 2) cee 7 MET ; fgg 7 Eug + up > 2'¥ cep we = Aq; - ay) +0 colmeaa Cui , Seg re” Me . The corresponding equations of equilibrium for the stress measures are GEN," +N,e “Nog HEP, =O > (Neg) NG +N, eo TPE Bea 2 8 (Q,) 495 tr2"N,, +2 Noo-91- (rtfg,)" Mp ~Myg 2'Fy ~F% 1 Gey) HMSg HMyg F2"Fe ~ 295 (xP,)' + PS + Fg Nye) = 0 where, for simplicity, we have assumed the absence of surface moment load components so that 4, = 4, 74,70 + These equilibrium equations are consistent with our strain displacement relations for shallow shells of revolution so that the virtual work axiom is satisfied. The equilibrium equations and strain displacement relations are supplemented by the twelve stress strain relations given in chapter 2. With suitably prescribed boundary conditions, this set of shallow shell equations defines a well~posed boundary value problem in linear PDE Along an xr = constant edge, the stress boundary conditions are the same as those for nonshallow shells. For the classical shell theory, the appropriate contracted stress boundary conditions consistent with the principle of virtual work are e N, ON, = = AM. =0 ANge = aNeg = 8G, = AMEE with e Qt rt, . Evidently the boundary condition on the effec- e i See i : tive shear resultant N,, = Ney Meg/Bag is replaced by Neg itself The displacement boundary conditions along an r = constant edge are the same as those for nonshallow shells of revolution. For the classical theory, the corresponding strain boundary conditions are-92- e ‘i A e with 2g = Ag — Ege + The condition on Ky, = Kye ~ Cge/Byq for nonshallow shells is replaced by a condition on «, itself. 6e The twelve strain measures defined in terms of the six displacement components satisfy six compatibility equations which are the static geometric duals of the equilibrium equations. The equilibrium equations may be satisfied identically by suitable stress function representations of the stress measures. Except for particular solutions associated with the surface load terms, the stress function representations are the static geometric duals of the strain displacement relations. These additional relations will also be useful in the reduction of the boundary value problem to its canonical form.-93- 2. Reduction of Equations for the Classical Theory 0, we have With Y, = Yq (8.1) 1 ere = (swt + : Ko6 CS ad ) (8.2) These relations are identical to the corresponding relations for the transverse bending of flat plates in polar coordinates. With the three moment equilibrium equations also identical to the flat plate case when 0 , we get from these equations Me otke 6 Gsuceee pC Ag)" (8.3) - -2, yo where V2() = ()"+r 1 jt +2°°(.)" . The expressions for Q, and Q may be used in the third equilibrium equations. However, unlike the case of a flat plate, this equation now involves the stress resultant Np, and Ngg and thus cannot be an equation for w alone. oe As fora flat sheet, the stress resultants may be expressed in terms of the Airy stress function F through the static-geometric dual of the development leading to the expression for K,, » etc-, the only difference being the appearance of inhomogeneous terms associated with the surface load intensities. If p, and Pp, can be expressed in terms of a load potential* L with *The load potential was denoted by -@ for the flat plate case so that L=- for flat plates.then the first two equilibrium equations are satisfied identically by setting (8.4) Now the remaining force equilibrium equation becomes a single fourth order PDE for w and F: spv2v2y = p+ LF] - Ge" + Bet)L (8.5) with Litsel = (tet + Seen + Chet + Ree" (8.6) keeping in mind that z is independent of ® for shells of revolution. The dual development of the compatibility equations gives a second PDE for w and F. The fourth and fifth compatibility equations give ay 2 ALG?F) - G- vou! , a, =-4ivF- G- vgn. 7) 8 & The third compatibility equation may then be written as-95- -av2v2F = L[w3z] - AQ - v VL e (8.8: If 20, the two equations (8.5) and (8.8) uncouple and become pv2v2w = Po +av2v2F = A(1 - v VL which are the governing PDE for plate bending and for generalized plane stress, respectively. The two PDEs (8.5) and (8.8) may be combined into a single fourth order complex PDE for a complex potential X= w+ ivA/DF : etx + Erie) = 5, - PE + tyd YDA - 2. (= vv? In dimensionless coordinates, the term L[X;z] would be multiplied by the Reissner number which is large compared to unity. The PDE for X therefore has the structure of a singular perturbation problem as L is a second order differential operator while ¥?v?_ is a fourth order differential operator.-96- 3. Shallow Spherical Shells For a shallow spherical shell frustum, we have 2' = Eqr/ry 5 r,Srsry- With x= t/t so that r,/t) = <1, the partial me differential equations for X becomes A 1 2 24x = dtp - a?X + is2aX = DIP, -1 wh 2 = = + here g2 = Egtg/VDA and aC) = C) HX Cag +E C)ogg ts the two-dimensional Laplacian in dimensionless polar coordinates. The complementary solution x, of the above equation may be written as a sum of X and x : with X + in?k = 0 a= 0, : aX, + 10°k, ( 8) For a shell frustum complete in the circumferential direction and subject to self-equilibrating surface and/or edge loads, it suffices to consider the case X = X(x)cos(né) . In that case, we have Ry (x) + Eo) where 2 - Fk, t+ in?K = 0 (a = 0,8)-97- or n n = + (A, + dA)x" + GB, + Bx D (ber, (Bx) — ibed, (8x)] oA (c, + i + (D, + 4D,) (ker, (8x) - tkei (6x)] > Bs i? or? Ot » etc. are The eight real constants of integration A, , to be determined by relevant boundary conditions at the circular edges of the shell frustum. Boundary value problems of this type have been treated in [19-24] and will not be discussed further. We only note here that exact solutions are also possible for shallow conical shells and shallow logarithmic shells (with z' = Eory/t ) and that asymptotic methods of solution are applicable to cases where the Reissner number 8 is large compared to unity.-98- 9. Spherical Shells 1. Governing Equations for Spherical Shells As before, we let & be the slope angle between the tangent to a meridian and the horizon. Take the cylindrical coordinates of a point on a (nonshallow) spherical middle surface in the form 2 = - acost ys 8s &) " — and & = 8 , we have ri sacose , a where ( )' = 3( )/88, = 2( -)/2E , and ( )° = 3( )/88 - For the surface coordinates (£,8) , the strain displacement relations for a spherical shell take the form ) = 2 a > Vg Bo, + Glescéw" + uy a 2 (escéog + coté},)-99- coe Les +4) 5 Ke FE ceseto; = cotEb, - &) ce, = taut - w) e,. = E(esckur + cottu, - w) ee ae > & eq = 9 So8ehNy e eee = Lescéut - fr ate > &o¢ q cesebuy cotéu,) + = Lym - = 1¢ . Ag Zl" = 9) > A (csctw" + ¢,) The equilibrium equations take the form GN,,)! + NJ, ~ F'Ngg — FQ, + tap, = 0 (eNzg)' + aNgg + T'Nge - FQq + rap, = 0 5 (xQ,)' + aQ + (Ng +N,,) + rap, = 0 (Mee)! + aMGe — rMMyg — TAQ, - EP, + rag, = 0 (rHg)" + aly + xy, - raQ, + rP, + raqy = 0 & (xP, )" + aP; gt Fa(Neg — Nog) — Fey — Mae) + rag, * 0 BE The stress and strain measures are related by the system of stress-strain relations given in chapter 2, Supplemented by appropriate boundary conditions these equations define a well-posed boundary value problem in-100- linear PDE. The solution of this BVP provides adescription of the elasto- statics of the spherical shell. For the solution process, additional relations such as compatibility equations and the stress function representation of the solution of equilibrium equations will also be useful. The compatibility equations are the static-geometric duals of the equilibrium equations while the stress function representation take the form 1 a ela ~Zlesckes + cote.) +N Nye = (ey +2) 1 5 > Neg = qtesebee cotée, - 2) Loy > My = (Uz - FD where quantities with a subscript p denote any particular solution of the equilibrium equations for the given surface loads. If we have 1aL r = 2b 5 ae > Py“ xr 38 then it is not difficult to see that we may takere as a part of the particular solution. Apart from the particular solution terms, the stress function solutions are the static-geometric duals of the strain-displacement relations.-102- 2. Reduction for the Classical Theory With y, = y, 705 we have as usual 1a L - eo GO ted os = (csckw' + u,) which in turn give Kp, > Keg > Kye amd Ky in terms of up uy w, and w (and their derivatives). It turns out that for spherical shells, the expressions for the strain resultants e »€, ae . ® Cece > fan? ee Egg ) Bay be used to eliminate u, , uy and w from the expressions for the four strain couples to get Liye aa kee 7 GEO AD = See c= - A(esc2ew" + cotew! + w) - 88 a > -- i cS rye 2 Cezgoege) ~~ qeesce et - cotew") = Zlepgetyg) + By the static-geometric dual development, we have also the following stress function representation for the stress resultants as consequences of 5 7 0 and the stress function solutions for the stress couples: l 1 i “4D + SM 5 Nyg 7 az + 2g - Mop) - © y.. = Ap(csc2eF” + cover! + F) + 204, -M,) - 8 ce a ave EP fi ly 1 - L ey - 4 . coteE’) + Mg My) +N, - GM i — {N, 4 sNye) = gzescé (F-103- a i M, Next, we use the stress-strain relations to get Mp + Msg > Meg > ead as in terms of F and w alone. From = dpa - o ei Meg = Me = ZD(L- 4) (Keg + Koz) > Soe 7 fee 7 ZAC +V,) (Neg +Nge) > we get with the help of results already obtained in this section M,, + +Da-v a ke =~ 3G -v, eset (w!* ~ cottu") og 1 1 5 +) a2 Aaa _)M, -e ne = + (BY -cot&F*) - tam}. Za(ity IMpg ~e5¢ = GEACH,) (eset (FI* - cottF”) aN, +N} ca These two equations may be solved for M., and €,, giving us saz eva, esed (w"* ~ cottw") Mee = My Loo 0 . 2 + +e2esck(F'" - cotEF’) - e2(aN, - M,)} a ¢ coteF’) (ax, > 1 1 a fi 5 { + 1°. coteF’) - a2N. + Teer! Gran, lescé F cotéF’) - a?N, + all] + de2esct (w"* - cottw)} where ©? = DA(1 - v2)/a? = h?/12a2 << 1. Similarly, we get from the remaining four stress strain relations (+e?) = - EPG" +0) + v,(csctew" + cotew! + wy] e 162 fesc2EF” ' 2 Ze? lese®sF’ + cove! + F] + e?(ab + M,,)-104- (1 +67)Mgg = ~ Aviv cw" +w) + (csc7Ew + cottw' + w)] -1.2[py 4 7] + ea + M, ) a ®p (ter)eg, = ater" + F) - vi (esc76F” + coteF" + F)] fala eee o ene ' - - loa -v2 ale [csce*Ew + cotéw' + w) - A[(1 vb + ava ven) (a te)ep, Peal (esc?er” + cotEF! + F) - vi (F" + F)] 4.2 fy S = Log - v2 gow" tw] - ALG = v)L + 2DG - v?)ee] where Kk. = (M, E ot -y2 = - -y2 Ep ep ~Ybllap)/D(L-v?) and «; M, vyit, ,)/D(L -v?) ep ‘ep db Meo * Mog are identical, so that we have with » the stress function solutions for N,, and N,- Neg = Given q, = 0 and Nyc a arte te naians satisfied identically, Similarly, we have from e,, = ¢,, the symmetry condition <,, — Kg, So that the sixth compatibility equation is also g satisfied identically. The remaining two moment equilibrium equations (with P, =P, = 4, = 4,70) give Q@ and Q, in terms of w and F E and surface load terms. Remarkably, all F terms cancel out in these expressions for the case of uniform shell thickness and material properties leaving us with 2)9 = -tepry2 pene eet eee on 1 + 2)Q, Gedlvew +2w]! +e? [LI + SMI + Zeotk ) ae 8p + Fesce ( - any)-105- 1 i 1 2 +e2)Q, © = v2 2 hens! er (1 +e2)Q, = -GresctD[Vw + 2w}’ + e715 OH - aN) qcote (M, - aN,) + Feseg (al" + M5.) a P where V2 is the dimensionless Laplace's operator in surface coordinates (E,8) 3 v2(.) = ("+ cote( )' + esc2E( DT - The expressions for strain resultants and stress couples obtained above can also be used in the expressions for the strain couples and stress resultants to get these quantities in terms of w and F alone: Ay (eseF” + cotgF’ + F) - G+ iy) L+e2)N G+ Ne ep - Apotow" + w) + vy (esc%ew + cottw! +] oy 2 haven - 1 (1te2)N, = Se" +P) - + SMD) op = Apply (w" + + (cac?ew” + corew' + ¥)] 2 e- Su" Ata- Loa -v2 (1+e%)ee, > gre" tw) + al-vL + S00 -v?)«, 1 ' Aytiesc2er” + covert + F) - vy(F" + FD) 2 = - dp (oac2ew” w + 4Ata- log -y2 Ate? egg gz (ese"gw + corew + w) + qld-vE+ gpa -¥ egy! Ate + F) - vi (esc?EF” + cover! + F)]-106- ; cE il With Qs Q) 5 Neg and Nyg all expressed in terms of w and F , the third (force) equilibrium equation now becomes a PDE for w and F. In the absence of surface loads, this equation takes the form (2 + 2192 + GQ + yyw - $F By the static-geometric duality, we have from the third compatibility equation a second PDE for w and F. Assuming again the absence of surface loads for simplicity, this second equation is evidently the static-geometric dual of the first equation: (2 +22 + Q-vIF+ fub =O. The appearance of Poisson's ratio notwithstanding, these two fourth order PDEs for w and F can be combined into a single fourth order complex PDE for a complex potential X=w+AF. It can be readily. seen from this complex equation that, for spherical shells complete in the circumferential direction, the solution for X (and therefore for w and F) may be obtained in terms of associated Legendre functions of real and complex degrees. Furthermore, the results are consistent with previously obtained results for axisymmetrically and laterally loaded spherical shells (provided we include suitably multivalued stress functions which lead to axial force, axial torque, side force and tilting moment). It is also possible to eliminate the Poisson's ratio term from-107- both PDEs for w and F by using a modified stress function F=F+D(1-v,)w/a instead of F. The elimination is achieved at a price, namely, a loss of the static-geometric duality between the two governing PDEs. The details of these observations will be left as exercises.-108- 10 - Circular Cylindrical and Conical Shells 1. Equations for Higher Harmonics of the Shell Response To the extent that we have completed the analysis for shells of revolu- tion with axisymmetric loads and with lateral loads, it suffices to consider only the solution process for a typical higher harmonic in the Fourier decom- position of the shell response mentioned in Chapter 1 of these notes. For n> 2, the n-th Fourier component of the load distribution is self-equilibrating and there exists no first integral to reduce the order of the shell equations, as there are for the n = 0 case and the n = 1 case previously analyzed. Thus, it is appropriate to seek for the classical theory a reduction of the shell equations to two simultaneous fourth order differential equations for the transverse displacement component and its dual stress function analogous to those obtained in the last two chapters for shallow shells and for (non- shallow) spherical shells, respectively. For the purpose of the contemplated reduction, we choose to satisfy all six equilibrium equations by way of six stress functions as indicated in chapter 2. Along with their dual strain displacement relations, the stress function representations transform the stress strain relations of Chapter 2 into twelve first order differential equations for the stress functions and their dual displacement components. The contemplated reduction can then be accomplished by using ten of the twelve equations to eliminate all but the transverse displacement component and its dual stress function from the remaining two equations. This process is simplified somewhat by the following observations for the classical theory. The conditions of vanishing transverse shearing strains, y, = Yq = 0, immediately allow us to express 4, and 6, in terms of-109- Ug, Ug and w while the symmetry condition cpg = eg¢ determines w in terms of the same three midsurface displacement components. Static-geometric dual con- siderations give the stress functions $¢, ¢g and 9 in terms of Ug, Ug and F (as well as the prescribed surface loads). These results effectively leave us with six (stress-strain) relations for us, Ug, wand their dual stress func- tions. For a typical harmonic of the shell response of the form {ug wWUgsF} = Cun(€,)s Wn(E,)+ Un(E,)» Fr(é,)}cos (ne), {ug, Up} = (n(,)> Vn(E,)}sin(ns), (n > 2) four of the six remaining stress strain relations for uz, Ug, w and their duals may be re-arranged to give Uy, vn, Un and Vp separately in terms of un, Vn> Uns Vas Yao Was Fas and Faz Une na oO uy = -YsCE Un + PE vn] = CR ACLS) FRE Un + Fay Yn (10.1) é = Luluns Yas Uns Yn> Wns Fao Fa] rm n vba. = -vpl Un + #2 Vn) - ER - Re dFn + ypmep aert 2 f +D(1-5) EER Un + Ty vat ae Wy - wd (10.2) = LylUns Vno Une Yn» Fos Wns Wa2 1-85 na iret 7 + 4napAs (10.3) r(1te = LyLuns Yn» Vav Wns Mp» Fre Fal eg 4nDsAs 301 ' ~Fo Na ri ‘ na vi =o My, + ly, -— - BF, - = Fd nae f r r(1te1) Re Rene 4napDs 4nDs vor 12DsAs (10.4) - eS ne m= F Wn) + Re anp i r(1tey) r(1ter) LyCUns Va> Yn> Fas Fas Wns WA] where wl il 20 3.) 1) 3 12 P= RS aE 67 OSASCRS - weMCRS - Re) 817 OsAsles - we) (10.5) ne = rol fMpr-3/"Unepgen2h, )rade, - (eG, - Sng. + B,)Irade ep 7 0! Porm’ Map rade, - (RE ep = 7 Mop * Peirace, Bo. - np = -r-2{"1(rig - niigp)radéy, gp = -SBq (10.6) : gp = -c2r- 1°12 Bprade Jade £2. dep = - tf le“ Bprade,-111- with s = sing = z'/a, c = cose = r'/a and ( )' = d( )/dé, and with prescribed surface load distributions in the form {Pe sPn} = (Be(E,)+ Pn(Z,)}cos (ne) Po = Bo(E,)sin(ne). (10.7 We now use (10.1) - (10.4) to eliminate u,, U, and Vj, from the remaining two stress strain relations to get n= Lyla Yar Mae “ne Une Va> Fae Fad ; (10.8) Ft bp LUn> Yao Fue Fae Une Ypo Me Wad We will not write out the expressions Ly2[.+.] and Lp[...J on the right hand side of (10.8) and merely note that they are linear combinations of the arguments. (See [14] and [15] for a more detailed discussion of the steps leading to (10.1) - (10.8)). The remaining steps in the reduction process consist of (i) differentiat- ing both sides of the expressions for “ and Fe in (10.8) and using (10.1) - (10.4) to eliminate uj, v,, U, and V, from the resulting expressions to get 2 Une Va F; (10.9) 2 Mae Wyle > Une Vas (ii) differentiating both sides of (10.9) and again using (10.1) - (10.4 to eliminate u,, v,, Un, and V,, from the resulting expressions to get wet ftusvewswWewywyU,V,F, F wal n n n n n n n n Fe Uns Var Was Wye-112- (iii) solving the four equations (10.8) and (10.9) (which are linear in Un> Vn> Uns Yn) for Uns Va» Uns and vp and using the results to eliminate these same four quantities from (10.10) to get two fourth order ODE's for Ww, and Fy. The reduction process will be illustrated in the next section by applying it to the case of a circular cylindrical shell with uniform materia properties and constant thickness. Once we have Wn and Fy (and therefore Un, vps Un and Vp), the stress re- sultants and couple are determined by fl n2 rt n Neg = (> Fy > pe Fn ~ gang Un ~ ag Yn + Mep?eos(n9) Noo = Fu. + Ly ‘ty + ngptcos (ne) 98 n a(R) Un + Mop fo gt one 1 2 lye ot n Nas Fa Fn ag fm * Za (Re Bey = VQ) ~ Bene Un + Mp}sin{na) ap nl Neo = a Fy - tag Fn + zag(Yn - Fo Ya) + oC - Hen + np}sin(ne) = & Vn ~ Un}sin(no) eho gt my, art n2 96 = faRS Fn * ara Yn * pez, Yn * Coe : ean + dzp}cos(ng) Qe= dt y+ So nt Be Und +i ppig - 2 vayisin(na) Re ae Pn Ry Yn The strain measures can either be obtained from the stress measures by the stress strain relations (and compatibility equations) or from displacement components by the dual strain displacement relations.-113- 2. The Classical Theory of Circular Cylindrical Shell With r(é,) = a and 2(¢,) = ag, for a circular cylindrical shell where the positive constant a is the radius of the circular cylindrical midsurface, we have L Re aa, Equations (10.1) - (10.4) simplify in this case to read = 2 a(i-v2)tn?Fq - aVp] + vslwn - On] c 2 2 : Anas vy neh, | (Bee) eee eee en ae 0” ~ ated) (Un ~ Fd 3(ieg) "n * 3(aeea) (10.11) Us = 20(1-v2)[n2Hq = vA] = vplFy - nV] 2 2 + 4nDs ‘ 4neyy . n(3-e7) vie MOS tue owt + ML pty Me n* Z(aeedy Cen “nd * 3qqyezy fn * 3CGeezy OP 1 1 1 where now 2_9 9 h2 =X dhs = 2 & * az OSS © ag(av2) a2 and where we have omitted all surface load terms to simplify the presentation. Appropriate equations for circular cylindrical shells including load terms may be obtained by specializing the results of the next section. Equations (10.8) become = = 5 (Vp Fn) + vp(n2W%p = Vn) n : 10.12 FY = 3 (nvq = Wn) = v5(n2Fp ~ 1Vq) (10.12) We now differentiate both sides of (10.12) to get = % (nv, = Fa) + vp(n2w, = nv.) ema (10.13) nie (nvp = wh) = ys (n2F 5 nv)-114- The expressions for V, and v, in (10.11) are then used to eliminate these quantities from (10.13) to give 2a*(w! 3 (aR 2A = n2A%(wh = Uy) + 5 (AZFA ~ MPA3U,)» (10.13') = n2A,(F! 2 (Agwt = n2 m= n2Ai(F\ - Un) - (AowWh = n?Asu,). The expressions for A,, A,, A, and their duals a, ay and ay in terms of D, A. Dg, Ag and vp = vg can be found in [14]. Next, we differentiate (10.13') and use (10,11) to eliminate uy and U/ from the resulting expressions to get it 2 ete a * " wim = n2(BiwWa + Bowe) - 5(n2B2F py + BFA) s an? + Biv + Sy BEV (10.14) F™ = n2(ByFy + Bof") - 2(n2Bowq + Bunn) 3 + n?BeVq - “F~ Ben The constants B., ..., B, and their duals Br. odds BE can also be found in [14]. We can then solve (10.12) for Vp and Vq and use the result to eliminate these two quantities from (10.14). In this way, we get (1-v2e2)nvq = (Len2v2e2)Wn + vgeZwh + Las (n@-1)Fq + Feds (10.15) ee ‘ (1-v2e2)nVq = (1-n?v2e2)Fq - vpeSFy + SOL vp(n2-1)wy + wT, where ek = DA/a2, and ma ay eee" oe wy = CML + Cin = S(CKF" + C2Fn) (10.16) Fe = cof + CiFy = ~ S(CuWl + C3Wn) where c,, c,, C,, c, and their duals are given in [14]. Fp-115- It should be noted that the reduction described above is exact; it involves no approximation whatsoever. Similarly, the stress, strain and dis- placement measures of the shell can be calculated exactly by the auxiliary formulas given in the last section, specialized to the case of a circular cylindrical shell, once we have Wj and Fy. However, the exact equations for Wn and Fy, (10.16), contain many terms which are of magnitude no larger than the error inherent in shell theory (as an approximation of the exact solution of three dimensional elasticity) and may be omitted without loss of accuracy when viewed in the context of an approximate solution for the three dimension- al problem, When these negligibly small terms are deleted, equations (10.16 simplify to read man? + n2(n2-L)Win = FF (10.16') one" 4 n2(n2— 2-2 FM. an2F" + n2(n2-1)Fy = - Note that the left hand side of the two equations in (10.16') corresponds to the binarmonic operator applied to w = Wa(é,)cos(nd) and F = Fy(£,)cos(ns), respectively. Therefore, we may also write these equations as v2yay = 2 FH, v2y2F = a w" (10.16") o> Equation (10.16") are identical to those obtained in [11] by a different meth- od which also makes allowance for the inherent error of shell theory. In the form (10.16') or (10.16") the two governing equations for w and F may of course be further reduced to a single complex fourth order equation for a com- plex potential. The exact reduction procedure for circular cylindrical shells leading to the two simultaneous equations for wry and Fy given by (10.16) applies to any shell of revolution including those with inhomogeneous material properties and nonuniform thickness. The final two equations for wy and Fy will normally-116- contain many terms of the order of the error inherent in shell theory which may be deleted without affecting the accuracy of the final solution. It is generally possible to obtain the same two simplified (but accurate) equations for wy and F, by deleting the same type of small terms at the intermediate stages of the reduction process and thereby reduce the amount of calculations involved. This simplified reduction process has been applied to the case of conical shells with uniform material properties and constant thickness. The results will be reported in the next section.-117- 3. Conical Shells and Generalized Hypergeometric Functions For a conical shell, we may describe its middle surface by Perot Ch, 27% +88 (10.17) where rg, Zo, C and s are fixed geometrical parameters and, in terms of the meridional slope angle €, ¢ = cose and s = sing. Correspondingly, we have = 1 1 eae ae re (10.18) where Re and Rg are the two principal radii of curvature. Evidently, the independent variable <, is the meridional arclength measured from the end r= tg and z= Zo. For our particular a, Re and Rg, the two equations (10.1) and (10.2) for us and U, simplify to read: uy = = vgt-3(cun + vq) + A(1-v2)nsr-2V_ + vgsr-twy + A(L-v2)(cr-PF) = n2r-2Fq + nep) (10.19) UL = vpr72(cUn + nVq) - D(1- 5)nsr-2vy = vpst}Fy = D(1-vg)(cr-tw, = n2r-2Wq) + vpmep with the inhomogeneous terms agp, etc. given by (10.5), while (10.3) and(10.4 become Vi = > 4nsr-2AsUn + cr-lvq + nrcluq + 4nr-Aas(F, - cr!Fy) + Lansr-? DgAs(w = cr-2Wn) + 4Asnp (10.20) Vs 4nsr~2Dguy + cr7}Vp_ + nem tUn - 4nr-2Ds(w, = cr-lwn + L2nse-2DsAs(F, - cr-2Fy) + 12DsAgsr-3np