Advanced Strength of Materials

ADVANCED STRENGTH OF MATERIALS

J. P. Den Hartog

Professor Emeritus of Mechanical Engineering

Massachusetts Institute of Technology

DOVER PUBLICATIONS, INC.

New York

Copyright © 1952 by the McGraw-Hill Book Company, Inc.

All rights reserved.

This Dover edition, first published in 1987, is an unabridged and unaltered republication of the work first published by the McGraw-Hill Book Company, N.Y. in 1952.

Library of Congress Cataloging-in-Publication Data

Den Hartog, J. P. (Jacob Pieter), 1901-

Advanced strength of materials.

Reprint. Originally published: New York : McGraw-Hill, 1952.

Includes index.

1. Strength of materials. I. Title.

TA405.D38   1987     620.1′12

87-6746

eISBN-13: 978-0-486-13872-5

Manufactured in the United States by Courier Corporation

65407908

www.doverpublications.com

PREFACE

This book deals with material which is covered in two courses at M.I.T., each of a semester’s duration. The first of these, taken in the senior year, deals with Chaps. 1, 2, 3, and 5; and the second, given as a graduate course, covers the remaining chapters. As the title indicates, the book cannot be used by a beginner; it is aimed at the student who has had the usual one-semester course in elementary strength of materials. In writing this text I have followed the notations of my previous elementary Strength of Materials (1949), and Mechanics (1948), but the present book can be used after the study of any other elementary exposition.

Many good textbooks on elementary strength of materials are readily available and, on the other hand, the mature student can find all that is wanted in the series of advanced books by Timoshenko on elasticity, plates and shells, and elastic stability. The difference in level between those books and the elementary texts, however, is formidable, and with this text an attempt has been made to bridge the gap and supply something of intermediate difficulty.

I express my gratitude to the friends and students who have generously given me their advice and help during the writing, particularly to Mr. Iain Finnie, who worked out all the problems, and to Mr. Mauricio Casanova, who drew all the illustrations.

J. P. DEN HARTOG

CAMBRIDGE, MASS.

April, 1952

CONTENTS

Preface

Notation

CHAPTER I.TORSION

1.Non-circular Prisms

2.Saint-Venant’s Theory

3.Prandtl’s Membrane Analogy

4.Kelvin’s Fluid-flow Analogy

5.Hollow Sections

6.Warping of the Cross Sections

7.Round Shafts of Variable Diameter

8.Jacobsen’s Electrical Analogy

CHAPTER II.ROTATING DISKS

9.Flat Disks

10.Disks of Variable Thickness

11.Disks of Uniform Stress

CHAPTER III.MEMBRANE STRESSES IN SHELLS

12.General Theory

13.Applications

14.Shells of Uniform Strength

15.Non-symmetrical Loading

CHAPTER IV.BENDING OF FLAT PLATES

16.General Theory

17.Simple Solutions; Saint-Venant’s Principle

18.Circular Plates

19.Catalogue of Results

20.Large Deflections

CHAPTER V.BEAMS ON ELASTIC FOUNDATION

21.General Theory

22.The Infinite Beam

23.Semi-infinite Beams

24.Finite Beams

25.Applications; Cylindrical Shells

CHAPTER VI.TWO-DIMENSIONAL THEORY OF ELASTICITY

26.The Airy Stress Function

27.Applications to Polynomials in Rectangular Coordinates

28.Polar Coordinates

29.Kirsch, Boussinesq, and Michell

30.Plasticity

CHAPTER VII.THE ENERGY METHOD

31.The Three Energy Theorems

32.Examples on Least Work

33.Proofs of the Theorems

34.Bending of Thin-walled Curved Tubes

35.Flat Plates in Bending

CHAPTER VIII.BUCKLING

36.Rayleigh’s Method

37.Coil Springs; Beams on Elastic Foundation

38.Proof of Rayleigh’s Theorem

39.Vianello’s or Stodola’s Method

40.Rings, Boiler Tubes, and Arches

41.Twist-bend Buckling of Beams

42.Buckling of Shafts by Torsion

43.Twist Buckling of Columns

44.Thin Flat Plates

CHAPTER IX.MISCELLANEOUS TOPICS

45.Mohr’s Circle for Three Dimensions

46.Torsion of Pretwisted Thin-walled Sections

47.The Theorems of Biezeno and Spielvogel

Problems

Answers to Problems

Index

NOTATION

CHAPTER I

TORSION

1. Non-circular Prisms. The most useful and common element of construction subjected to torsion is the shaft of circular cross section, either solid or hollow. For this element the theory is quite simple, and we remember that the shear stress in a normal cross section is directed tangentially and its magnitude is given by the expression

where R is the outside radius of the shaft, r (smaller than or equal to R) is the radius at which the stress is measured, and Mt is the twisting moment on the shaft. We also remember that the angle of twist θ is determined by

In the derivation of these two equations it was assumed (or rather it was shown by an argument of symmetry) that plane cross sections in the untwisted state remain plane when the twisting torque is applied and also that these cross sections remain undistorted in their own plane.

We now propose to find formulae for the twisting of shafts that are not circular in cross section but that are still prisms, i.e., their cross sections are all the same along the length. For such shafts it is no longer possible to prove that plane cross sections remain plane or that they remain undistorted in their own plane. In the proof of these properties for a circular cross section the rotational symmetry of that section about its central point is essential.

If plane cross sections remain plane and undistorted, it follows logically that the shear stress must be along a set of concentric circles, as in the round shaft. Now it can be easily seen that this cannot be true for a non-circular shaft, because then the stress (Fig. 1) would not be tangent to the boundary of the cross section, and would have a component perpendicular to that boundary. Such a component would be associated with another shear stress on the free outside surface of the shaft, which does not exist. The shear stresses on a cross section thus must be tangent to the periphery of the cross section. As a special case of this we see that the shear stress in the corners of a rectangular cross section must be zero, because neither one of its two perpendicular components can exist.

FIG. 1. If the shear stress ss in a peripheral point of the cross section is perpendicular to a radius from the center of twist C, then it can be resolved into tangential sst and normal ssn components. The normal component must have a companion stress on the free outside surface, which does not exist. Hence the normal component sn must be absent.

Figure 2 shows a circular shaft and a square shaft being twisted. In both cases longitudinal lines on the periphery which were originally straight and parallel to the shaft’s center line become spirals at a small angle γ as a result of the twisting. In the round shaft (Fig. 2a), an element dr rdθ dl, in which all angles are 90 deg in the untwisted state, then acquires angles of 90 – γ, as shown in Fig. 2c, because the plane cross section remains plane. This angle of twist γ of the particle is associated with the shear stress of twist. Now let us look at the corner particle dx dy dl of the square shaft. Again the spiral effect of the twisting couple causes the small angle γ, and if plane cross sections would remain plane, then the angle γ would necessarily be associated with the shear stress of Fig. 2c. But, as we have seen, this is impossible because the shear stress ssn on the free outside surface does not exist. Hence the only possibility is shown in Fig. 2d; the upper face of the cross section also must turn through an angle γ, to keep the angle at 90 deg, so that no shear stress occurs. This means that the corner element of area of the cross section is perpendicular to the spiraled longitudinal edge, and since this must be the case at all four edges, the plane cross section is no longer plane but becomes warped vertically.

FIG. 2. If in a twisted bar of square cross section a plane cross section should remain plane, there would be shear stresses in the corner, as shown in (c); a zero shear stress in the corner is possible only when the upper surface of (d), that is, the normal cross section, tilts up locally. Only with a circle (a) is a plane cross section possible.

For the circular cross section it was shown that plane cross sections remain undistorted in their own plane. This means that if we draw on that normal section a network of lines at right angles (such as a set of concentric circles and radii or also a square network of parallel x and y lines), then these right angles remain 90 deg when the torsional couple is applied. We cannot prove that this property remains true for non-circular cross sections. However, in Fig. 3 we see what a distortion of the normal cross section implies. With such a distortion shear stresses appear in sections parallel to the longitudinals, while no shear stress in a normal section is necessary. Only these latter stresses can possibly add up to a twisting torque, and the stresses of Fig. 3 are useless for resisting a twisting torque. Later, in Chap. VII we shall see that such useless stresses never appear. Nature opposes a given action (here a twisting torque) always with the simplest possible stresses: to be precise, the resisting stresses are so that they contain a minimum of elastic energy. The stresses of Fig. 3 add to the stored elastic energy in the bar, while they do not oppose the imposed twisting couple. Although this argument does not constitute a proof, it makes it plausible that normal cross sections do not distort in their own plane.

FIG. 3. If in a twisted shaft normal cross sections should distort m their own plane, there would be shear stresses on sections parallel to the longitudinals of the bar. Such stresses do not contribute to a twisting moment. In fact they do not exist, and normal cross sections do not distort in their own plane.

FIG. 4. Saint-Venant assumes that a cross section turns bodily about a center, without distortion. Thus a point A turns to B through an angle θ1z. When the displacements are called u and υ, this turning is expressed by Eqs. (2).

With this preliminary discussion we are now ready to start with the theory of twist of non-circular cylinders or prisms. This theory is due to Saint-Venant and was first published in 1855.

2. Saint-Venant’s Theory. Let x and y be perpendicular coordinates in the plane of a normal cross section with their origin in the center of twist, i.e., in the point about which the cross section turns when twisting. Let z be the coordinate along the longitudinal center line, and at z = 0 we place our section of reference which is not supposed to turn. Then Saint-Venant’s assumption for the deformation (Fig. 4) can be written as

Here u, υ, and w are the displacements of a point x, y, z from the untwisted state (A in Fig. 4) to the twisted state B, in the x, y, and z directions, respectively, and θ1 is the angle of twist of the shaft per unit length. The expressions for u and υ state that a cross section at distance z from the base turns about the origin through an angle θ¹z in a clockwise direction. The sign of υ is negative because for positive x, a point moves in the negative y direction when it turns clockwise. The third expression w = f(x, y) states that a cross section warps by an amount w in the longitudinal z direction; that this warping is different for different points x, y, following an as yet unknown pattern f(x, y). It further states that all cross sections warp in the same manner since w is independent of z.

The next step is to express the assumed displacements, and also γxy = 0. The third of Eqs. (2) states that all planes warp in the same manner; hence the z . Thus there are only two strains left: γxz and γyz. In order to express γxz we study a section in an x-z plane (y = constant) which is parallel to the longitudinals of the bar, shown in Fig. 5. An element ABCD goes to A′B′C′D′ because of twisting, and the two originally plane sections z and z + dz become the warped ones shown in dashes. The horizontal distance between A and A′ we have called u. The horizontal distance between C and C′ can be written as u + (∂u/∂z) dz, because point C differs from point A by the distance dz only, while A and C have the same x value. The horizontal distance between A′ and C′ then is (∂u/∂z) dz, and the (small) angle between A′C′ and the vertical is ∂u/∂z. We now repeat this whole story for the vertical distances (instead of the horizontal ones) of the points A, B and A′, B′. The reader should do this and draw the conclusion that the (small) angle between A′B′ and the horizontal is ∂w/∂x. Hence the difference between ∠ C′A′B′ and ∠ CAB is (∂u/∂z) + (∂w/∂x), and by definition this is the angle of shear γxz in the xz plane. For the yz plane the analysis is exactly the same, only the letter x is replaced by y (and consequently u is replaced by υ) in all of the algebra as well as in Fig. 5. Thus we arrive at

FIG. 5. Derivation of the expressions (3) for the strain γxz. An element dx dz in the vertical xz plane has the contour ABCD before the bar is twisted and goes to A′B′C′D′ when the bar is twisted.

Now Eqs. (2) state what the displacements are. Substituting Eqs. (2) into the above leads to

With Hooke’s law these strains are expressible in the stresses

FIG. 6. Definition of the signs of the shear stresses (ss)xz and (ss)yz. The z axis points upward; the three axes form a right-handed system, and the two subscripts xz mean that the shear stress chases around a small element in the xz plane. The × means that we are looking on the feather end of an arrow and • looks on the point of an arrow.

where the signs of the stresses are shown in Fig. 6.

The next step is the derivation of the equation of equilibrium. It is clear that the shear stresses are not constant across a normal cross section xy but differ from point to point. (In the circular section the stresses are zero at the center and grow linearly with the distance from it.) Thus the stresses on opposite faces of the dx dy dz element of Fig. 7 are not exactly alike but differ from each other by small amounts. If, for example, the stress on the dy dz face for the smaller of the two x values is denoted by (ss)xz, shown dotted in the figure, then the stress on the opposite dy dz face (which is distance dx farther to the right) is somewhat different and can be written as

or more precisely as

FIG. 7. Derivation of the equilibrium equation. Vertical equilibrium (in the z direction) of the element requires that Eq. (5) be satisfied. Horizontal equilibrium, both in the x and y directions, is automatically satisfied because the stresses are functions of x and y only and do not vary in the z direction by Eqs. (4) and the last of Eqs. (2).

The extra, unbalanced, upward stress on the pair of dy dz , and since this stress acts on an area dy dz, the unbalanced force . The reader should now repeat this argument for the two dx dz . There are no other vertical forces or stresses on the element, and an element in a twisted bar is obviously in equilibrium. Setting the net upward force equal to zero and dividing by the volume element dx dy dz gives the equilibrium equation

This is a partial differential equation in terms of two unknown functions (ss)xz and (ss)yz, both depending on two variables x and y. The problem of finding a solution would be very much simpler if we had to deal with one single function of (x, y) instead of with two. Here we come to the vital step in Saint-Venant’s analysis. He assumes that there is a function Φ(x, y), such that the stresses can be found from it by differentiation, thus:

The definition (6) of the new function Φ has been chosen cleverly: by substituting (6) into (5) we see that Eq. (5) is automatically satisfied for any arbitrary function Φ, provided that

which is always true if Φ is a continuous function. The function # is called the stress function of the problem; here it is Saint-Venant’s torsion stress function. Other examples of stress functions will be seen later, on pages 42 and 174.

Now we can begin to visualize the situation geometrically. The value Φ can be plotted vertically on an xy base and thus forms a curved surface. Then Eqs. (6) state that the (ss)yz stress, i.e., the stress in the y direction, is the slope of the Φ surface in the x direction, and vice versa, that the shear stress in the x direction is the (negative) slope of the Φ surface in the y direction. We shall now prove that this statement can be generalized and that the shear stress component in any direction equals the slope of the Φ surface in the perpendicular direction. Before we proceed to the proof, we notice that by Fig. 1 (page 2) the shear stress normal to the periphery of the shaft is zero; hence the Φ slope along the periphery must be zero, which means that the Φ height all along the periphery must be constant. The Φ surface then can be visualized as a hill, and if we cut this hill by a series of horizontal planes to produce contour lines, then the shear stress follows those contour lines. For a circular cross section the contour lines are concentric circles, and the Φ hill is a paraboloid of revolution.

Now to the proof. Consider in Fig. 8 an element at point A with the stresses (ss)xz and (ss)yz. Draw through A the line AB in an arbitrary direction α, and let AB be dn, the element of the normal direction. Perpendicular to AB is the line CD. When AB = dn, then AE = dx and EB = dy, and we see from the figure that dx/dn = cos α and dy/dn = sin α. Over all these points is the hilly surface of the stress function Φ, and we have in general

or

Transcribing this by means of Eqs. (6) leads to

Now, looking at Fig. 8 we see that

or equal to the component of stress in the direction DAC, which is perpendicular to AB. But dΦ/dn is the slope of the Φ surface in the AB direction, which completes our proof.

FIG. 8. Toward the proof that the slope of the stress function surface in any arbitrary direction α equals the total shear stress component in the perpendicular direction.

Now we possess the single stress function Φ, with which we shall operate instead of with the pair of stresses (ss)xz and (ss)yz. The first thing to do is to rewrite all our previous results in terms of Φ. Turning back, we first find Eq. (5), which we have seen is automatically satisfied by the judicious definition of Φ. Next we find Eqs. (4), which now become

In these the quantity w is the warping of the cross section. We do not know what w looks like, but it is certain that w and its derivatives are continuous functions of x and y: the warping will have no sudden jumps or cracks in it. One mathematical consequence of this continuity is

We now operate on Eqs. (7) in two ways. First we apply ∂/∂y to the top equation, ∂/∂x to the bottom equation, and subtract them from each other, which gives

Second we apply ∂/∂x to the first of Eqs. (7), ∂/∂y to the second, and add the results together, which leads to

The last result is a partial differential equation for finding the warping function w. We shall postpone discussing it until page 31.

The result, Eq. (8), is a partial differential equation for the stress function Φ. From the previous derivations we understand that any arbitrary function Φ, whether it satisfies Eq. (8) or not, leads to stresses [by the process of Eqs. (6)] that satisfy the equilibrium equation (5). If the function does satisfy (8), it leads to stresses which correspond to continuous warping deformations w, whereas a Φ which violates (8) gives equilibrium stresses all right, but the corresponding warping function is discontinuous. Therefore Eq. (8) is an equation of continuity, or, by a special term used in the theory of elasticity, Eq. (8) is known as the equation of compatibility. If for a given cross section we can find a Φ function which satisfies Eq. (8), and which also satisfies the boundary condition Φ = constant along the periphery, then the stresses derived from that Φ are the true solution to the torsion problem.

FIG. 9. The transmitted torque across the section is found by integrating the contributions of all elements dx dy of the section.

Torque in the Shaft. Before we proceed to find a few solutions of this kind, we first derive one more result: for the torque transmitted by the shaft. The clockwise torque about the origin furnished by an element dx dy (Fig. 9a) is

To find the total torque, this expression must be integrated over the entire cross section. We first calculate the first term in the bracket, i.e., the torque caused by the (ss)xz stress alone.

Substituting Eq. (6), we find

Integrate this first along a strip of width dx parallel to the y axis (Fig. 96). Then x is constant, and we can replace (dΦ/dy) dy by dΦ without ambiguity. Thus, integrating by parts,

The first term in brackets, when taken between the limits A and B, is yBΦB – yAΦA. Now we know that Φ must be constant along the boundary, so that ΦA = ΦB. We have not so far decided what absolute value we shall assign to Φ at the boundary. For simplicity we now decide to set Φ equal to zero at the boundary. We are allowed to do this, because Φ after all is only an auxiliary function, existing only for convenience. We are decidedly interested in the stresses, which follow from Φ by differentiation by Eqs. (6). Suppose we add a constant to Φ, that is, suppose we raise the entire Φ surface. Then the slopes remain the same everywhere, and hence the stresses remain the same. The Φ function has the nature of a potential function, such as gravitational height, electric voltage, or the velocity potential of hydrodynamics. Its absolute value has no meaning; only its differences are important. Hence we are free to assign a zero value to Φ at one location wherever we please; then the function is fixed at all other locations. We choose for zero Φ the boundary of the bar; then ΦA = ΦB = 0 (Fig. 9b), and the above integral becomes

which is the volume under the Φ hill. The reader should now calculate the torque caused by the (ss)yz stresses in the same manner, by integrating first over a strip dy parallel to the x axis. He should find the same result with the same sign. Thus it has been proved that the transmitted torque Mt equals twice the volume under the Φ hill, provided that Φ is taken equal to zero at the boundary:

3. Prandtl’s Membrane Analogy. Saint-Venant in 1855 found solutions of this torsion problem for a number of cross sections, such as rectangular, triangular, and elliptic sections. Those solutions were found by complicated mathematical methods, often involving infinite series. Many important practical sections such as channels and I beams cannot even be reduced to a mathematical formula, so that approximate methods of solution are desirable. The best method of this sort among the many that have appeared is due to Prandtl (1903). Prandtl observed that the differential equation (8) for the stress function is the same as the differential equation for the shape of a stretched membrane, originally flat, which is then blown up by air pressure from the bottom. This remark will give us an extremely simple and clear manner of visualizing the shape of the Φ function, and the stress distribution. We therefore now derive the equation of a thin, weightless membrane initially with a large tension T (expressed in pounds per inch, and having the same value in all directions), blown up from one side by a small excess air pressure p (expressed in pounds per square inch). By a large initial tension T we mean such a tension that its value will not be changed by the blowing-up process. The membrane is an elastic skin, of rubber, for example. If in the unstressed state we draw on it a network of squares, then these squares have to be deformed into larger squares in order to get tension in it; in other words, there must be initial strain connected with the tension T. This initial strain must be the same in all directions, and hence T is also the same in all directions and is expressed in pounds per running inch of (imaginary) cut in the membrane. When the membrane is blown up from the flat shape into a curved surface, being held at the edges, obviously the lengths of lines drawn on it increase, so that the blowing-up process causes more strain. We now prescribe that the air pressure p must be so small and the initial tension T must be so large that the blowing-up strain is negligible compared with the initial strain and that consequently T remains constant during the blowing-up process.

FIG. 10. An originally flat membrane with large tension T under the influence of a small air pressure p from the bottom assumes a shape z with small slopes, satisfying the differential equation (11).

Derivation of the Membrane Equilibrium Equation (11). Let in Fig. 10 such a membrane be shown, lying originally flat in the xy plane and then having air pressure p blowing it up to ordinates z. The periphery of the membrane is fixed so that the peripheral points remain at z = 0 when the interior is blown up. Since the pressure p is small, the ordinates z likewise will be small. The equation of the blown-up skin then will have the form z = f(x, y), and the slopes of this shape, ∂z/∂x and ∂z/∂y, will also be small. Consider a small element dx dy of the membrane. It will have acting on it two forces T dy in the x direction on the two opposite cuts dy, two forces T dx on opposite faces dx in the y direction, and finally a force p dx dy perpendicular to it, practically in the z direction. Now we resolve these various forces into components in the x, y, and z directions. These components are the forces themselves, multiplied by the sine, cosine, or tangent of the slope. Since, for small angles,

we can say that the cosine of the slope equals unity and that the sine (or tangent) of the slope equals the slope itself. Then we are correct up to magnitudes of the first order small, neglecting quantities of the second and higher orders. In that case the horizontal equilibrium, either in the x or in the y direction, of the dx dy element is automatically satisfied, because the horizontal components to the right and left are both equal to T dy, while the x component of the air-pressure force is two orders smaller than that and hence negligible. However, the z equilibrium gives a good equation, as follows: The z component of T dy on face A (Fig. 10) is T dy (∂z/∂x) downward. The z component on the opposite B face would be the same (upward) if the slope ∂z/∂x were the same, but in general it is not. That component can be written as

The net sum upward of the membrane tensions at A and B together thus is

and we see that it is proportional to ∂²z/∂x², which is the curvature (for small slopes). Similarly we deduce that the net upward force resulting from the two dx faces is

The third upward force is the air pressure p dx dy, so that, after dividing by T dx dy, the z equilibrium equation becomes

or, in words: The sum of the curvatures in two perpendicular directions is a constant for all points of the membrane.

Now, if we adjust the membrane tension T or the air pressure p so that p/T becomes numerically equal to 2Gθ1, then Eq. (11) of the membrane is identical with Eq. (8) of the torsional stress function. If, moreover, we arrange the membrane so that its heights z remain zero at the boundary contour of the section, then the heights z of the membrane are numerically equal to the stress function Φ; the slopes of the membrane are equal to the shear stresses (in a direction perpendicular to that of the slope); the contour lines z = constant of the membrane are lines following the shear stresses, and the twisting moment is numerically equal to twice the volume under the membrane [Eq. (10)].

Practical Use of the Membrane Analogy. To calculate the shape of a membrane for a given cross section by integration of Eq. (11) is, of course, just as difficult as to calculate the stress function from Eq. (8). But with the membrane analogy we can do two things: we can measure experimentally, or, more important still, we can visualize intuitively.

Experiments have been made, using stretched rubber sheets or soap films for membranes. At first thought it would seem necessary to know the value of the tension T and to regulate the air pressure p of the membrane so that p/T = 2Gθ1. This would be complicated and is usually avoided by taking a large membrane and by blowing up side by side two cross sections: the one to be investigated and a purely circular one. Since the p/T value is the same for both, because the same membrane is used, the corresponding 2Gθ1 is also the same. Then if we measure the volumes under the two hills and also the slopes in each, we conclude that in

the constant is the same for both sections. Since we know all about the circular section, the constant is easily calculated and then we have the torque/stress ratio for the other section. Also,

We now propose to calculate the stress and stiffness of a few simple sections by the membrane analogy. Before doing that, let us first see what happens to the circular section. On account of symmetry the height z of the membrane there does not depend on the two numbers x and y but can be said to depend on the single quantity r only. Cutting a concentric circle out of the membrane and setting the downward pull on the periphery 2πr equal to the upward push of the air pressure on πr² gives

or, in words: The slope of the membrane is proportional to the distance r from the center, and hence the