]\gt ; INDEX SLIP .
FILON , L. N. G.\mdash ; On the Besistance to Torsion of certain Forms of Shafting , with special reference to the Effect of Keyways .
Trans. , , vol. 193 , 1899 , pp. Torsion , of Prisms of particular Forms .
Effect of Keyways .
Points .
FILON , N. G. .
Trans. , , vol. 193 , 1899 , pp. Conjugate Functions ; application to the Problem of Torsion .
FILON , L. N. G. Phil. Trans. , , vol. , 1899 , pp. [ 309 ] X. On the to Torsion of Certain Forms of ) to the Effect of L. N. G. 1LON , Siudent of 's College , Cambridge , of University College , , 1851 Exhibition Science Research by M. J. M. HILL , F.lf .
Received June li 1899 .
S 1 .
Object Methods of the THE object of the present paper is to obtain solutions of the problem of torsion cylindelB whose cross-sections are bounded by confocal conics .
It is mainly an extension of DE SAINT-VENANT'S investigations , and is based upon his general equations of torsion .
The method employed depends upon the use of conjugate and , such that const .
represents confocal ellipses and const .
confocal hyperbolas .
The use of conjugate functions for the torsion problem has suggested by THOMSON and TAIT ( Natural Philosophy by CLEBSCH Theory der Elasticitat fester Korper , ' SS 33-35 ) , and by BOUSSINESQ ( ' Journal de Mathematiques , ' pp. , Serie III .
, vol. 6 ) .
CLEBSCH has used such elliptic coordinates to solve the torsion problem for hollow cylinders bounded by confocal ellipses , and DE has applied conjugate functions to the same problem for shafts whose sections sectors of circles ; curvilinear coordinates have also been employed by Mr. H. M. cDoNALD Ou the Torsional Strength of Hollow Shafts 'Proc .
Camb .
Phil. Soc vol. s , , p. 62 , et , but I am not aware that the actual solution has yet been obtained for sections bounded by both ellipses and hyperbolas .
The work proceeds on lines analogous to those developed by himself , in his solution of the ) roblem of torsion for the cylinder of rectangular cross-section .
The strains and stresses are expressible in terms of infinite series involving circular and hyperbolic functions .
The boundaries of the section are given by constant values of and .
The values of are taken to be The conditions from which the unknown quantity ( the shift parallel to the axis ) is determined are 29.12.99 .
throughout the section ; and along the boundary , where element of the outwards normal to the boundary , is the angle of torsion per unit length , and are the direction-cosines of Now in the present case where at the boundary where const .
, and the boundary where const .
, the sign being so determined that is positive .
By adding suitable terms to , we can reduce one or other of the boundary conditions to the form where suitable terms .
Suppose we make Expanding now in the form of a series , the differential equation and the first boundary condition are identically satisfied .
When this value is substituted in the second boundary condition , we get an equation expressing a given function of in a series of sines of odd multiples of between the limits and But such an expression can be definitely obtained by a method analogous to that for FOURIER'S series .
Comparing coefficients , we obtain relations which determine completely all the constants in the expression of is then known .
The shears and torsion moment are then deduced by differentiation and a double integration .
S2 .
of the Resutts .
The cross-sections which are dealt with in the present paper are of very great generality , and they include as special cases many of the cross-sections which SAINT- VENANT has worked out , for instance the rectangle and the sector of a circle .
The first section of which I treat is that bounded by an ellipse and two confocal TORSION OF CERTAIN FORMS OF SHAFTING .
hyperbolas .
Although the analysis is worked out for the case where the two hyperbolic segments are not symmetrical , I have not given any numerical examples of this case , as the sections obtained by taking two hyperbolas curved the same way , as in fig. 1 , do not correspond to any intersecting practical case : the section is too broad at the ends and too narrow at the bend to be any fair representation of the angle iron .
The section ( fig. 2 ) bounded by an ellipse and the two branches of a confocal hyperbola is , on the other hand , an approximate representation of a well-known section , much used in engineering practice , the rail section .
This section I have worked out for various values of the eccentricity of the ellipse and of the angle between the asymptotes of the hyperbola .
The four sections in fig. 2 , where this angle is , give the best representation of the rail section .
The numerical results are tabulated so as to show the ratio of the torsional rigidity of this section to that of the circular section of the same area , and also the same ratio for the maximum stress .
The ratio of these two ratios gives us a kind of measure of the usefulness or " " efficiency\ldquo ; of the section .
In the case of the sections of fig. 2 I have investigated at length the position of the -points , or points of maximum strain and stress , the maximum strain , in the case of torsion , being coincident with the maximum stress .
It is found that for the two smaller ellipses the maximum stress occurs at the point where the section is thinnest .
For the two larger ellipses the maximum stress occurs at points , symmetrically distributed round the contour , and lying on the broad sides of the section .
The critical section , when these two cases pass into one another , can be calculated and is shown as in fig. 2 .
In figs. the corresponding points belonging to the different sections are distinguished by suffixes .
The changes in the stresses are shown by the curves in fig. 9 , ( p. 340 ) in which the abscissa represents the quantity whose hyperbolic cosine and sine are proportional to the major and minor axes of the ellipse respectively , and in which the ordinates represent the stresses at , divided by the maximum stress of the circular section of equal area .
The curves are in certain parts only roughly drawn , but they suffice to show the manner in which the stresses vary .
It is seen that the stress at separates from the maximum stress after the critical value , and gradually dimimishes , compared with the stresses at A and F. This result might have been expected from the investigations of DE SAINT-VENANT upon certain sections bounded by curves of the fourth degree .
These investigations appear , however , not to have been sufficiently noticed .
THOMSON and TAIT , in their ' Natural Philosophy , ' and BOUSSINESQ , in his researches on torsion ( ' Journal de Mathematiques , ' Serie II .
, vol. 16 , p. 200 ) , both conclude that the fail-points are at the points of the cross-section nearest to the centre , and BOUSSINESQ even gives an apparently general proof of this proposition .
His proof , however , is subject to certain restrictions which I point out , and which prevent it from being applied to the sections I am dealing with .
The sections are sensibly less useful than the circular section , their torsional rigidity Fig. 2 .
Fig. 4 .
being always diminished and the maximum stress very often increased .
This remark , I may add , applies to all the sections dealt with in this paper .
This usefulness or efficiency decreases as the neck of the section becomes more narrow , as , indeed , might have been anticipated .
Other sections worked out are those corresponding to angles between the asymptotes of ( fig. 3 ) , ( fig. 4 ) , and ( fig. 5 ) ; in the latter case the sections degenerate into ordinary elliptic sections with two straight slits , or indefinitely thin TORSION OF CERTAIN OF keyways , cut into them along the axis , as far as the foci .
The stress at the foci , however , is then theoretically infinite .
It is interesting to see how .
as we make the bend the foci ) , the values of , for which the two -points break up into four , becomme and until , when the angle between the asymptotes of thu hyperbul .
is less , the greatest stress always occurs at the neck of the section .
The limiting case of such sections , when the angle the asymptotes is very small and the eccentricity of the ellipse nearly unity , the distance between the foci being very great , gives us the rectangle .
I then pass on to the section bounded by one ellipse and one confocal hyperbola .
In the limiting case when the ioci coincide , we obtain the sectul of V0L .
CXCIlI .
Of this I have worked out numerically three cases , in each case taking two ellipses ( 1 .
) The semi-ellipse ( fig. 6 ) .
( 2 .
) The ellipse with a keyway cut into it of the shape of a rectangular confocal hyperbola ( fig. 7 ) .
( 3 .
) The ellipse with a single slit cut into it ( fig. 8 ) .
The most striking of the results is in reference to the reduction of the torsional rigidity of the ellipse in case ( 3 ) .
This reduction of rigidity decreases rapidly as the depth of the notch decreases .
The rigidity , which is reduced by as much as 23 per cent. when the depth of the keyway is as great ( semi-major axis ) , falls only about 1 per cent. when this depth is 12 ( semi-major axis ) .
Possibly this may throw some light on the fact that the effect of cutting such slits into the material does not always give in practice the reduction in the torsional ligidity which should have been expected from SAINT-VENANT'S results for the circle .
Clearly the depth of the keyway is a factor of the very first importance , and keyways of moderate depth will produce a comparatively small effect on the torsional rigidity .
It is also shown that the effect of cutting two equal and opposite slits is practically equal , iu the two cases which I have calculated namely , , to twice the effect of a single slit .
It seems , thel.efore , that the study of these sections brings to light several interesting facts in the of elasticity , will weII repay the trouble involved in dealing tho long and somewbat tedious algebra and arithmetic which lead to these results .
S3 .
Statement of Notation .
, In what follows the axis of will be taken parallel to the generators , the axes of and in one of the terminal cross-sections .
The origin , however , will not necessarily be at the centroid of the cross-section .
The shifts of any point of the material parallel to the axis will be denoted as usual by , and for the stresses I shall use the notation of TODHUNTER and PEARSON'S ' History of Elasticity , ' denoting the stress , parallel to , ; across a plane element perpendicular to Then if , following SAINT-VENANT , we suppose the terminal cross-section to be fixed that , then if be the angle of torsion .
unit length , .
( 1 ) .
If be the modulus of rigidity , ( 2 ) , and all the other stresses are zero .
TORSION OF CERTAIN FORMS OF SHAFTING .
is then determined from the body stress equation or which holds at all points of the cross-section , and from the -stress equation that is , ( 4 ) at the boundary , where is an element of the outwards normal , and are its direction-cosines .
The above conditions allow us to determine uniquely .
They are associated with the condition that the parallel to the generators through the origin remains fixed .
If , however , we take any other parallel to the generators through the point to remain fixed , then , and the equation at the boundary becomes , .
Now , if instead of .
we write ( ay\mdash ; bx ) , then and the equations to determine are the same which we bad before for .
It follows that the stresses in the cylinder are unaltered , whatever be the parallel to the generators about which it is twisted , the effect of the change being merely to introduce a term ay \mdash ; into , which corresponds to a rigid rotation about an axis joining to the origin .
It follows from this that in dealing with stresses due to torsion we may take our origin wherever it is most convenient . .
L. N. G. FILON ON THE RESISTANCE TO S4 .
Analytical Work for Sections Bounded by One Elliptic and Two Hyperbolic Arcs .
onsider now the transformation If we allow to vary between and , and between and , the point will move within the space contained between the ellipse the two hyperbolas Using then the coordinates instead of we find that our equations for become for . . . . .
( 3 ) .
Also , when ( 4 ' ) .
, when or Write now Then ( 5 ) .
( 6 ) .
\mdash ; Let us assume where Then conditions ( 5 ) and ( 3 ) are identically sfied .
Let us now determine the coefficients A and so that ( 6 ) shall also be satisfied .
TORSION OF CERTAIN FORMS OF SHAFTING .
We have to expand between the limits in a series of smes as follows : The coefficients , .
are found in the usual way also we have , substituting in ( 6 ) , equating coefficients of , and , , whence , solving for and , we find sech ( 7 ) , ( 8 ) , whence cosh ( Q ) .
.318 .
L. N. G. FILON ON THE RESISTANCE TO Having obtained ?
, the shears are easily deduced by simple differentiation ( 10 ) , where These again may be put into the slightly different form sech 2 sech 2 .
The next quantity which we require is the moment of the shears sech 2 sech 2 sech 2 sech 2 TORSION OF CERTAIN FORMS OF sech 2 \mdash ; But the series can be summed in finite it is tound to be - sech .
Substituting this value in the expression the torsion we find the two series terms .
Whence , after some obvious reductions - cos ) ( 11 ) .
*\amp ; e CnRySTAL'S ' Algebra , ' vol. 2 ( ting the result marked ( 8 ) on 1 ) .
338 ) .
S5 .
Alternative Solution for the \amp ; ctions .
There exists also an alternative solution ; it is general of a less convenient fonn than the one last given .
It may , however , be useful in certain cases .
eturn now to the boundary conditions ( 4 ) and write we then obtain when or when The latter condition may be written .
Now let us write where and But ( 12 ) sin . . .
( 13 ) when and Assume we have now to express between limits TORSION OF CERTAIN FORMS OF SHAFTING .
We find Substituting in ( 12 ) and ( 13 ) and equating coefficients we obtain sech sech whence sinh ' .
( 14 ) .
It may be noted here that may vary between and .
It follows that may have the value , and in that case the denominator of the first term under the first 2 becomes zero .
The same happens to the denominator of the first term of the second 2 when has the value .
Further , in this latter case the first term in also becomes infinite , so that the expression ( 14 ) is apparently no longer applicable .
It is easy to see , however , that the terms , which are apparently infinite in ( 14 ) , exactly cancel each other .
If we write , where is small , simplify and proceed to the limit , we find that when the two infinite terms reduce to : \mdash ; ( 15 ) , which is finite .
VOL. CXCIII.\mdash ; A. 2 In like manner when the limit of is easily evaluated .
For all other values of and the series in ( 14 ) are absolutely convergent as they stand , for all points withit ?
or on the boundary of the section .
The same holds of all the series in the last paragraph for all values of The expression for being given , the shears and torsion moment are obtained as before \mdash ; The integrations are all easily effected , and we find Remembering that , reducing , and regrouping the terms , we find finally ( 16 ) .
We may test the correctness of the expressions ( 11 ) and ( 16 ) by remembering that TORSION OF CERTAIN FORMS SHAFTING .
when we make the distance between the foci very great , and and very small , the section reduces to a rectangle , of which the half sides and are given by The first and last terms in ( 16 ) are ultimately of negligible order , when muItiplied by The second and third reduce to \mdash ; The fourth term gives Hence which is one of SAINT-VENANT'S expressions for the torsion moment of a rectangle of sides If we treat in a similar manner expression ( 11 ) , neglecting terms of ordel greater than four in , we get the other expression for the torsion moment of the rectangle .
S6 .
Recapitulation of Results for the Symmetrical Case .
By .
the most important case we have to deal with is that in which the sections are symmetrical .
We have then , and therefore Both solutions then simplify a good deal , and we have the equivalent expressions \mdash ; sech ( 17 ) .
sech 2 siu .
( 18 ) .
sech 2 ' ' ( 19 ) .
- .
( 20 ) .
TORSION OF CERTAIN FORMS OF SHAFTING .
.325 S7 .
Importance of he Maxi , mum Stress .
to I now pass on to the numerical determination of the torsion moment and stresses , and in particular of the maximum stress .
SAINT-VENANT has shown , in his memoir on torsio that if we assume an ellipsoidal distribution of limiting stretch , stretch such that , when it is exceeded , the elasticity of t , he material is impaired ) , then , in the case of a shaft under torsion , we have at any point where are the shearing strains in the planes respectively , is the value of the timiting shearing of the material , and is the maximum value of the ratio of the stretch in any direction to the stretch in that direction .
The condition that there should be no failure of elasticity is herefore that Therefore and , since The points where this condition will first be broken are called by SAINT-VENANT the fail-points points dangereux They are clearly the points where is a maximum , i.e. , where the resultant stress across an element of the section is a maximum .
Hence the importance of determining the points of maximum stress .
Strictly speaking , the latter give us no certain information as to where , or how , rupture will actually take place : all that they tell us is where linear elasticity begins to fail .
But they will , in general , give us a useful clue to the regions where breaking may be expected to occur , and , in absence of definite theory of plastic deformation and upture , we must be content to be guided by the results of elastic theory .
I have worked out 1lumerically the values of the stresses at the points and .
These give the .
points in which the axes meet the boundary of the cross-section .
I have denoted them by A and respectively .
The boundary is convex at A and concave at B From considerations of symmetry it follows that A and must be points of ?
aximum or rainimum stress , and the stress being zero both at the and at the centre , it will often hapl ) that they are points of maximum stress .
When this is the case those of the points A and where the stress is numerically greater , will give us the .
But there is an obvious exception , when there are two points of maximum stress on either side of the mid-point , and this is a case which , we shall see , does occur in these sections .
* Memoires des Savants Etrangers , ' , vol. , pp. also TODHUNTER and PEARSON , Hist. Elast vol. , part , pp. 7-10 .
S8 .
Methods of Calculation .
The symmetrical sections selected for numerical treatment are those for which and , and , sixteen in all .
These sections are shown in figs. 2 to 5 .
of course is the complement of the half angle between the asymptotes , and all sections having the same have been collected in one figure .
The numerical calculations were generally based upon the formulae of S4 , as they did not require modification for the value of .
In many cases , however , the alternative series were used also , in order to test the results obtained .
The calculation of the terms of the series in the expression for the torsion moment was carried on until was so great that could be taken sensibly equal to unity .
The remainder of the series , namely , was then obtained by expanding the denominator by the binomial theorem , thus The successive terms were easily calculated from the values of given in CHRYSTAL'S ' Algebra , ' vol. 2 , chapter XXX , S15 .
The stresses at A and were calculated from formulae ( 18 ) and ( 19 ) .
If denote these stresses , we find easily .
( 21 ) , sech 2 ( 22 ) .
The first expressions for were used in each case as the main basis of the calculation , but the results were partly verified by means of the second expressions .
TORSION OF CERTAIN FORMS OF SHAFTING .
With regard to the accuracy obtained , I may say that , where I was able to verify , I found the values of the stresses correct to five , and sometimes to six , significant figures .
In the values of the torsion moment the first five figures generally agreed , so that , on the whole , the results may be considered correct to the number of figures given .
The first series for and the second series for very slowly convergent indeed .
The method adopted in dealing with them was the following : The series was broken up into two series The latter series converges rapidly , and was easily calculated .
The former was calculated to an even number of terms , and the remainder obtained by means of the Euler-Maclaurin sum-formula , thus : Now 2 \mdash ; .
, where , \amp ; c. Let me write Then whence Put therefore , .
In like manner if , These series converge fairly rapidly if is at all large , and thus the remainder can be obtained .
Even with the help of all these devices the labour of calculat the moment and stress for the sixteen sections was considerable .
The values of the hyperbolic functions were taken from Tables ( ' Theory der Potenzial oder Cyklisch-hyperbolischen Functionen and from GLAISHER'S and NEIIMAN'S Tables of the Exponential Function ( ' Cambridge Phil. Trans. , ' vol. 13 ) .
S9 .
Values of the The first quantity calculated was the torsion moment .
The values found are shown in the table below .
TABLE of It is interesting to compare this table with the table of values of the torsion moment , as given by empirical formula , torsion moment where area , moment of inertia of section about its centroid .
Calling this value of the torsion moment , we have whence we obtain the following set of values:\mdash ; TORSION OF CERTAIN FORMS OF SHAFTING .
TABLE of \mdash ; - If we compare this with the preceding table , we see at once that although the agreement is fairly good for the more compact sections , empirical formula utterly breaks down for deeply indented sections .
That , indeed , might have been expected , since it takes no account of slits cut into the material .
One rather noticeable feature in the comparison is that SAINT-VENANT'S formula always gives too high a value for the torsional rigidity .
S10 .
Comparison with the Circle .
" " ) \ldquo ; Rigidity .
In order , however , to compare properly the efficiency or usefulness of these various sections , it was found advisable to refer each of them to some kind of standard , or unit .
The most obvious standard , as I thought , was the circular section , this being the one whose torsion obeys the most simple laws .
I determined , therefore , to compare every section with the circular section of the same area .
Now if be the radius of this circular section , its torsion moment and the maximum stress .
To find , we have the equation : area of given section , whence The values of for the various sections are easily found .
VOL. CXCIII.\mdash ; A. 2 TABLE of whence we obtain the following table giving us When we look at this table , we observe immediately that the torsional rigidity decreases , compared with the torsional rigidity of the circular section , as we increase , that is to say , as we decrease the thickness of the neck with regard to the other linear dimension .
This indeed might have been expected , for it is clear that such a process must weaken the rigidity enormously , inasmuch as it tends to render the two halves of the section independent of each other .
When , the ratio is greatest .
In this case the section does not deviate very much from a square .
( For very small values of and the section is , of course , a rectangle .
) This result shows us therefore that , so far as torsional rigidity is concerned , the square is a more efficient form of section than any one of those dealt with in the present paragraph .
That the circle is a more efficient type of section for rigidity is quite evident from the table , since all the values in it are less than unity .
It may be interesting to note what are the values of for the ellipse .
When we use the values given by SAINT-VENANT in his memoir on torsion and , if TORSION OF CERTAIN FORMS OF SHAFTING .
.331 Hence we have , for the full ellipse If we compare this with the previous table , we see that for the flattest ellipse , , the ratio of the torsional rigidity to the torsional rigidity of the circular section of equal area , which I propose to call for brevity the relative torsional rigidity , is greater for the truncated than for the full ellipse , except in the last case , This last must necessarily be , since the strength of the section should be reduced by cutting two slits into it along the major axis .
For the higher values of we see that the relative torsional rigidity is always greater for the full than for the truncated elipse .
S11 .
of the Stresses at the Points of Symmetry .
Passing now to the values of the stress , the values of are given , for the sixteen symmetrical sections , in the table below .
TABLE of The values of were only calculated for and , it being clear that for the given values of would be less than for .
The values of are given in the following table:\mdash ; TABLE of The values of , as calculated from the formula ( 21 ) above , actually turn out to be negative .
The sign is , however , of no importance .
Several important results are seen to follow at once from these tables .
In the first place , when a keyway cut into a shaft of elliptic cross-section reduces to a mere slit , the stress at the inner extremity of the keyway is seen to be infinite , although this keyway is the limit of a single continuous curve and not of two curves making a sharp angle , as is the case for a slit along a radius of a circle , obtained as the limit of a keyway of the shape of a sector of the circle .
Such slits are thus bound to produce rupture plastic flow of the material at their deepest points , whatever be the manner in which we approximate to them in practice .
The second point of importance , which these tables bring out clearly , is that the maximum strain .
and stress do not always occur , as most of the results obtained by SAINT-VENANT would lead one to suppose , and as THOMSON and TAIT ( ' Natural Philosophy , ' vol. 1 , Part II .
, S710 ) , and BOUSSINESQ ( ' Journal de Mathematiques , ' Serie II .
, vol. 16 , p. 200 ) assumed , at the point of the boundary nearest the centre .
Indeed SAINT-VENANT himself , in his edition of NAVIER'S Legons de Mecanique ' ( S33 , p. 313 ) , has given an example to the contrary , and it happens that the section dealt with in this example is closely analogous to the sections of fig. in this paper .
The shape of the section is reproduced in fig. 11 ( p. 342 ) , from ' Legons de NAVIER .
' He calls it a " " section en double analogue celle d'un de che de \ldquo ; He gives two numerical examples in which the ratio of breadth to length of the section is 20 and 14 , corresponding for our sections , when , to and respectively .
He finds in these two cases that the failpoints are not at the point of symmetry on the contour which is closest to the centre , but at points on the contour at a distance from the axis of symmetry of 46 and 52 of the half-length respectively .
Now it is easy to see from the tables above that the result which one would expect according to the ordinary rule , namely , does hold in fifteen out of the sixteen cases , but there is one exception in the case of the section , when the greater of the two stresses is found to occur at , the point further from the centre .
I was much struck at first by this apparently solitary deviation from the rule , and was inclined to ascribe it to some error in the arithmetic .
In order to test this , I calculated the values of and for the neighbouring section , .
I found the previous exception .
I then took the expressions ( 21 ) and ( 22 ) for and , and tried to determine the limits to which they tended , when was made very great .
TORSION OF CERTAIN FORMS OF .3.3.3 Clearly , when we look at the second of expressions ( 22 ) , we see that the second term must ultimately become vanishingly small , provided , or For such values of , then , tends to the definite limit , i. e. , tends to the limit for large values of For values of it would seem that increases numerically to an indefinite extent .
Consider now the second expression for .
Clearly , if exceed a certain yalue , approximates so closely to unity that we may replace it in the serice by unity , and the error will only be a small fraction of the series itself We then find tends to } when is large .
Substituting , and using the Euler-Maclaurin sum-formula to calculate the series , I find , and remembering that , if large , , if large , we see that , when is large , , and , therefore , increases numerically nitely .
On the other hand , if be very small , we get a flat section , A being now the ) nearest to the centre .
Looking at expressions ( 21 ) and ( 22 ) we see easily that tends to and to sech 2 or , neglecting squares of compared with the first power , Now lies between 1 and 8/ 9 , hence .
N. .
FILON ON THE RESISTANCE TO and being always is always numerically greater than for small values of .
This confirms the usual rule , which we should expect , since the section approximates in this case to a flat rectangular section .
S12 .
Discussion of the Variations in these Stresses .
The variations in the stresses are shown in fig. 9 ( p. 340 ) .
This figure gives , not the values of the stresses themselves , but the ratio of the stresses to the maximum stress in the circular section of equal area .
This ratio is plotted as ordinate to the various values of as abscissae .
Of course , , as given by the expressions ( 21 ) being negative throughout , the curve shows the ratio and not The diagram is comparatively rough , especially near the origin , owing to the very limited number of points which I could calculate .
The value of selected was When is it is not difficult to show that , for and when is large These last enable us to see the form of the curves near the origin and at a great distance from it .
They are perpendicular to the axis of at the origin , and at first the curve of lies below that of .
At some point between and they cross .
This corresponds to the case of the square for rectangular cross-sections .
is now less than , but instead of its remaining so , as we should have expected , the curves cross again near th .
value of .
The curve of now tends to become practically a straight line parallel to the axis , at a distance from it , and the curve of approaches the axis asymptotically .
The values of are given in the tables below .
TABLE of TORSION OF CERTAIN FORMS OF SHAFTING .
TABLE of S .
Investigation of the Fail-Points other than the Points of Symmetry .
The question now arises , are A and really points of , and therefore shear ?
Assuming that the fail-points do occur on the contour , are we sure that there are not other points on the boundary where greater maxima occur , and may we not be in presence of a case like that of SAINT-VENANT'S section en double spatule , To see whether this is so , it is necessary to go at some length into the equations determining the maximum stress .
Consider first the sides of the section .
It is clear that the resultant stress along this side will be simply the component parallel to the contour , since the normal component must vanish in virtue of the boundary conditions .
Calling this resultant stress , we find easily where .
Hence we have to make a maximum with regard to We have therefore that is Take now the second value of as given by ( 17 ) , and we get 2 \mdash ; Hence the equation giving the maxima and minima is sech ) .
One root of this equation is .
Now it is clear that since is zero , and therefore a minimum at , if it be not a maximum at , then there must be a maximum somewhere between and To find whether is a maximum or not , we have to investigate the sign of i.e. , we have to investigate the sign o \mdash ; all the other terms vanishing when .
Now being always positive , we have to investigate the Slgn of \mdash ; when and therefore ultimately the sign of TORSION OF CERTAIN FORMS OF SHAFTING . .
sech sech I may remark here that all these differentiations are permissible , provided that be less than by a Jinite , however small , the series being then uniformly convergent .
We may not , therefore , make actually zero in the last expression .
If , however , is small but finite , then it is easy to see that must be negative if .
Also if algebraically increases continuously , until for a certain value of it reaches the value zero .
For all higher values of it remains steadily positive .
S14 .
Critical Values of and It follows , therefore , that in all sections for which , the maximum stress along the sides occurs at the point , until a certain critical value , is reached , when the stress at becomes a minimum , and we now have two points of maximum stress on either side of B. When is apparently infinite , but it eally tends to a finite .
If we put where will ultimately be made very small , and neglect terms in \amp ; c. , we find that the expression becomes sech sech sech \mdash ; sech 3 \mdash ; secl ] When is very great , this will ultimately be positive .
Hence , here also we shall have a critical value .
This value , however , is easily seen to lie outside the values of taken in this paper .
I have investigated so carefully the cases when , as , for the particular sections selected , the maximum stress certainly occurs at B. One point , however , is .
When is very large , the second settle the sign of E. The sign of these terms , settled of the leading terms .
is negative if is positive , i.e. , il VOL. CXCIII . .
2 X This will always be the case , provided be greater than the root of the equation This root is nearly .
Hence , for values of the maximum stress always occurs at the thinnest point of the section .
This is the case for the sections of fig. 4 with broad keyways , .
which Returning to the case of sections for which we have given by the equation Putting in this we have for the transcendental equation giving in this case sech Now , for fairly large values of , we may neglect all terms in this summation except the first .
We then have ( sech sech ) whence I find Hence for values of greater than this the maximum occurs at certain points on the contour , given by previously obtained equation in S15 .
Calculation of the Position of Fail-Points and the Jlagnitude of the Maximum Stress for the Sections We have , therefore , if we wish to find the maximum stress for the sections and when , to solve this transcendental equation : which may be put into the slightly simpler form TORSION OF CERTAIN FORMS OF SHAFTING .
I find the roots of this to be approximately given by when 821 when The corresponding values of are found to be and respectively .
Comparing these with the values of given on p. 331 , we see that the values of the stresses corresponding to the maximum on the broad side are the greater , and hence we have really a case absolutely analogous to section en spatule , with four fail-points symmetrically distributed , all of them lying on the broad sides of the contour .
S16 .
is great .
It is interesting to see to what limit the fail-points tend , when is great .
We have Replace now by its equivalent .
Now if be great , we have approximately .
Hence , if we suppose where is finite , so that we are dealing with points whose distances from the centre bear a finite ratio to the dimensions of the section , we find terms negligible in comparison and ultimately vanishingly small when is made infinite .
Hence we have to make a maximum .
Differentiating , the series being absolutely and uniformly convergent , we . .
This is the equation giving .
When we put in it I find from this approximately , and the corresponding value of .
Poeferring to the value of given on p. 333 , we see that the real fail-point is on the broader side .
The curve of true maximum stress is shown also in fig. 9 .
It joins on to the Showing the variation in the value of as increases , at the three points , of Fig. 2 .
const .
curve of at the point corresponding to the value .
It then remains above the curve of , tending ultimately to a straight line parallel to the axis , at height .
above it .
S17 .
Proof that there can be no other It remains to show that the point corresponds to a maximum along the side , and also that there is no maximum of the resultant stress inside the section .
The stress along is given by TORSION OF CERTAIN FORMS OF SHAFTING .
The equation giving the maxima or minima is found as before .
It is \mdash ; or , using the first of expressions ( 17 ) for The left-hand side is always positive .
Hence there is no root except , and that corresponds to a maximum .
That no absolute can exist inside the section can be proved as follows:\mdash ; We have Suppose at any point inside the section is a maximum .
The above forms for being independent of axes , let us take for axes of and the direction of the resultant stress across the section at and the perpendicular to it .
So that at P. Consider a near point .
Let be the stresses at .
Then , since is a maximum at but Therefore or is a numerical maximum .
But since therefore also ; and it is well known that no function can have an absolute maximum or an absolute minimum inside a region where it satisfies LAPLACE'S equation .
This is in fact a particular case of the general theorem that a potential cannot have a maximum in free space .
: Hence we have proved that the only fail-points are those which we have already investigated .
S18 .
Deductions from the above and Criticism .
of BOUSSINESQ'S Proof of the Position of Fail-Points .
Thus we see the study of these symmetrical sections is extremely instructive as regards the position of the fail-points .
They show us the connection between the rectangular section and the section with a neck , and they give us the limiting cases , the four fail-points coalesce into two , and vice versed .
The four fail-points begin to occur after the ratio of the long to the short axis of the section exceeds a certain value , which depends upon the angle of the bounding hyperbolas .
As the indented appearance of the section becomes more obvious , that is , as increases , this limiting value becomes greater and greater until , when the angle between the asymptotes is less than the fail-point is always at the vertex of the hyperbolas .
But in no case are the fail-points on the convex sides of the sections , unless the ellipses are so flat that the points A are nearer to the centre than the points B. M. BOUSSINESQ has given ( ' Journal de Mathematiques , ' Serie II .
, vol. 16 , p. 200 ) a sketch of a proof that the fail-points must be sought for " " sir les petits diamStres Fig. 11 .
des sections As this statement is in opposition with the results of the present paper and with DE results for the rail sections already mentioned , I venture to suggest that M. BOUSSINESQ'S reasoning hardly holds in the case of sections part of whose contour is convex and part concave , for the following reason .
The problem of torsion is mathematically analogous to that of a cylindrical vortex of uniform strength , whose cross-section is that of the shaft considered .
The motion being in two dimensions we have a stream function , and the resultant stress at any point in the torsion problem is the same as in the hydrodynamical analogy , being an element of the normal to any line .
Now if we draw the stream-lines for equidistant values of , they will , says M. , " " reproduce the irregularities of the contour , but more and more faintly , so that the curves are spaced at greater intervals along the large , then along the small diameters In consequence , , or the stress , is greater in the latter case .
The above argument assumes that the same number of stream-lines cross each TORSION OF CERTAIN FORMS OF SHAFTING .
diameter .
This will undoubtedly be the case , and M. BOUSSINESQ'S proof will hold , if the boundary of the section be everywhele convex , for then it is evident that each stream-line will consist of a single simple oval ( fig. 10 ) .
But when we deal with sections like those of the present paper it is by no means so clear that this will be the case .
There may be stream-lines which consist of two separate ovals ( see fig. 11 ) , and it then becomes very much open to question towards which part of the section the lines will be most crowded .
In fact in some cases , as we have seen , this will occur at intermediate points , ( fig. 11 ) , and in other cases the narrowness of the neck or the sharpness of the bend will counterbalance the limited number of the stream-lines through the neck , and the fail-points will be at , B. S19 .
Comparison with SAINT-VENANT'S Results .
" " Efliciency\ldquo ; of the various Sections .
Wheri we compare the results given by these sections bounded by confocal ellipses and hyperbolas with those given by DE SAINT-VENANT in his edition of the ' cons de NAVIH : R ' for the " " sections en double spatule we find very good agreement .
Thus the critical value of the ratio of breadth of neck to length for which the fail-points split up each into two is given by SAINT-VENANT as 3247 .
For the sections of this paper , when find this critical value to be 3215 .
The following are the principal numerical points Lecons NAVIER , tion of breadth of nerk to longth . . . .
1250 . . . . .
3955 . .
tio stance o-points from short axis to length of longer semi-axis . . . .
4486 On the whole the sections of the present paper appear the more useful , and the quantity being greater than for DE SAINT-VENANT'S rail sections .
This .
L. N. G. FILON ON THE RESISTANCE TO quantity I opose to call the " " efficiency\ldquo ; or " " usefulness\ldquo ; of the shaft .
It is equal to , and gives us the ratio of maximum stress to .
torsional rigidity , compared with the same ratio for the circle of equal area , that is , it gives us a measure of how much torsion we may put into the shaft without impairing its elasticity .
Thus if be the maximum torsion moment which the given shaft will bear without failure of elasticity , the maximum torsion moment which the shaft of equal circular section ( made of the same material ) will stand , then the corresponding stresses , are each equal to the limiting elastic stress of the material and efficiency or the limiting torsion moment of a shaft of any section is obtained from that of the circular section of the same area by merely multiplying by the " " efficiency\ldquo ; as thus defined .
TABLE of " " Efficiency\ldquo ; of the sixteen given sections .
A glance at the above suffices to show that the efficiency of these sections is , in general , about one-half , that is , on the whole , this form of shaft is about half as useful as the shaft of circular section .
The zero efficiency in the case of the slit is due to the infinite stress in the keyway .
S20 .
Analysis for the Sections bounded by one Elliptic and one Hyperbolic Arc. I now pass on to the consideration of cross-sections , bounded by an ellipse and a single branch of a confocal hyperbola .
Such sections are shown in figs. 6-8 .
In this case we shall find it more convenient to define and by the equations ( 23 ) where TORSION OF CERTAIN OF SHAFTING .
We find in the same way as before - .
There is , however , in the present case , a further condition .
For , referring to figs. 6-8 , when we cross the line being that focus of the conics which is inside the section , is continuous , passing through the value , but is discontinuous and changes from positive to negative .
We have to ensure that the function of and , which shall represent , shall be continuous in crossing , and that its space-differential coefficients shall also be continuous .
The latter ldition implies that must ; this being so , all the conditions ) satisfied .
Let us write conditions ( 24 ) then become when .
when If we now assume then the conditions of continuity and the first of ( 25 ) are identically satisfied .
is found from the second equation of ( 25 ) in the same way as in S4 .
And thus \mdash ; ( 26 ) .
\mdash ; VOL. .
L. N. G. FILON ON THE RESISTANCE TO The shears are given by the equations sech 2sech 2 , here J now stands for the quantity The torsion moment sech 2sech 2 sech 2 sech 2 When this is integrated out and reduced as before , it is found that .
S21 .
Alternatire Solution for these Sections .
For this type of section also we can find an alternative solution .
Suppose we assume These conditions ( 24 ) reduce to : ( 29 ) .
AIso the condition of continuity requires that must be odd in OF AIN 1IS OF AFTING .
We assume , therefore , But since . .
, where ' we find easily the first equation of ( 29 ) , comparing coefficients ' and .
The stresses and torsion moment are deduced without difficulty .
They are : ( 31 ) . .
( 32 ) . .
The same remarks which were made concerning the solution in S 5 apply here .
The critical values are .
The limits to which the values of the stresses , given in ( 30 ) , tend , are easily obtained if required .
In practice , however , the other solution would probably be used .
S22 .
Resulfs .
Effects of Keywa.upon Torsional Rigidity .
I have worked out numerically six cases of this type of section .
The sections selected are shown in figs. 6-8 .
The bounding ellipses are and , the bounding hyperbolas are and , giving respectively ( a ) the half ellipse , ( b ) the ellipse with a rectangular hyperbolic keyway , ( c ) the ellipse with a single thin keyway or slit .
The values of the torsion moment and of its ratio to the torsion moment of the equal circular section , are shown in the tables below .
TABLE of TABLE of When we look at these results we see that the torsional rigidity of these sections is always less than that of the circular section .
The sections consisting of a complete ellipse with one fine slit up to the focus are weaker than the half-eHipss or the sections with a broad keyway .
This is more particularly shown in- the case of the more elongated ellipse , With regard to the effects of slits , or thin keyways , it is interesting to compare the values of for ellipses , when are no slits , ( ii .
) there is one slit , ( iii .
) there are two slits TORSION OF CERTAIN OF We find .
It follows that , in the first case , the cutting of one thin keyway lowel .
S the of the section by per cent. , and of two keyways by per cent. Hence the effect of two such keyways is slightly greater , if anything , than twice the effect single keyway , in the case of the more elongated ellipse .
The difference is , however , practically negligible .
In the other case the result is different .
The reduction of the torsional rigidity is very great : it amounts to per cent. for one keyway and to per cent. two keyways .
Here we see the effect of two keyways is rather less than twice the effect of one .
We may infer , however , from these two results that we may in , without very large error , if we have a number of keyways cut symmetrically into a section , and we know the effect on the torsional rigidity of any single keyway , assume that the effect of all the keyways is the sum of their separate effects .
Another important point which is brought out by the above results is that the effect of such a keyway upon the torsional rigidity is by no means simply proportional to the depth of the keyway , but increases according to some much more rapid law .
Thus , for the ellipse , the depth of the keyway ( semi-major axis ) .
For the ellipse , the depth ( semi-major axis ) .
Thus , when the depth of the keyway is only decreased to one-fifth of what it was before , the reduction of torsional rigidity falls from 23 per cent. to 1 per cent. , nearly in the ratio of the squaj.es of the depths of the keyways .
This result may explain the fact that , when keyways of only moderate depth cut into shafts , the decrease of torsional rigidity is by no means so great as would have been inferred from DE results for a circular section , with a thin keyway or slit extending right up to the centre .
If we suppose , which appears reasonable , that the effect of such a slit upon an ellipse , which is not very elongated , does not differ much from the effect on a circle , we see that a keyway , whose depth equals about one-eighth of the radius , will decrease the torsional rigidity by only about 1 per cent. Now , when we make , we get the case of the clrcle with a keyway going right up to the centre .
The reduction of torsional rigidity is then 44 per cent. about .
MR. L. N. G. FiLON oESISTANCE ?
Hence we have , roughly , If the reduction of torsional rigidity were simply proportional to the depth of the keyway , the last two results ought to be per cent. and per cent. respectively .
It may be objected to these deductions that in the above sections , the fact that the stress at the vertex of the keyway is infinite , violates the physical conditions assumed by the theory of elasticity and renders our results untrustworthy .
My answer to this is that these cases should really be considered as limiting cases .
If , instead of considering a section where the keyway is actually a straight line , we consider a hyperbola with a very sharp bend , we can easily ensure , provided the angle of torsion be not too great , that the physical conditions shall not be violated , and , on the other hand , the values of the torsional rigidity ( since they tend to a finite linnt ) will differ but little from the values obtained above .
The very fact that the torsional rigidity tends to a finite limit shows that , even in the extreme case , the area where the physical considerations are violated is infinitesimal .
Finally , in drawing conclusions from such results we shall only be following the example of SAINT-VENANT and of THOMSON and TAIT , who have not hesitated to use results found in a precisely similar way for such keyways cut into a circle .
S23 .
Values of the Stresses .
I have also determined the stresses at three points on the boundary of the section , in order to find where the stress was greatest .
The points selected are denoted in figs. 6 , 7 , and 8 by the letters , C. A is the vertex of the hyperbola , is the point opposite to , and is the point corresponding to A is given by , and by If denote the corresponding stresses , I find sech 2 TORSION OF CERTAIN IS OF SHAFTING .
( 35 ) , sech 2 ( 36 ) , whence I find\mdash ; TABLE of Stresses .
We see that the greatest of the three stresses occurs at , i. e. , at the vertex of the hyperbola , and it is probable that A is the true fail-point .
The following are the values of having the same meaning as in S12:\mdash ; of ON THE RESISTANC TO TORSION OF CERTAIN FORMS OF SHAFTING .
The values of the " " efficiency\ldquo ; are easily obtained .
TABLE of E. The above results do not need any detailed discussion .
We see that in all cases the maximum stress is greater than the maximum stress for the circular section .
Also the efficiency is always less than unity .
If we compare these values of the efficiency with those in S19 , we see that , on the whole , the rule holds that the more compact section the higher its efficiency .
On the other hand , by indenting a section we render it less efficient . .
Conclusion and Looking back upon the results of the paper , we see that the study of these special forms of cross-section sheds new light upon several little-explored parts of the theory of elasticity .
It confirms to a great extent DE SAINT-VENANT'S investigations concerning the behaviour under torsion of a rail , or of shafts of similar section .
Owing to the great generality of the forms treated , it enables us to correlate the results previously obtained for sections of various shapes , especially with regard to the maximum stress .
It shows us what type of cross-section will give us four failpoints not at the points of the contour closest to the centre ; within what limits we may expect to find this exception to the ordinary rule ; and in what manner this case passes into others .
Again , with regard to the effect of keyways upon the torsional rigidity , the results of the paper tell us that , without risk of sensible error , we may in practice , in order to get the joint effect of two indefinitely thin keyways or slits , add the known effects of each keyway , taken separately .
Further , these results bring the theory of elasticity into better accordance with observed facts , by showing that the effects of keyways of moderate depth are comparatively much smaller than would have been expected ffom the results for the circle .

