PROCEEDINGS OF * THE ROYAL SOCIETY .
Section A.\#151 ; Mathematical and Physical Sciences .
On the Viscous Flow in Metals , and Allied Phenomena .
By E. 1ST .
da- C. Andrade , B.Sc. , Physics Research Scholar , University College , London .
( Communicated by Prof. F. T. Trouton , F.R.S. Received March 9 , ^-Read April 14 , 1910 .
) It has been observed* that for a lead wire , loaded well beyond the elastic limit , the extension after some time becomes proportional to the time , or the flow becomes viscous in character .
The rate of this viscous flow varies with the load , and the following work was undertaken to investigate the law of this variation , and the phenomenon in general .
Methods of Experiment .
The Hyperbolic eight .
The experiments were all done on wires of the metal , and the preliminary observations were all made on lead .
To suspend the wires , they were soldered into stout brass hooks .
In the first rough experiments the extension registered itself on a clockwork-driven drum ; these experiments showed that the rate of extension for a given load became constant for a time , but finally increased , and continued increasing .
This is due to two causes : ( 1 ) As the wire stretches , the length of wire being experimented on at any moment increases .
( 2 ) As the wire stretches , the cross-section diminishes , and thus the load per unit area or stress increases .
This is by far the more disturbing cause , since , as shown later , the rate increases much more rapidly with the stress than would be given by a linear law .
To * E.g.f Trouton and Rankine , " On the Stretching and Torsion of a Lead Wire beyond the Elastic Limit , " ' Phil. Mag. , ' October , 1904 .
VOL. LXXXIV.\#151 ; A. B 2 Mr. E. N. da C. Andrade .
On Viscous [ Mar. 9 , obviate this difficulty an automatic method of keeping the stress constant was devised .
Constant stress was obtained by letting the weight producing the stress sink into a liquid ( fig. 1 ) as the wire stretched , the form of the weight being so chosen that the upthrust at any moment was such that the effective load was inversely as .
the length of the wire at that moment , and thus was directly as the cross-sectional area .
The required shape is easily shown to be given by a hyperbola of revolution , = ./ S. J_ y v U+x ' where M is the mass of the load , Z0 the initial length of the wire , and the density of the liquid .
As water was always used as the liquid , p is fixed .
Having chosen a particular M and l0 , the exact size of the weight is given .
From the equation we see that a weight , once constructed , is only exactly efficient for one particular initial load , but with reasonable approximation the same weight can be used over a limited range of loads .
An additional advantage accruing from the device of the hyperbolic weight was soon discovered , viz. , that stresses could be freely applied much greater than can possibly be used with constant loading .
The possibility of working at much greater known stresses than is otherwise feasible opens up a new region for experiment .
j Description of Experiments .
Experiments were made on three kinds of wires :\#151 ; ( 1 ) Approximately pure lead .
( 2 ) " Fuse wire .
" An alloy of lead and tin .
( 3 ) Approximately pure copper .
Fuse wire was used , because it shows the viscous flow to a remarkable extent ; copper as being a metal usually supposed not to show a viscous flow .
The first experiments were made on lead .
The preliminary tests were made on ordinary commercial wire ; the last specimen , here called lead D , f Flow in Metals , and Allied Phenomena .
1910 .
] was supplied as pure lead by the London Electric Wire Company .
Most of the numerical results here given were obtained from this specimen , which on analysis was found to be 99'8 per cent , pure lead .
The analysis was kindly done for me by Mr. I. G- .
Rankin , B.Se .
, who also analysed the fuse wire , and found it contained lead , 21'5 per .
cent. ; tin , 78 ; 5 per cent. To observe the extension a cathetometer reading to 0*01 mm. was used .
The water into which the hyperbolic weight sank was contained in a vessel of much larger diameter than the weight , filled up to the level of a hole in the side , through which the displaced water escaped , thus securing constant level .
The weight was adjusted so as just to touch the surface by observation of the image in the water , the weight being previously dry ; this was to avoid uncertainty caused by the surface-tension rise .
Three different hyperbolic weights were used to cover the range of loads for the lead , a fourth for the fuse wire , and a fifth in the high temperature experiments , which were done with lead at 162 ' C. In these experiments the temperature was maintained by an oil bath kept at constant temperature by a Riechert regulator .
The wire was pulled upwards , the lower hook being hooked on to a horizontal bar running across , the bottom of the bath .
Short lengths of lead wire were used , about 10 cm .
, so that the temperature throughout might be constant , and the load was applied by a fine steel wire running over two pulleys , to which was attached a hyperbolic weight .
The friction of the steel wire on the pulleys was less than 10 grammes weight .
j Experimental Results.\#151 ; The general nature of the stretch in a lead wire loaded beyond the elastic limit has been noted by previous workers.* An immediate extension is followed by a continuous stretch which decreases in rapidity to a final rate , constant if the extension is sufficiently small to make negligible the increase in stress due to thinning .
The first result from the use of the hyperbolic weight was that even for large extensions the rate settles down to a constant value , if the stress is kept constant .
In fig. 2 the extensions are shown in the two cases of constant load and constant stress for the same initial length , and the same initial load .
It is to be noted that even when the stress is kept constant we should expect the rate of flow to increase again , owing to the length flowing at any moment increasing with the time ; this effect , however , would not be very noticeable in the case shown , as the rate of flow at the constant stage is not very rapid .
In the case of fuse wire , however , where the viscous rate is large , this effect is marked .
It wras found that taking the rate at different times in the viscous flow , and dividing by the mean length pertaining to each rate , * Trouton and Rankine , loc. cit. B 2 Mr. E. N. da C. Andrade .
On the Viscous [ Mar. 9 , BREAKS FIG 2 Tl M E ( mins ) a rate per uuit length was obtained , which was constant at different stages of the flow .
It was thus established that under constant stress the rate per unit length , once the initial effect has died out , is constant right up to breaking.* A series of results for lead under constant stress at various stresses was obtained .
The experiments at high temperatures gave similar extension curves , but the initial effect was much less marked .
In the extension curves for fuse wire the initial effect was smaller still ; the increase in viscous rate , owing to increasing length , can be easily seen in fig. 3 .
I was led to dividing the extension into three parts :\#151 ; ( 1 ) The immediate extension on loading .
( 2 ) An initial flow which gradually disappears .
( 3 ) A constant flow , taking place throughout the extension .
I shall call ( 2 ) the / 3-flow , as a measure of it is given by the constant / 3 in an equation to be dealt with later .
The constant flow per unit length has already been mentioned as , viscous flow .
A discussion of a conception of the mechanism of these flows is given in the theoretical section .
A , few experiments were done on copper with the automatic recorder before mentioned .
P. Phillipsf states that " the copper wires made no attempt * Breaking in such a case is merely due to a cumulative local thinning at some initial irregularity ; in the case of fuse wire an extension of more than 50 per cent , of the original length was frequently obtained .
t " The Slow Stretch in Indiarubber , Glass , and Metal Wires , when subject to a Constant Pull , " P. Phillips , JVJ .
Sq , , * Phil. Mag. , ' April , 1905 .
1910 .
] Flow in Metals , and Allied Phenomena .
\#151 ; FUSE WIRE w\gt ; TIME ( mins ) towards approximation to a linear function " ; I found , however , that when loaded near the breaking stress , annealed copper wires gave a linear flow which was still continuing after seven days .
A peculiarity of copper wire was that it showed sudden slips at irregular intervals , which , perhaps , may be called " copper quakes , " as being analogous to the geological slipping supposed to result in earthquakes .
These were not due to the recording apparatus .
In fig. 8 , which represents one of the tracings from the drum , both these | FIG 8 z TIME slips and the existence of a final viscous flow are evident .
In copper , the viscous flow is very small , the immediate extension very large in comparison to a metal such as lead .
Before considering variations with the stress , 'it was thought advisable to obtain an equation to the curve of length against time at constant stress , to see how to divide up the effect into initial flow and viscous flow .
For large values of the time , the length l is given by Cekt , where C is approximately constant , since , experimentally , \\lt ; j[i = cons^anb Since Phillips gives l = a-\-b log t for the case when there is no viscous flow , i.e , when kss 0,1 tried l = A { a + log t ) ekt .
Mr. E. N. da C. Andrade .
On Viscous [ Mar. 9 , This was not very successful .
After various attempts , l\#151 ; Z0 \#151 ; A was found to give a good representation of the initial part of the curve ; the form of equation Z= Z0(1 + / 3ZV ' was therefore attempted .
This fitted our lead curves surprisingly accurately throughout the whole range .
To show the closeness of the fit , the following tables for two stresses chosen at random are given , since such divergencies as there are , are too small for convenient diagrammatic representation :\#151 ; Table I.\#151 ; Lead D. Initial load , 3600 grammes .
Initial length , 38-8 cm .
Average temp. , 14'-5 C. Time ( in minutes ) .
Length observed .
Length from formula l = ( 39 -13 + 0 -57 Vi ) eO -0000695* . .
Stress .
\ ( .
-\#166 ; i 0 38 -8 + 39 T3 3600 1 39 *73 39 '70 2 39 -87 39 *85 4 40-04 40 *04 5 40*12 40*12 15 40-56 40 *57 * 25 40-85 40 *86 3597 35 41 -07 41 *09 40 41 -17 41 T8 80 41 -81 41 '81 130 42 -38 * 42'39 3597 250 43 -48 43'47 300 43 -86 43 -85 420 44 -68 44 -69 3601 500 45 -19 45 -19 550 45 -51 45 *50 3602 ' v. - 1 Stress given in grammes weight per area of cross-section of unstretched wire .
Table II.\#151 ; Lead D. Initial load , 1800 grammes .
Initial length , 9-9 cm .
Average temp. , 162 ' C. Time ( in minutes ) .
Length observed .
Length from formula l = ( 9-94 + 0-43 VZ ) e'-0106i . .
Stress .
0 9-9 + 9 -94 1800 1 10 -46 10 -48 2 10 *70 10 '71 3 10 -90 10 -90 5 11 -27 , - 11-26 1800 8 11 -76 11 -76 10 12 -08 12 -08 12 ; 12-40 12'40 14 12 -73 12 -73 15 12 -90 12 -90 1800 1910 .
] Flow in Metals , and Allied Phenomena .
7 The extension curves for fuse wire , when the stress was truly constant , were equally well represented by the formula .
Since the equation l \#151 ; l0(1 + ftt^)ekt fitted my results so well , it was thought well to try the equation l = Z0 ( 1 + ft$ ) for the case of a pure / 3-flow , when there is no appreciable final viscous flow .
Taking a set of readings from Phillips ' paper , * the equation / = 70(1 + was found to fit quite as well as his logarithmic equation .
To investigate this a little further , I did one experiment on / 3-flow in copper .
The following table exhibits the result:\#151 ; Table III.\#151 ; /3-Flow in Copper .
Time ( in minutes ) .
Heading .
3 -089 \#151 ; reading = Z\#151 ; Z0 .
0 -131 v# .
0 Loaded 2 2 *924 0-165 0-165 3 2 -900 0-189 0 -189 4 2-881 0-208 0-208 5 2 -863 0-226 0*224 6 2 *850 0*239 0-239 7 2 *840 0-249 0-250 8 2 -827 0-262 0*262 9 2 -814 0-275 0-273 10 2 -807 0*282 0-282 15 2 -767 0-322 0 *322 20 2 -740 0-349 0*355 Phillips states that there were considerable deviations from the logarithmic law when the stretch had become very slow , but , unfortunately , he does not state in which direction the deviations were , so it cannot be surmised whether my formula would afford a better fit .
The tables given , which are representative , suffice to show , I think , that l \#151 ; Z0 ( 14- ftt^ ) ekt is successful in representing the observed facts .
Iq is not the exact unstretched length , but a number slightly larger , probably the immediate length on loading , which could not be measured , being fairly small compared to 1 , 1+/ 3 ft becomes nearly constant for large values of t , as ti then increases very slowly .
We have for convenience called that part of the flow represented by the coefficient ft the / 3-flow ; is a measure of the purely viscous flow .
To see how ft and kvaried with the stress , and the quantity ( stress ) , or the " fluidity , " were plotted against stress .
Fig. 4 shows and kj ( stress ) for lead D at 15 ' , and fig. 5 the same quantities for lead D at 162 ' .
It is to be especially noted that ft tends to become constant for large values of the stress .
A form of curve similar to our ft-curves was found by Philiipsf for th * Loc .
cit. , p. 525 .
t Loc .
cit. , p4 526 .
Carves III a , 6 , e , d. Mr. E. N. da C. Andrade .
On the Viscous [ Mar. 9 , FIG 4\#151 ; --LEAD AT 15 ' STRESS FIG 5 \#151 ; LEAD AT 162 STRESS variations of his coefficient b with the stress in his experiments on copper , where he was dealing with a case involving no sensible viscous flow .
( Owing to the similarity of our equations , b and yd are measures of approximately Flow in Metals , and Allied .
1910 .
] the same effect .
) It is remarkable that / 3 tends to the same limit in the experiments at 162 ' as in those at 15 ' , though the corresponding values of k are very much greater in the former case .
Fig. 6 shows the variations of k/ s ( where s is the stress ) for fuse wire .
It is of exactly the same character as in lead , fi is not here shown , as it is so FIG 6--FUSE WIRE AT 15 ' STRESS 200C small in the case of fuse wire that satisfactory determinations were not made .
The curves of k/ s against s can be approximately fitted by hyperbolae having one asymptote parallel to the axis of s , and the other asymptote inclined at an acute angle to this axis .
Fig. 7 shows k/ s for an early specimen of lead , A. k was computed from the later readings of the extension , it being assumed that the / 9-flow had died out .
The values are therefore only approximate ; the figure is given as a LEAD A AT 15 STRESS 10 Mr. E. N. da C. Andrade .
On the Viscous [ Mar. 0 , confirmation of the character of the fluidity curve , the specimen being a softer lead than D. Table iV.\#151 ; Lead D. Stress = s in grammes/ initial area .
jjj k. ' 1 \#151 ; x 10s .
S Temperature , 15 ' C. 2000 0 -0008 1 2600 0 '0044 3000 0-0077 0 -000035 1 16 3600 0 -0145 0 -000069 1 -92 4000 ' 0 -0210 0 -000110 2 -75 4400 0 -0290 0 -000202 4-60 4600 0 -0334 0 -000322 7 -00 5200 0*0424 0 *00207 40 -0 5400 0 -0420 0 -00253 47 -0 5600 0-0440 0 -00345 61 -6 Temperature , 162 ' C. 700 0-002 o-oooi 14 1010 0 -0045 0-00095 94 1400 0 -021 0-0032 230 1500 0 -024 0 -0048 320 1600 0-041 0 -0060 375 1700 0-041 0-0083 488 1800 0 043 0 -0106 588 2000 0-043 0 -0216 1080 Approximate cross-sectional area of nnstretclied wire ** 2 *16 sq .
mm. Table V.\#151 ; Fuse Wire .
Temperature , 15 ' C. Stress in grammes per initial area .
k. II - x 106 .
t ' 5 730 0 -00072 1 -o 840 0 -00116 1-4 960 0 -00197 2 -05 1250 0 -00590 4-7 1280 0 -00628 4-9 1480 0 -0132 8-9 1505 0 -0146 9/ 7 1580 0 -0174 11 -o 1680 0 -0241 14 -3 1790 0 -0369 20-6 1885 0-060 31 -8 1985 0-090 45 -0 2090 0 -136 65 -0 Approximate cross-sectional area of nnstretclied wire = 0 *64 sq .
mm. 1910.1 Flow in Metals , and Allied Phenomena .
11 Theoretical Considerations .
There appears to have been but little previously written on the flow in metals which will aid in formulating a theory to accord with these results .
Boltzmann 's well-known paper on " Elastische Nachwirkung " * offers no help .
It is interesting , however , to consider the phenomena from the point of view suggested by Ewing in his Presidential Address to Section G- of the British Association on the structure of metals.f In this he insists upon the crystalline nature of the structure .
We may look at .
the initial flow as resulting from the addition to a permanent viscous flow of a flow which dies out with the time .
This latter may be taken as arising from a rearrangement or rotation of small parts of a crystalline nature in the material , much after the fashion of the rotation seen in Ewing 's magnetic model illustrating hysteresis in iron ; the / 3-curve is of the same form as the curve of magnetisation .
I assume the rotation of the crystals to be opposed by some kind of elastic force proportional to the displacement , but , as in the magnetic model , a succession of equilibrium positions is passed , through which return is not spontaneously possible on the removal of the stress .
The viscous flow may be regarded as taking place in a viscous matrix , in which the crystals are embedded .
If we take the formula l = 1$ ( 14-/ 3^ ) ekt , we may write rj]ft dt J 0 since / 3 and Jet are small within the range . !
The first term represents a purely viscous flow , the second the flow due to rearrangement or rotations , if we assume that the number of crystals rotating per unit time diminishes as t~i .
The third term , which is small , may be regarded as a small correction to the permanent flow arising from the rotations increasing the amount of material which at any moment is in a suitable attitude for purely viscous flow : it is proportional to the number of crystals which have rotated .
( The formula is not presumed to do more than represent the phenomena throughout any convenient experimental range ; for very large values of t it is incorrect , as it would give the number of crystals rotating iu an infinite time as being infinite , should probably be replaced by a function of t which approximates * L. Boltzmann , 'Wiener Sitzungsberichte , ' October , 1874 .
t J. A. Ewing , ' Report of the British Association , ' 1906 , p. 657 .
t For example , take lead for stress 4600 grammes per initial area .
Here the largest value of let was 0*079 , dl/ dt = 0*0065 + 0*0054 ( 1 + 0*079+ \#163 ; ( 0*079)2 + ... ) , and the omission of squares and higher powers of makes an error of about 0*15 per cent , in the flow .
12 On the Viscous Flow in Metals , and Allied Phenomena .
closely to ft for such values as are here considered , but which as t increases ultimately becomes constant .
) The fact that ft tends to the same limit at both 162 ' and ,15 ' is significant , as suggesting that the division of the flow into / 3-flow and viscous flow is valid physically , / 3-flow being dependent on a purely geometrical arrangement in the structure of the metal .
Such a process as the rotations suggested would give the experimentally observed constancy of ft for large stresses .
The following is a brief summary of the paper :\#151 ; ( 1 ) Establishment of the existence of a viscous flow in lead , in a lead-tin alloy , and in copper .
( 2 ) Realisation of the necessity of working at constant stress , and the device of the " hyperbolic weight " for this purpose .
( 3 ) Demonstration that at constant stress the flow is purely viscous right up to breaking for the metals used .
( 4 ) The empirical formula l \#151 ; U(1 -f ft ft ) ekt is shown to fit very closely all the extension-curves .
The flow can be divided into / 3-flow and viscous flow .
( 5 ) The / 3-flow is experimentally shown to tend to a limit for large stresses .
( 6 ) The y\#163 ; ?-flow is shown to tend to the same limit at 162 ' as at 15 ' C. ( 7 ) The curve of fluidity against stress is shown to be roughly a hyperbola , with one asymptote parallel to , and the other steeply inclined to , the stress axis .
It is a great pleasure to me to offer my thanks to Prof. Trouton , to whose unremitting kindness and encouragement whatever merit there may be in this work is largely due .
I also owe my thanks to Assistant-Professor A. W. Porter for many suggestions , and for reading through the paper .

