131 A High-Speed Fatigue-Tester , and the Endurance of Metals under Alternating Stresses of High Frequency .
By Bertram JIopkinsox , F.R.S. ( Received November 20 , \#151 ; Read November 23 , 1911 .
) The deterioration of metals under the action of stress varying rapidly between fixed limits has been the subject of much experimental investigation .
It is found that the important factor in the rate at which this " fatigue " goes on is the algebraic difference of the limits between which the stress varies , usually called the " range of stress " ; and that the absolute position of these limits matters little , provided , of course , that the mean stress is not too large .
The number of applications of a given range of stress required to fracture the piece increases as the range is diminished , the general nature of the relation between the two being as shown in the curve ( fig. 1 ) , which represents the results of a series of tests -S * $ -a 1 " EgJL . ?
Test on , mild Steel bars by Dr. Stanton , i ~ 1IOO r.ft .
7TL 'Tension , ZzmzZ = / .
09 times comfiression limit .
Jtrobatble vatue of hrruZing range 24'5 tonsfier sq.in , .
3 * s \amp ; 7 Jlever\amp ; aZs , hundreds of thous-ccnds made by Dr. Stanton on mild steel .
In these observations the stress alternated between compression and tension , the ratio of the compression and tension limits being 1*09 .
The form of the curve suggests that a range of stress not much below 25 tons , which in an average specimen would just cause fracture after a million reversals , could never break the bar , however often applied .
One of the chief objects of the fatigue tests hitherto made has been to discover this " limiting range .
" At an early stage in these investigations the question was raised whether K 2 Mr. B. Hopkinson .
[ Nov. 20 , the endurance by the material of a given cycle of stress is affected by the rate of repetition of the cycle .
Besides its intrinsic interest , this question is of importance because on the answer to it depends the possibility of reducing the excessive amount of time taken to carry out fatigue tests .
The determination within a few per cent , of the limiting range requires several separate tests in which the cycle is repeated at least a million times , and even that number is not always sufficient to give a reasonably close approximation .
Wohler worked with 60 to 80 reversals per minute , and he found that the same wrought iron which could just sustain a million applications of a range of 23 tons broke after 19 million repetitions of a range of 17| tons .
The more recent machines have been run at much higher speeds , and there is now a machine of the Wohler type at the National Physical Laboratory which gives 2000 cycles of bending stress per minute .
Even at this speed , which I believe is the highest yet reached under conditions admitting of accurate measurement , it takes eight hours to do a million reversals .
Within these limits it does not appear that frequency of application has much effect on endurance , that is , it takes the same number of repetitions of stress alternating between given limits to rupture a bar whether they be performed fast or slow .
It is true that the experiments of Keynolds and Smith* suggested that endurance was less at high speeds , but this has not been confirmed by more recent work.f The weight of evidence seems to favour the conclusion that speed is without effect on endurance , at any rate up to 2000 applications per minute .
Eor reasons which I discuss in detail below I think that this independence of endurance and speed is to be expected a priori so long as the speed does not exceed a certain limit , but that if that limit be passed the number of repetitions of a given range required to produce fracture will begin to increase , and it seemed probable that the increase might become noticeable if the speed were so high as 7000 per minute .
It was mainly with the object of testing this point that I constructed the machine described in this paper , which gives alternations of stress at the rate of over 7000 per minute , or three times as fast as the highest speed hitherto attained .
I may say at once that , in the case of the mild steel which has been examined , the endurance is undoubtedly greater at the higher speed , not only in number of reversals , but also in the actual time taken to produce fracture .
Hence the machine does not , as I thought it might ( if the speed effect * ' Phil. Trans./ A , vol , 199 , p. 265 .
t See " The Endurance of Metals , '5 by Messrs. Eden , Eose , and Cunningham , Institution of Mechanical Engineers , October 20 , 1911 .
1911 .
] A High-Speed Fatigue-Tester , etc. 133 revealed were zero or small ) , serve the practical purpose of reducing the time required for fatigue tests .
It seems rather to open anew field for investigation , which differs from that covered by the older slow-speed machines .
Description of the Machine .
The apparatus ( which is shown semi-diagrammatically in fig. 2 ) is designed to give alternations of direct stress between any desired limits of tension and compression at the rate of 100 complete cycles per second , or more .
The test-piece A is fixed at the lower end , by means of a nut B , to the screwed pillar C , which again is fixed into a massive block of cast-iron D , forming part of the base .
The base is bolted to a concrete foundation .
At its upper end the test-piece carries an iron weight E ( about 180 lbs. ) , to the upper side of which is fixed the laminated armature F by means of yokes Gr .
An electro-magnet H is carried by the pillars and cross-pieces I , above and independently of the armature , so as to leave a small air-space across which the magnetic pull acts .
The magnet is excited by alternating current in the winding H ' and exerts a periodically varying pull along the axis of the piece , whose frequency is twice that of the current.* * When the experiments described in this paper were practically complete I learnt that Prof. Kapp had also \amp ; 134 Mr. B. Hopkinson .
[ Nov. 20 , The test-piece behaves as a spring , the lower end of which is held fixed by the inertia of the masses to which it is attached , while the upper end carries the mass E. The adjustments are such that the natural period of vertical oscillations of this system is nearly equal to the period of the varying magnetic pull .
The latter then sets up large forced oscillations of its own period .
If the two periods are so far different that the dissipative forces have not much effect , the amplitude of the range of pull experienced by the piece is to the range of pull exerted by the magnet approximately in the ratio 1/ ( 1 \#151 ; n2 ) , where n is the ratio of the two periods .
The range of stress in the piece can be adjusted coarsely by variation of the frequency of the magnet current , and more finely by changing the E.M.F. applied to the coils .
By means of wire guys Y the weight is constrained to move parallel to the axis of the piece .
Owing to the smallness of this movement the wires exert no appreciable pull in a vertical direction .
The weight is carefully balanced so that its centre of gravity is in the axis of the test-piece .
The extension of the piece is measured by means of a pair of extenso-meters L , which are shown in greater detail in fig. 3 .
The concave mirror of each of these instruments casts on to a transparent scale an image of a horizontal lamp filament .
When the instrument is in action the line image is drawn out into a band , whose ends are sharply defined , and whose length measures the range of extension in the piece .
The extensometer readings are calibrated by means of the micrometer screws 0 ( fig. 3 ) , and were checked when the machine was in action by observing , with microscopes , the movement of needle points attached to the blocks P , P. The extensometer always agreed with the microscopes within 3/ 10,000 inch , or 3 per cent , of the range , and the measurements of extension are certainly correct within that amount .
The form of test-piece used is shown in fig. 3 with the extensometers attached .
The section is 0*049 square inch .
Current is supplied to the machine from a 4-pole rotary converter , which takes continuous current at 100 volts from a storage battery .
Constant frequency in the supply of alternating current is , of course , of vital importance , and this is perfectly secured if the converter is driven from a battery which is not being charged or doing other work .
The converter has a weak field , and is therefore largely self-compensating as regards speed , a drop in battery voltage producing a constructed a fatigue-tester , in which direct stress was produced by the pull of an electromagnet excited by alternating current .
The principle of resonance is not , however , used in this machine .
A description of Prof. Kapp5s machine appeared in the ' Zeitschrift des Yereines Deutscher Ingenieure , ' Aug. 26 , 1911 .
1911 .
] A High-Speed Fatigue-Tester , etc. Mr. B. Hopkinson .
[ Nov. 20 , corresponding drop in the field current .
It is found that , once set going , the machine will continue to run for several hours , giving a constant stress within 2 per cent. , without being touched in any way .
The converter takes about 10 amperes at 100 volts when the machine is in action .
Test-coils of 100 turns each of fine wire placed round the magnet poles , just above the air-gap , are connected to an electrostatic voltmeter , and serve to measure the flux density , whence the magnetic pull can be calculated .
Typical curves showing the relation between frequency of supply current and the corresponding stress for a given magnetic pull are given in fig. 4 .
The pressure of supply was raised in each case in proportion to the frequency , Relation , behcreen frequency and sh Tflagnehc fudl constant for each curve:-Tbr Dj Sc 113 , 0 4 Ions fuzr scy .
ui td n " " " / J^requertcy x\gt ; f fwJJL , cycles fier second .
so that the flux density , and therefore the range of pull applied by the magnet , remained constant .
The frequency was accurately measured by counting the beats between the notes produced by the machine and by a standard tuning fork .
In one case ( Di ) the maximum pull produced in the piece at the point of resonance was about 75 times , and in another ( UB ) ( taken under similar conditions ) 35 times the magnetic pull .
The ratio between these quantities at the resonance point depends mainly on the rate of dissipation of energy by elastic hysteresis in the test-piece and its attachments , and by vibrations communicated to the ground .
A bar of steel subjected to alternations of stress at the rate of 100 cycles or more per second in this machine gets perceptibly warm even when the stress is apparently within 1911 .
] A High-Speed Fatigue-Tester , etc. 137 the elastic limit .
It was observed that much more heat was developed in the bar UB than in Di ( which was of a different brand of steel ) , and it is probable that the difference between the two curves is mainly due to this cause .
When these trials were made the machine was bolted to a very heavy block of concrete , and it is improbable that much energy was dissipated by vibration .
When the third curve ( D2 ) was taken , the machine had been shifted on to a lighter foundation .
The vibration in the surrounding floor was then perceptibly greater , and this is probably the cause of the smaller magnification of pull observed in this case ( about 17 times ) .
It will be observed that the resonance point is much less marked , the curve having a flatter maximum .
This is in practice an advantage , though more power is taken to drive the machine , because the stress is less affected by slight alterations of speed .
The pull of the magnet varies between zero and a maximum value , and is always such as to give tension in the piece .
The mean value of this pull is half the maximum value or range of pull , and ( in consequence of the magnification by resonance ) is small compared with the total range of stress .
Such as it is , it is more or less counteracted by the weight of the mass E , so that in the normal working of the machine the mean stress is almost zero , and the limits of stress are nearly equal in tension and compression .
The machine has been used in this way throughout the experiments described in this paper .
It is easy to apply any desired steady tension by means of a screw and spring balance ( see fig. 2 ) , and thus to alter the ratio of the tension and compression limits .
Measurement of Stress .
As already indicated , the quantity directly measured in this machine is the change of length of the piece in inches .
By applying a static pull to the same piece in a testing machine the relation between pull in pounds and extension in inches can be directly determined , and the stress can then be calculated from a measurement of the section of the reduced portion of the bar on the assumption that Young 's modulus is the same for the rapid ' alternating stress as for the steady pull .
The experiments of Sears on the longitudinal impact of rods* justify this assumption completely if the stresses are within the elastic limit .
In the present instance the stresses may be rather outside that limit , and the fact that fatigue , resulting ultimately in fracture , occurs , shows that at each application there must be a small extra-elastic extension or compression .
Bairstow , f measuring the * ' Camb .
Phil. Soc. Proc. , ' vol. 14 , p. 257 .
t 'Phil .
Trans. , ' A , vol. 210 , p. 35 .
Mr. B. Hopkinson .
[ Nov. 20 strain in a piece subjected to cycles of stress slowly performed ( about two per minute ) , and rather outside the limiting range , so that fracture would ultimately result , found that the range of strain , which was at first that corresponding elastically to the stress , slowly increased , owing to the formation of a hysteresis loop of the type shown in fig. 5 .
It consists of the two parallel elastic portions AB and CD which correspond to the normal Young 's modulus of the material and the curved portions BC and DA .
The error committed in calculating the stress from the strain on the assumption of perfect elasticity is represented very approximately by LM .
After a good many thousand reversals this loop settled down to a constant form which apparently persisted till fracture .
The extra-elastic strains at YAjifiarent 'Range , caLculalecL from strain -h m\#151 ; True Range of Stress-----------------\#151 ; *J .
these slow speeds are considerable ; for instance , in a steel whose7 limiting range was probably about \#177 ; 13 tons , the application of a range of \#177 ; 14 tons gave ultimately a hysteresis loop whose width ( LM ) was about lJ5t'ns\gt ; or 10 per cent , of the range .
It is , however , quite certain that the non-elastic portion of the strain in the high-speed machine is nothing like so great , because , if it were , the magnification of the pull at the resonance point would be much less than is actually found .
The magnetic pull , by its action on the moving weight , provides all the energy dissipated mechanically in the system , whether by defective elasticity of the piece , vibrations of the floor , or other causes .
The work done per cycle is easily calculated from the range of pull P and the movement of the weight d ; it is , in fact , \irVd if the frequency corresponds 1911 .
] A High-Speed Fatigue-Tester , etc. 139 ' to maximum resonance .
This sets an upper limit to the area of the hysteresis loop , and thus to the amount of extra-elastic strain .
It is not difficult to show that , even if the whole work done by the magnet is accounted for by hysteresis in the piece , the range of stress differs from the value corresponding to perfect elasticity by an amount not exceeding about 2 P* For instance , in the curves Di and UB ( fig. 4 ) , the error in measuring the stress must have been less than 0*8 ton per square inch , and was probably a good deal less .
In D2 , which was exactly the same steel as Di , the magnetic pull is much greater , but this , as has already been explained , is due to the greater vibration of the floor .
These considerations seemed to justify the view that , while the stress calculated from the change of length of the piece in this machine must be rather too high , the error could not be important , so long as the magnetic pull was but a small fraction of the range of stress .
It was , however , thought desirable to determine the acceleration of the moving mass , and thus to obtain a direct measurement of the pull exerted by it .
For this purpose a needle was fixed horizontally at a convenient point on the mass E , and a microscope , magnifying about 60 diameters , was sighted on the reflection , in the surface of the needle , of an electric lamp .
By properly adjusting the position of the lamp , a very sharp line could be obtained , whose position , when the machine was at rest , could be defined within 1/ 10,000th of an inch .
With the machine in action , this line was drawn out vertically into a band , whose length could be measured against a transparent scale in the field of the microscope correct to about l/ 5000th of an inch .
The microscope was supported for this purpose , with its axis horizontal , on a table loaded with iron plates , and resting on thick blocks of rubber placed on the heavy block of concrete to which the machine was bolted .
In this way movement of the microscope itself was completely eliminated , and the apparent movement of the needle in its field corresponded to the actual motion of the weight .
The wire guys V , being stretched fairly taut , prevent any angular motion of the weight , so that the movement of its centre of gravity is equal to that of the needle , wherever the latter be attached .
The sufficiency of this constraint was proved by attaching a mirror to the weight and observing through a telescope the reflection of a distant lamp filament .
Not the slightest tremor could be seen when the supply of current to the machine was suddenly * If d !
be the change of length of the piece in a cycle , and P ' the amount ( LM in fig. 5 ) by which the apparent stress calculated from the strain exceeds the true stress , the area of the hysteresis loop is greater than \ Hence ^7rPd ( work done by magnetic pull P per cycle ) is greater than JP 'd ' .
Now in this particular machine d is about %d\ hence P ' must be less than 2P .
Mr. B. Hopkinson .
[ Nov. 20 , thrown on , though an angular movement , giving a displacement of the needle ( relative to the centre of gravity ) of l/ 5000th of an inch , would \#166 ; certainly have been detected .
Having thus obtained the range of movement a of the weight , its acceleration / at each end of its travel can be calculated from the formula / = p1a , where 27r/ p is the periodic time of the cycle .
It is here assumed that the motion of the weight is simple harmonic .
The mechanical conditions of the problem are such as to justify this assumption .
We are dealing in effect with a weight attached to a spring , the other end of which is fixed , and this simple system has practically only one degree of freedom , so that its natural small oscillations will be nearly ^simple harmonic .
Further , the wave form of the alternating E.M.F. supplied is nearly simple harmonic , and the corresponding magnetic pull , which depends only on the flux , will be of a similar character , and will force approximately simple harmonic vibrations in the mass of a period equal to its own period .
Finally , under normal circumstances , the frequencies are adjusted so that the natural period of the mass is approximately equal to that of the force , and the selective action of resonance has the effect of accentuating that type of motion of the weight whose period corresponds with the fundamental period of the applied force .
A record of the extension of the piece in terms of the time was taken by causing one of the extensometer mirrors to throw a spot of light on to a revolving drum .
The curve so traced showed no appreciable departure from a sine wave .
It must , however , be remembered that in consequence of the increased importance in the acceleration of high frequency terms in the displacement , a departure from simple harmonic motion which could not be detected on the record might produce a considerable change in the acceleration .
The justification of the assumption that for this purpose the motion may be treated as simple harmonic therefore rests mainly on the theoretical considerations mentioned above .
The photographic records , however , were useful as showing the very regular character of the cycle of stress produced in the machine .
The forces acting on the weight are the pull in the bar and the pull of the magnet .
At the point of resonance these two forces will differ in phase by a quarter of a period , and consequently at the two ends of the travel of the weight , when its acceleration is a maximum , the pull exerted by the magnet will be zero .
At these points therefore the mass multiplied by the acceleration is equal to the pull in the rod , and the range of pull in pounds may therefore be taken as equal to 2 Map2/ g , where 2a is the range of movement of the weight observed in the microscope .
If the frequency of the magnet pull differs from the natural frequency by such an amount as considerably to 1911 .
] A High-Speed Fatigue-Tester , etc. 141 reduce the movement of the weight , the magnetic pull comes nearly into phase with the pull in the bar , the relation .
of signs being such that if the force frequency is the greater the magnetic pull assists the pull in the rod to accelerate the weight , while if the natural frequency be the greater these pulls are in opposite directions .
Thus in the first case the range of pull calculated from the movement of the mass should exceed , and in the second should fall short of , that shown by the extensometer by an amount equal to the range of pull exerted by the magnet .
The following table gives one complete set of observations made with the object of comparing the different methods of getting the stress .
The material used in this trial and in the endurance tests described below was a mild steel containing 018 per cent , of carbon , with a breaking stress of about 30 tons per square inch .
Frequency of stress cycle , per sec 100 .
j 100 .
116 .
120 .
134 .
Range of extension in piece :\#151 ; Extensometer A ( thousandths of an inch ) 1 *6 2 5 8*6 9*1 0*9 " B ( " " ) 1-5 2-5 8*2 8*9 1 *1 Microscope A ( , , " ) 1 *6 2*5 8*6 9*1 1 -o Mean 1 *6 2-5 8-4 9-0 1 *0 Stress calculated from extension ( tons per sq .
in .
) ... 5 5 8 *5 28-7 30 -7 3-4 Magnetic Rail :\#151 ; Volts on test-coil ( 100 turns ) 47 *5 58 *5 46 49 48 Maximum flux density ( kilo-lines per sq .
cm .
) 4-4 5*4 3*6 3*8 3*3 ( 1 ) Range of magnetic pull ( tons per sq .
in .
) 1 *5 2-3 1 -1 1-1 0-9 Acceleration Pull:\#151 ; Movement of mass observed in microscope 2 -4 3*6 12 -1 12 -7 1 *6 ( thousandths of an inch ) ( 2 ) Corresponding range of stress ( tons per sq .
in .
) 4'0 6-0 27 -3 30 -6 4-8 Sum or difference of ( 1 ) and ( 2 ) 5 .5 8-3 ( 28 -4 ) \#151 ; 3-9 i Range of stress calculated from frequency ( assuming that resonance occurs at frequency 119 ) 5*1 7-9 ( 29 0 ) 3 3 The point of resonance in this case was probably very near 119 periods per second .
At 100 periods and at 134 periods the magnetic pull would be very nearly in phase with the pull in the piece .
It will be seen that at the first of these frequencies the sum and at the second the difference of the magnetic pull and the acceleration pull agrees closely with the tension in the piece .
The frequency of 116 is near to but is certainly below the resonance frequency .
There will be a considerable difference of phase in this case , but not so much as 90 ' .
Consequently the stress in the piece should exceed that required to accelerate the weight by something less than the magnetic pull of IT tons per square inch .
In fact , it appears to exceed it by 1*4 tons per square inch .
At 120 periods we are very close to the resonance point , but probably rather past it .
Here the difference of phase 142 Mr. B. Hopkinson .
[ Nov. 20 , will be nearly 90 ' , and it will be seen that the stress required for acceleration is very near to that calculated from the extension .
The last line of the table gives the stress calculated from the magnetic pull P by the formula P/ ( l\#151 ; n2 ) , where n is the ratio of the forced to the free period .
This is , of course , not applicable to frequencies near the resonance point , when the amplitude is controlled largely by the dissipative forces , but it gives results in fair agreement with the observations at lower or higher frequencies .
A great number of comparisons of the stress estimated from the acceleration and that estimated from the extension have been made on different pieces of the same steel , at or near the point of resonance , and with a range .of stress of 30 to 32 tons per square inch .
These observations , taken altogether , reveal a systematic difference between the two ; the stress calculated from the extension being higher by perhaps f ton per square inch on the average .
It seems probable that this is mainly due to defective elasticity , which becomes apparent in this particular steel when the true stress exceeds 30 tons per square inch .
The endurance tests which follow were made on the same steel .
It may be observed here that the phenomenon of increasing hysteresis .under the influence of alternating stress which was observed by Bairstow at low speeds is completely reproduced at the high speed of this machine.* If the true stress , as shown by the movement of the weight , be maintained .constant at a fairly high value , the strain shown by the extensometer slowly increases , reaching a steady value after the lapse of some thousands of reversals .
At the same time the magnetic pull has to be quite noticeably increased in order to maintain the stress .
The piece also gets considerably hotter .
Comparison of Endurance at Different Speeds .
Some systematic endurance tests were then carried out on the same steel with the object of finding whether the high speed had any effect on resistance to fatigue .
Dr. Glazebrook kindly arranged that a series of trials of the material should be made at the National Physical Laboratory in the direct stress machine designed by Dr. Stanton , f whom I must thank for the great interest he has shown in the tests and the trouble which he has taken * Bairstow , ' Phil. Trans. , ' A , vol. 210 , p. 42 .
The increase of hysteresis under alternating stress has also been observed by Turner , who noticed the rise of temperature of a tube subjected to a Wohler test , 'Engineering , ' Sept. 4 , 1911 .
t Stanton and Bairstow , ' Inst. Civ .
Eng. Proc. , ' vol. 166 , p. 78 .
The first machine on these lines was made by Messrs. Reynolds and Smith ( see 'Phil .
Trans. , ' A , vol. 199 , p. 265 ) .
1911 .
] A High-Speed Fatigue-Tester , etc. 143 in carrying out this part of the work .
The steel contains 0T8 per cent , of carbon and 07 per cent , manganese and was supplied in the form of bright drawn bars of 1 inch or inch diameter .
The cold drawing to which these bars are subjected has the effect of raising the elastic limit of the bar as a whole , but that this change is not serious is apparent from the tensile tests given below , and anyhow it is of little consequence in a comparative trial .
Three bars of this steel in all were used in the test , two being of 1 inch in diameter , and the other ( D ) 1 ] inch .
These bars are referred to as .C , D , and E respectively .
Tensile tests were made on two specimens cut from bar D with the following average results:\#151 ; Elastic limit in tension ... ... . .
19'3 tons per sq .
in .
" compression ... ... .
13'4 " Maximum stress ... ... ... ... ... ... .
29 " Elongation on 8 in ... ... ... ... ... 16 per cent. Reduction in area ... ... ... ... ... .
60 " For the tension test the bar was turned down to a diameter of f inch , and for the compression to 1 inch .
The elastic limits correspond to the load at which permanent set first becomes distinctly apparent in a Ewing extensometer ; that for compression is not well defined .
A series of seven pieces cut from bar C and tested in the direct stress machine at the National Physical Laboratory gave the following results , which have been plotted in fig. 1 N.P.L. No. of specimen .
Mark .
Reversals per minute .
Range of stress , in tons per sq .
in .
Reversals to fracture .
Total duration of test ( hours ) .
216 C VI 1351 38 *8 , 6,613 0*08 217 C IV 1351 34 *9 7,633 0*09 220 C V 1246 29 *65 44,244 0*59 219 C I 1104 26 -45 197,761 3*0 218 C III 1063 17 *7 1,017,337* 15 *9 218 C III 1283 25 *7 150,981 1 '96 221 C II 1092 22-8 1,113,484* 17 *2 221 C II 1125 25 *5 1,003,610* 14 *9 225 C VII 1076 24 *6 326,550 5*0 * Unbroken .
Pieces of another bar ( D ) of the same material were tested in the direct-stress reciprocating machine and also in a high-speed Wohler machine with the following results :\#151 ; Mr. B. Hopkinson .
[ Nov. 20 , .
N.P.L. No. of specimen .
Mark .
Reversals per minute .
Range of stress , in tons per sq .
in .
Reversals to fracture .
Total duration of test ( hours ) .
Direct Stress Machine .
222 D I 1084 27 -7 120,000 1 *8 223 D II 1084 27 -7 119,200 1 *8 224 D J 1084 - 24 -6 326,560 5*0 Wohler Machine .
77 D J 2200 27 -0 1,637,500 12 *5 105 D J 2200 25 -5 7,000,000* 53 *0 # Unbroken .
The endurance tests on the new high-speed machine were carried out on pieces of the same bars .
It may be noted that the diameter of the test-piece\#151 ; \ inch\#151 ; is the same as that of the piece used in the direct stress machine at the National Physical Laboratory , so that exactly the same portion of the bar was tested .
The trials were net all continuous , but were split up into periods separated by intervals of rest ranging from a few minutes to several hours .
But most of the continuous runs covered at least a million cycles , and if a piece was not broken in one day , the trial was usually continued on the next , and so on , until fracture occurred .
In all the trials the piece got perceptibly warm , and if it reached a temperature ( judged by the hand ) of 60 ' or 70 ' C. at the centre , a jacket of blotting paper was applied and kept soaked with water .
It does not appear that the rise of temperature materially affected the endurance , for the pieces seemed to break indifferently at the centre or near the ends , where of course the temperature was by conduction kept nearly atmospheric .
Continuous observation was kept of the extensometer reading during a trial .
After the first few minutes this settled down to a constant value , , and very little adjustment was then required .
The " apparent stress " given in the second column of the following tables is that calculated from the extension on the assumption of perfect elasticity .
In the first series of trials , which were made on the " C " bar the strain was kept at rather a high value , and there must have been considerable departure from elasticity .
The true stress is , therefore , uncertain except in the one case C7 where the movement of the mass was measured .
Continuous observation was , however , kept of the E.M.F. on the test-coil and as pointed out in the last section the " apparent stress " cannot exceed the true stress by more than twice the range of magnetic pull calculated from these observations .
This is the basis of the assertion ( in cases other than C7 ) that the stress in the following series of trials never fell short of 3ft tons per square inch .
1911 .
] A High-Speed Fatigue-Tester , etc. No. of bar .
Range of stress ( tons per sq .
in .
) .
Time ( hours ) .
Millions of Remarks .
By extensometer ( apparent ) .
True stress .
reversals to fracture .
Co 32\#151 ; 34 Exceeded 30 15 6*0 Four periods .
6 C5 34\#151 ; 35 Exceeded 30 15* 6*2 Three periods , one of 12 hrs .
continuous .
c6 33\#151 ; 34 Exceeded 30 29 11 '5 Four periods , one of 14 hrs .
C- 31\#151 ; 33 30\#151 ; 32 ( by movement of mass ) 7 2*8 Three periods .
C8 32\#151 ; 33 Exceeded 30 8* 3*4 Two periods .
In most of these tests the piece heated to such an extent that it was necessary to jacket it with wet blotting paper .
Further trials were made in two pieces cut from another bar of the same material ( D ) .
Piece Di was the same as that used in the comparisons of different methods of getting the stress which have been described above , in the course of which the stress varied , rising sometimes to 31 tons per square inch .
On the conclusion of these experiments .it was put through a steady endurance test at an " apparent " range of stress of 30*5 tons per square inch , calculated from extensometer .
It broke after 10 million cycles taking about five periods of five hours each on consecutive days ( total , 25 hours ) .
As the piece kept cool ( it was not necessary to jacket it ) it may be inferred that the elasticity was nearly perfect to the end of the test , and that the estimate of stress derived from the extensometer was not far wrong .
Allowing liberally for all possible errors , it may be asserted that the stress exceeded 29 tons per square inch throughout this test .
The treatment of piece D2 was as follows:\#151 ; Range of stress ( tons per sq .
in .
) .
Total time ( hours ) .
Total reversals ( millions ) .
Remarks .
By extensometer .
By acceleration .
28 *2 26 *8 32 12*8 | Four periods , about 8 hrs .
each .
Four weeks ' rest .
29 *0 I 28*8 | I 15 1 6*0 | Three periods , about 5 hrs .
each .
Two days ' rest .
30 -0 28*2 14 5*6 Three periods , one of 12 hrs .
continuous .
31 *0 30 *0 15 -8 6*3 The piece was perceptibly hotter .
32 *0 29 *0 5*0 2*0 Two periods .
Broke .
vol. lxxxvi.\#151 ; A. L Mr. B. Hopkinson .
[ Nov. 20 , Until the last period of five hours the piece was only slightly warm , and the elasticity must have been nearly perfect .
During the last period it got very hot , and had to be jacketed .
It is probable that the stress did not fall short of 29 tons during the last 50 hours and 20 million reversals\#151 ; certain that it did not fall short of 28 tons .
Finally , a piece Ei , cut from another bar of the same material , was subjected to the following treatment:\#151 ; Range of stress ( tons per sq .
in .
) .
Total time Reversals By extensometer .
By acceleration .
( hours ) .
\#187 ; ( millions ) .
31 32 Not less than 30 Not less than 31 6 7 2-5 | 2*7 1 *0 ( broke ) Each of the runs was continuous , and they took place on successive days .
In each continuous observation was kept of the movement of the weight , and care was taken that the true stress ( calculated from the movement ) should never fall short of 30 tons and 31 tons per square inch respectively .
Deference to the curve ( fig. 1 ) on which the observations with the slow-speed machine ( 1100 revolutions per minute ) are plotted , shows that the probable life of this material under a range of 28 tons per square inch at that speed is about 100,000 reversals and one and a-half hours of time ; and with a range of 30 tons about half as much .
In the Wohler machine ( at 2000 revolutions per minute ) the endurance is apparently a little greater , but neither the number of observations nor the amount of the difference is sufficient ground on which to base an inference as to the effect of speed .
But at the higher speed of 7000 alternations per minute the evidence seems to be conclusive that the endurance is greater .
Not only is the number of reversals of a given range of stress required to break the piece greatly increased , but the actual time during which it must be subjected to alternating stresses is also increased .
It is not , however , proved that the limiting range of stress , namely , the maximum range which will not cause breakdown however often it be repeated , is greater at the higher speed .
Though the observations suggest that conclusion , it is quite possible that sufficiently prolonged vibration at a range of stress of 27 tons per square inch would ultimately have ruptured the material .
Since , however , it takes 10 million reversals and some 25 hours of time to break the piece with a range which certainly exceeds 29 tons , it would appear probable that the number of reversals and the time required with a range of 27 tons would be very great indeed , perhaps so great that for practical 1911 .
] A High-Speed Fatigue-Tester , etc. 147 purposes the piece might be regarded as unbreakable by this load if it be applied and removed sufficiently rapidly .
Probable Causes of Speed-effect .
In the hysteresis loop ( fig. 5 ) into which the stress-strain curve of a metal undergoing fatigue apparently always develops , the straight portions AB and CD correspond to perfect elasticity , but during the curved parts BC and DA the material is flowing or taking permanent set .
There seems to be little doubt that fatigue is the cumulative effect of the internal slips which accompany this flow , and that the rate of fatigue is determined mainly by the amount of the extra-elastic strain occurring in a cycle ( BD in fig. 5 ) which Bairstow has called the " cyclical permanent set .
" Accepting that view , there are two ways in which speed may be expected to affect endurance .
First , if the speed be high enough there may not be time for the full amount of flow corresponding to BO and DA to take place at each reversal .
The amount of flow , and therefore of damage to the material , occurring at each cycle will then be reduced .
So far as this factor is concerned , therefore , increased speed will make for greater endurance , provided that the speed is sufficient materially to affect the stress-strain relation .
Second , there is the phenomenon of recovery .
Any over-strained material tends to recover its elasticity with time .
This tendency must be supposed to be continually at work during the progress of a fatigue test , repairing the ravages of the successive over-strains .
Obviously , so far as it goes , its operation will be relatively more effective at slow speeds .
Hence the effect of this second factor is to associate greater endurance with slower speed , and is opposite to ' that of the first .
The combined effect of the two would be to cause the endurance ( always reckoned as the number of stress-cycles to fracture ) to diminish at first as the speed is increased , then to reach a minimum value and increase again .
But if the influence of recovery on the progress of a continuous fatigue test is small , then the minimum of the curve connecting speed and endurance becomes very flat and we shall get approximately constant endurance until the speed reaches a certain limit , after which the endurance increases .
The conclusion expressed in the last sentence is in accord with the facts , and the inference that recovery is not an important factor is at least probable .
It may be , however , that the opposed actions of recovery and of time-change in the stress-strain relation happen to cancel each other pretty accurately over the fiat part of the curve .
Finally , to settle the point it would be necessary to carry out comparative tests in which the speed is the same but the rate of recovery different .
The most obvious way would be to A High-Speed Fatigue-Tester , etc. try the effect upon endurance of raising the temperature of the piece , which is known to accelerate recovery .
Such an experiment would give very useful information .
As pointed out above , there is some evidence in the experiments with tire high-speed machine that the rate of fatigue , at that speed at any rate , is not much affected by temperature ; but the observations were not directed especially to this point , and the evidence is not conclusive.* The direct determination of the other factor\#151 ; the effect of speed on extra-elastic stress-strain relations\#151 ; also requires much more experimental study than it has yet received .
It is certain , however , that if the cycle of stress be completed in 1/ 100 second or less , the effect must be large .
The greater endurance at high speeds found in the present research is perhaps sufficient evidence for this assertion , but there is independent confirmation of it in the results of experiments which have been made on the extension of a wire under the action of a blow.f The wire ( stretched vertically ) was stressed by the blow of a weight falling against a stop at its lower end .
Measurement of the momentary extension , and calculation from the mass of the weight and the height of fall , alike showed that the average stress over the top 20 inches of the wire must have reached a maximum value exceeding by 50 per cent , the static elastic limit , and sufficient if long continued to break the wire .
Yet it was found that 11 such blows in succession extended the part of the wire under measurement by less than 1/ 2000 part of its length , and the wire was almost perfectly elastic up to the maximum stress applied by the blow .
The stress was outside the elastic limit for about 1/ 1000 second at each blow , which is of the same order as the time during which the over-stress lasts in each cycle in the high-speed machine .
The arguments here used would lead to the conclusion that though the rate of fatigue under the action of stress which ultimately ruptures the piece will be lower at high speeds of reversal , yet the limiting range of stress which the piece will sustain indefinitely without breaking will be independent of speed .
For if a single reversal produces any permanent effect whatever it is to be expected that a sufficient number must ultimately fracture the piece , and it is difficult to believe that time can be a factor determining whether or no such permanent effect shall exist , though it may affect its amount .
It is quite possible , however , that as the time occupied by each cycle is diminished the effect produced may diminish in greater \#166 ; * Some experiments on the effect of temperature on fatigue have been made by XJnwin , who found that the endurances of mild steel was slightly increased at a temperature of about 200 ' C. , 'Inst .
Civ .
Eng. Proc./ vol. 166 , p. 117 .
t Hopkinson , 6 Roy .
Soc. Proc. , ' vol. 74 , p. 498 .
Mechanism of the Semi-permeable etc. 149 proportion .
And the ultimate result may be that for practical purposes the limiting range is raised on account of the great length of time required to cause fracture .
Further , it is noted that when the rate of fatigue is very slow the property of recovery becomes relatively more important , and it is at least conceivable that it may be more than competent to overcome the fatigue altogether and so really raise the limiting range .
The observations described in this paper were for the most part taken and reduced by my assistant , Mr. H. Quinney , whom I must thank for the zeal and ability with which he has carried out the work .
I wish also to acknowledge valuable assistance given at various stages by Mr. C. Trevor-Williams , advanced student in the University of Cambridge , and by my brother , Mr. R. C. Hopkinson , Trinity College .
The Mechanism of the Semi-permeable Membrane , and a Netv Method of Determining Osmotic .
By Prof. F. T. Troutgn , F.R.S. ( Received November 14 , 1911 , \#151 ; Read January 11 , 1912 .
) Introduction .
For the theoretic exposition of osmotic pressure , no further attention need be paid to the membrane than to postulate its semi-permeability .
It is , however , desirable that the physical properties of the membrane should not be lost sight of , especially from the experimental point of view .
Leaving out of ' account for the moment the mechanical requirements of a rigid membrane , let us imagine a system in which we have a solution separated from the pure solvent by a thick layer of another liquid which can take up or dissolve the solvent , but not the solute .
Such a system might be sufficiently realised by means of an aqueous solution of cane sugar , ether , and water , since sugar is insoluble in ether , while ether dissolves a small quantity of water .
Both water and sugar solution take up some ether , but that is of no immediate interest in the present considerations .
This complication will be , however , dealt with later on .
If it is desired to imagine the necessities of gravity complied with , the sugar solution and the water can be supposed to be placed side by side in a vessel and kept from mixing by an impervious vertical partition reaching to

