THROW-TESTING MACHINE FOR REVERSALS OF MEAN STRESS. 
283 
If the mean variation of speed is taken at 5 per cent., the variation of stress 
range due to this will be 10 per cent. This great variation would seem to reduce 
the value of these tests, but the author introduced them mainly to enable one to 
compare results got in this way with those obtained after the fluctuations of speed 
had been reduced to a minimum. In some cases, this fluctuation is of no 
consequence, as will be seen from the following. 
On comparing Tests 18 and 20 (Table I.) we see that for a drop of '91 ton per 
square inch in the range of stress, the reversals for rupture have increased from 
7'1 X 10 5 to over 3 - 9 X 10 6 . If, then, a fluctuation of range of stress of 10 per¬ 
cent. had been taking place for a test at about this particular range, it is evident 
that only a very small fraction of the actual reversals recorded could be effective in 
damaging the material. One thus sees that, as long as the rate of change of 
reversals with range of stress is small, slight fluctuations of velocity will not 
appreciably affect the results, whereas, when this rate is great, it is important to 
keep the speed as steady as possible. 
If the limiting range of stress increases as the speed diminishes, it will be more 
rapidly approached with the method of lowering the range used here, than in that 
used by Wohler, for the diminution of range is got by diminishing the speeds when 
the specimens are of constant diameter. Hence, after a certain point, it will not be 
worth while doing long tests, since these fluctuations of velocity, however small, 
would render the results doubtful. 
Finding that the reversals for rupture with any given range of stress are diminished 
with the speed, the author decided to limit his tests more particularly to cases for 
which the reversals were less than one million. In a few cases, however, specimens 
have been subjected to a greater number of reversals. 
Nearly the whole of one year was spent in an attempt to get more regular results 
by improving the method of preparing the specimens, by annealing, by subjecting 
each specimen to a number of reversals with a small range of stress, and by 
diminishing the fluctuations of speed in the manner described previously; and, 
strange as it may seem, the results were not so regular in many of the final series of 
tests as in those recorded in Table I. The only possible explanation is that the 
material used for Experiment I. was of more uniform quality than that used 
subsequently. 
On the Restoring Effect of a Period of Rest. 
Since in the long tests, namely, those extending over several days, the experiments 
could not be conveniently carried out without stopping, and therefore allowing the 
specimen to rest, it is important to find the effect of these periods of rest on the 
total reversals required for breaking. With this object in view, tests were made 
with a number of specimens of mild steel, some of which were broken without 
stopping the machine, while others were allowed to rest for various periods after 
2 o 2 
