From the design point of view, the effect is different depending upon whether the loading is constant or variable amplitude.
Under constant amplitude loading it has been shown, by crack propagation testing, that there is a certain stress range (So), which varies both with the environment and with the size of any initial flaws, below which cracks do not propagate. Under such conditions, an indefinitely large number of cycles can be sustained. For design purposes it can be assumed, both in air and in sea water with adequate protection against corrosion, that if the applied stress range is less than that corresponding to 107 cycles on the relevant design S-N curve, then the possibility of fatigue failure can be ignored. However, for unprotected joints in sea water it should be assumed that all stress ranges are potentially damaging (i.e. that S o=0).
Under variable amplitude loading, where the stress spectrum includes ranges both greater and smaller then So, it has to be assumed that the stresses greater than So will cause the flaw to grow so that a progressively larger proportion of the smaller stresses will also become active (i.e. So will progressively decrease). BS 7608 (Clause 4.4) defines a suitable design method for dealing with this situation. This is to assume that the design S-N curve is bent from slope (m) to a shallower slope (m+2) at N=107 cycles. In consequence, the values of N in the Miner summation
are increased (compared with those that would be obtained from a straight line S-N curve) for stresses below So.
More recent research has, in fact, suggested that it would be more logical to ignore the proposed bend and assume that the straight line Log S v Log N design curve extends downwards ad infinitum.
See BS 7608: 1993 Code of Practice for 'Fatigue design and assessment of steel structures'. Also see FAQ: In fatigue analyses, what account should be taken of the compressive component in wholly or partially compressive stress cycles?