Just a clarification on breakeyp's excellent post (He's gone to an extreme "shorthand" to keep from filling pages!):
Steel parts will theoretically survive an infinite number of stress repititions or cycles IF the applied stress is below a certain level. Above that point- failure will
happen. How soon can be predicted with increasing certainty as the variables are controlled (or are known), reducing the "factor of stupid". (A phrase I picked upfrom a favourite pre-PC professor)
Some parts can be designed to last millions of cycles, others don't need to last ten! It's maddenly hard to control ALL variables, though!