Fatigue failures are an art that many people have received PhD's by the varied aspects of their occurance. However, as it turns out for steel, if stresses are kept below certain levels, the failures will not occur, or will not occur within a reasonable time frame, like say 35 years of service. This does not occur for aluminum, brass, some other metals or composites - lower stresses increase time to failure, but fatigue failure will always eventually occur for those materials. All I know is that we did everything we could to reduce stress and/or cycles to reduce the oportunity for fatigue failure cracks to expand. The elimination of fatigue failures is pretty much an art and a lot of investigationaly work and experience. The international marine standards took experience and created rules for allowable stresses in various parts of an engine and ship, by following those rules, you will be able to have your marine component approved for use in the sea and the vessel will be insurable.
As for the SD-75 engines, there were several variations in the models sold (EMD - Every Model Different). I do know that some of the engine components for which I had design responsibility failed due to fatigue. When we gathered all the information about the locomotives, the power the engines generated in five years on the 4500 Hp engines was what a 4300 Hp engine would be expected to make in ten years. This is to say that the railroads were using these locomotives about as efficiently as could be done, short of running them at full power 24/7 for several years. While the RR was unhappy about the premature wearing of the engines, the engines gave them the power (and low failure rates) that was contracted in the sale, except the the time occurred sooned than expected because the locomotives had been used so effectively that they needed maximum power. At 4300 Hp those failures did not occur, even after more than ten years or equivalent amounts of power generated.
Most failure rates (failures vs time in service) for a part are U-shaped, i.e. there is a relatively high rate for parts with little service time (called infant failures - say only a few days or weeks in service), then there are few failures for one to five years of service, finally there are many failures at the end of useful service life, say after five years. The maintenance instructions would call for replacement of parts before the failure rates would increase depending on service life. We found out for those SD-75s that the parts that were failing were at the end of service life based on the amount of power generated, not calendar time.
There are many factors which can contribute to early failure, if the technician building the part was having a bad day and did not assemble and fabricate everything perfectly, it could break quickly. For many things we are never able to determine exactly what constitutes a perfectly (or even acceptably) fabricated part, we just tweaked things until the failure rates dropped to a long enough time duration. This experience would translate into higher prices for our parts, but they lasted longer than the competition.
One railroad Director of Locomotive Maintenance spent a lot of time screaming at us (EMD) about the poor service our parts were giving his locomotives. We had them shipped to our plant to inspect them to see what the problems were - THEY WEREN'T OUR PARTS!! We told him to buy our parts and his failure rates would go away - he was so upset that we wouldn't tell him why those other parts were failing (we did know - they were cheaper) that he swore he would never buy ours again - big deal he wasn't buying ours anyway.