Originally Posted by
easyupbug
It is true that aluminum has no lower limit for stresses and a finite fatigue life; however, it is not a factor in consideration of aluminum frames built for purpose and used appropriately.
That's an excellent and concise summary.
More details, for the skeptical. One of the points of confusion that is often ignored is that fatigue analysis is based upon stress, which is the force per area. This can be expressed in pounds per square inch which is a pressure unit, so we could also use Pascals, megaPascals, or other units. Because aluminum has lower ultimate and yield stresses (the stresses at which the material breaks and bends, respectively), aluminum tubes are thicker than steel. So in a properly designed Al frame, the stress levels are much lower because thicker tubes have more cross-sectional area to divide the same forces over. A key point is that the ultimate and yield stress considerations by themselves require aluminum tubes to have a larger cross section. At this point in the discussion, assume that sizing doesn't consider fatigue.
A fatigue curve is a graph of stress (vertical axis) vs number of cycles before a fatigue failure occurs. See below. The steel curve "levels out". That is, below a certain stress level, steel can flex a huge number of times without breaking. There's some controversy as to whether steel will "last forever" below a certain stress level, but it is certain that there appears to be a fatigue stress limit, below which our bike frame will outlast us.
The confusion probably originates when we look at the curve for Al. It shows no "leveling out". That is, it appears that even modest stress levels, repeated enough times, will cause Al to fail in fatigue mode. If we naively assume that the stress level is the same for steel and aluminum bikes, we'd freak out and assume that our Al bike is going to fall apart after a few years. The reason that this does not happen is because of the cross-sectional area, mentioned above. We could take the figure below and create a new graph with force on the vertical axis and number of cycles to failure on the horizontal axis. Then we could assume assume two bicycle frame tubes, one steel and one aluminum, both sized (having adequate cross-sectional area) to handle normal bike forces. The traces on the new graph would have the same shapes, but the aluminum trace would be shifted up above the steel trace. It would start at the left significantly above the trace for steel, and the aluminum trace would not dip below steel until a huge number of stress cycles.
In the above, I've stated that on the basis of keeping our bike from breaking or bending (ultimate and yield stress) our aluminum frame tubes are thicker-walled. In practice, the walls might be made thicker still to further shift the aluminum frame's force/cycles fatigue curve further "up" on that new graph. That is, the thickness of the tube is selected as the minimum that satisfies all three failure modes: acute breakage, bending, and fatigue failure.
Al frames can and do fail in fatigue, and if fatigue is seen (As a fatigue crack) at one place on the frame then other cracks will likely be found elsewhere in fairly short order. And welding a fatigue crack weakens aluminum alloys used in bikes. To regain strength, the weld (in fact, the whole frame) needs to be heat treated. Practically speaking, an Al frame showing fatigue is toast.
If you find a 2nd hand Al bike of good quality,that is not damaged, and you carefully inspect for damage and for fatigue cracks and find none, you may be ok. But any sign of hairline cracks means you should walk away.
BTW, some of the responders above use a really, really bad set of logical/statistical fallacies. Words to the effect "My XYX 2000 aluminum bike has 100,000 miles on it and hasn't failed - aluminum frames are great". The first error is what is called the n=1 error. With one sample (your bike) you can make no generalizations about other similar bikes of the same make, model, and size, much less can you say about other bike makes and models. Another error in that statement regards extrapolation outside the sampled population. Even if you found 100 people with the exact same bike in the same size, and found a very small amount of fatigue failure, it's not statistically feasible to infer how other makes and models perform. In fact, its not even proper to infer fatigue life for the SAME make and model bike in a different size. It would be inapt to infer (for example) expected lifetimes for cheaply-made aluminum BSOs using data from owners of high-quality, well-designed and well-made aluminum bikes
