Finite Element analysis even if carried out in the manner we suggest would still not examine the connection details. Prompted by an epidemic of fatigue cracking in 1980's built tankers, class introduced "fatigue analysis" to check these areas. Given the societies' inability to enforce qualitative standards, they probably had no choice. Fatigue analysis can in theory help to improve the robustness of connection details, such as bracket toes, stiffeners to web connections, etc. But the practice leaves much to be desired.

Prior to about 1985, there was no such thing as fatigue analysis. Designers would select connections and bracket details that were known to be robust and, if they didn't, Class would require improvement. The goal was to eliminate failure due to fatigue rather than to increase fatigue life. Thanks to advances in computer power, it became possible to consider simulated voyage trading patterns, the resultant cyclic loadings and their application to the transfer functions of the proposed structure connections. A number of heroic assumptions are required in this process, but the end result is a guess at the mean time to failure, called the fatigue life.

The rule standard fatigue life is 20 years. Most tankers operate for 25 years or more. It has always puzzled us as to why Class would set an average time to failure less than the expected ship life. When Hellespont specified a 40 year fatigue life, some class software had to be recoded to accept the larger number. But an equally pressing problem is that fatigue life depends critically upon the trading pattern that was implicitly included in the specification. According to both LR and ABS, a ship trading in the North Atlantic has a very different fatigue life than one trading in Indian Ocean. In this case, the class software is almost certainly correct, at least in a qualitative sense. The most graphic example of this is the American flag tankers trading on the Alaska-West Coast route. In this very severe environment, some of these ships turned out to have fatigue lives of one or two trips. Closer to home for Hellespont, if a new VLCC trades Mongstad to Philadelphia rather than Ras Tanura-Yokohama, it's fatigue life is halved. The nominal 20 year life become 10.

This whole approach assumes that our knowledge of ocean sea spectra throughout the world is so complete that we can fine tune our structures right up to the point of planning for an average time to fail that is well within the expected life of the ship. The transfer functions themselves are gouges that are based primarily on offshore oil industry studies of generic joints and brackets and then extrapolated to actual ship details in a semi-judgemental manner. The mechanism as to how sea state spectra is converted to wave forces and then to shear and bending moments on the hull and then transferred via primary and secondary structural members to the connections is at best an idealization. Loading and discharging loads are completely ignored despite the fact that this cycling is implicated in the early cracking seen in North Sea shuttle tankers. There is no consideration of ballast exchange and its effect on the loading of the structure.

Here is just one example of how divorced from reality the Class rules have become. In computing the "fatigue life", the rules add port time to the fatigue life. In a 10 voyage year with 5 days port time per voyage, the naively anticipated fatigue life of 20 years is actually 315 days x 20 or about 17.5 years. This sort of adjustment assumes we can predict the fatigue life within 15 percent. We have proven over and over again we cannot. In reality, fatigue life is a misnomer. Fatigue life is nothing more than an index, 50 is better than 30 which is better than 20. That's about all you can say.

Our whole approach to fatigue is overly aggressive. It seems as if every time we have a choice between a conservative assumption and an unconservative, we choose the unconservative. This is probably the result of inter-class competition. In any event, the ships are still cracking early in their lives. Given all the uncertainties a far more conservative approach is the only prudent alternative. At a minimum Class should require a guessed at mean time to failure, well in excess of the ship's expected life, at least 40 years. We should run the voyage simulation over all reasonable trading patterns and design to the resulting fatigue envelope rather than some average. Given intelligent design, the increase in overall ship price will be barely noticeable.