Another problem with Class's use of finite element is the way the results are interpreted. Sometimes our philosophy seems to be: if the stresses come out low, reduce scantlings; if the stress comes out high, it's an artifact of the model. Before finite element analysis came along, naval architects were acutely aware of the fact that they couldn't predict stress with any degree of accuracy. Therefore, for the most part they adopted conservative practice, used upper bound estimates of stress, and were careful not to move very far away from established practice.

But even so we made some mistakes, and came up with some very marginal ships. Under severe economic pressure from the Japanese, the European yards pushed the envelope and made a number of big tankers in the mid-70's that had systemic structural problems. We've had some of these ships. Some of the failures were fatigue problems but others were more general including systemic cracking in the upper web corners, stringer buckling, etc.

When the first pictures of finite analysis of tanker structure became available in the early 80's, we were blown away. The FE models generate pictures which show stress level in colors: cool colors for low stress areas, and warm colors for the high stress areas. It was amazing; all the green and blue were in areas that had given us no problems; and all the problem areas were yellow and orange. In the usual color scheme, stresses above the legal level are red and stresses just below the legal level are orange. For the first time, we could see the stress flow and understand why we had failures in the corners of the upper webs and problems in the stringer toes. That's where almost all the yellow and orange were. It was obvious that this was a great tool. Now that we knew where all the yellow and orange areas are it would be an easy matter to make them green and blue. In fact, that was ABS's original idea: use FE only to increase scantlings, and not allow any decreases.

But that's not how finite element ended up being used. Instead of making all the orange, green and blue; the industry used finite element to make the whole structure orange. This is known as structural optimization. And the yards became very good at optimizing structure. Way too good. A structure can be meshed any number of different ways. The yards are experts at coming up with the model that minimizes calculated stress. And if they can't get the stress down to the number they want, then they go running to Class and ask that element stresses be averaged or in some cases be simply ignored as a model artifact. They never come running to Class pointing out the stresses look suspiciously low.

In 1999, when Hellespont went to the yards and told them that we wanted to reduce the maximum design stress by 10% (roughly make the ship yellow), they found that they had to increase the steel weight by about 7%. In other words, 70% of the structure was in the orange.

This is a prescription for disaster. The models simply aren't that good. Even the fine mesh recommended above leaves out all sorts of important details. And we don't know the loads or stresses that well. Anybody who thinks so should watch the yards during block fit-up. Often this requires a whole series of jacks and wedges and come-alongs. The induced deflection is far more than occurs in any design case. God knows what the residual stresses are. Two years ago an ABS VLCC suffered extensive stringer buckling during the stagger test. This was before even leaving the yard. The failure was blamed on moving an access ladder hole from one location to another without redesigning the stringer. But if the structure had been anywhere near robust enough to go to sea, a minor change like this -- locally compensated -- should have had no effect. When you go to sea you need margins, and we don't have those margins.

Steel is cheap. Steel prices fluctuate considerably. But on average the marginal cost to the yard of increasing a scantling is a good deal less than $500 per ton. If we increase the steel weight of a VLCC by 10% in an efficient manner, we will get a far more robust ship at a cost of about two million dollars, a little over 2%. That's intelligent regulation. Class should adopt a more conservative design criteria. We recommend an average reduction in design stress of about 10%.

This will get us back to the good ships of the mid-70's, which by the way were not over-built. We see evidence of this on even the best of our mid-70's built ULCC's. Almost all these ships developed some cracks by age 15. In the "good" ships, these cracks are limited to a handful of localized areas. The owners eventually learn where these areas are and expect to have to repair a crack or two in these areas every docking. The mid-70's ships were much weaker than earlier generations. A 40,000 ton tanker built in the late 50's had a bottom plate thickness of about 35 mm. A very good mid-70's 400,000 ton ULCC -- ten times larger -- had a bottom plate thickness of 28 to 30 mm.

And if we are going to depend on finite element, then we should use the numbers that the FE models generate. There should be no averaging of stress across elements nor any rounding down of scantlings. In absolutely no case, should the design stress be more than yield.