4. IMPLEMENT MUCH BETTER FINITE MODEL EXTENT AND MESH SIZE

Once we have a reasonably broad and conservative set of design conditions, we have to translate the resulting loads into actual structure. The main tool we have for this purpose is finite element (FE) analysis. One problem with this is that FE model extent and mesh size that is required by Class is hopelessly antiquated.

When Finite Element first became feasible 15 or so years ago, computational resources placed a severe restriction on the effort. Because of the computer constraints, Class elected to go with a model that was limited to two or three midships tanks -- usually one side only -- and a mesh size that was one web frame longitudinally (about 5 m) and three or four stiffeners girth-wise. This may have been a reasonable decision 15 years ago but it resulted in an extremely limited tool:

  1. A tool that could give only a very gross picture of the stresses in the mid body completely ignoring many critical "details" such as web and stringer corners.
  2. A tool that required a great deal of judgment in applying boundary conditions, especially at the fore and aft ends.
  3. A tool that could not model all sorts of interesting loading conditions, including ballast exchange.
  4. A tool that said nothing at all about the forebody and the aftbody nor the connections between the middle of the ship and the ends.
(1) was addressed by a second layer of FE models which used the results of the first to obtain boundary conditions for much more detailed models which analyzed a single web ring or stringer. Given the limits of the computer in the early 80's, this was reasonable but the process of converting the Phase I model deflections into Phase II model boundary conditions was labor intensive, error-prone, and required a great deal of judgment. One problem with all these judgment calls is that they are subject to commercial pressures (see below). More basically, the process inevitably introduced errors and uncertainty into the analysis.

But a far more fundamental limitation of this approach is that many important problems simply cannot be analyzed at all. Ballast exchange which involves asymmetric loading of tanks extending the full length of the cargo area is an obvious example. At some point in the ballast exchange operation, we must empty the 1S, 3P, and 5S ballast tanks? This ballast exchange condition typically involves 5 degree heel and an overall level of torsional stress that the yards estimate is around 15% yield. Nobody to our knowledge knows how these torsional stresses are going to be distributed, nor how they combine with all the other stresses because we don't have a proper FEM. The standard model required by class hasn't a clue.

Still more importantly, the forebody and aftbody were ignored completely. The forebody and aftbody are at least as critical as the midbody. These are areas where we see far more problems than the midbody. The forebody is subject to the toughest external loads of any portion of the ship. Deflection in the aft body is critical to the all-important shaft and main engine reliability issue. The standard class FE model simply can't address these issues.

It is no longer necessary to accept the constraints of this approach. Nor has it been for some time. Computers are at least 100,000 times more powerful than they were when FE was introduced into the Rules. The only proper model extent is the full ship. The proper mesh size girthwise is every stiffener (about 1 meter). The proper mesh size longitudinally is every frame, except where the frame spacing is more than three or four meters, it should be every half frame, but in way of the stringers it should be every quarter frame. ( One can reasonably argue it should be every quarter frame everywhere in the cargo tank length to keep the element aspect ratio nearly square.) For our ULCC newbuilding program, Hellespont belatedly developed a model that almost met this spec. It has about 300,000 nodes. It takes about two hours to solve a load case on a PC costing less than $2,500. Such a model is not only now computationally feasible; it's dirt cheap.

Class should require that all new tanker designs be modeled to the above level of detail. Paradoxically, by eliminating all the Phase II work, this will probably reduce rather than increase design cost. The yards resists better finite element modeling because and only because they know it will result in more steel.

PREVIOUS NEXT