Laatste update: 2004-09-11
Samenvatting / Summary
Titel / Title: “Doordringen tot de Probabilistische Lekberekening”
The probabilistic damage stability is one of the subjects in ship design that take much computational capacity and time. This calculation determines the ship safety after a collision according to IMO regulations.
The basic idea behind the probabilistic damage stability is to list all potential damages and to assign a probability of occurrence and of survival to each of them. Multiplying both quantities yields the safety index for every damage case, and summing the indices eventually gives the attained subdivision index A. There is only one requirement for this number: it should exceed the required subdivision index R.
The calculation is not as straightforward as described above, though. The probability of a damage case depends on its length and position, but the effect of decks and longitudinal bulkheads has to be taken into account separately by means of correction factors. Plus, the number of damage cases is usually large. The water intake should furthermore be modelled as a quasi-dynamic process and non-watertight openings and pipes could make the calculation even more complicated and time-consuming. Some aspects of the probabilistic damage stability are even erroneous and have not been corrected yet in spite of several reports on these issues. A designer should not be surprised to see negative or otherwise impossible probabilities in the calculation output.
In short, one could say that the probabilistic damage stability acts like a black box, of which the input are the ship size, form and subdivision; its output is the attained subdivision index. The ship designer cannot quickly estimate the value of this safety index to find out whether he is making good progress, nor can he conclude what is causing the problem whenever his ship does not comply with the requirements.
These characteristics make the probabilistic damage stability well suited for an optimisation method applied to a metamodel. This means that a number of experiments is selected, which are ship sub-divisions and their corresponding safety indices, and that a model is constructed matching the original data - it may be regarded as a multi-dimensional best fit curve. The next step is to study the newly obtained model that will provide the designer with a fast way of determining A with a certain error for every possible ship subdivision (and therefore of seeing how design changes affect the safety). A more important consequence is that the designer will be able to choose the best ship among all feasible designs, i.e. the one with the largest safety index. This knowledge will for example enable him to enlarge the ship's load capacity.
Since calculating the safety index consists of many repetitions of fairly simple steps, it should be committed to a computer programme. In this project I have used PIAS for this purpose. Composing a model and performing an optimisation are tasks COMPACT was especially designed for, albeit that its applications have not lain in the shipbuilding industry so far. A combination of both programmes was supposed to show whether the aforementioned expectations could be met. The method would only be declared successful if the calculation time required was not much more than it usually takes to design the subdivision in practice (by trial and error and with the aid of experience). Therefore, the programmes needed to interact automatically. Especially the experiments, which are all expensive damage stability calculations, had to be carried out without expecting the user to be constantly monito-ring the process. That is why I developed a tool that is able to streamline the optimisation. The user is led through the following steps:
- choosing which part of the ship subdivision may be changed in order to increase the safety (this boils down to selecting the variables: deck and bulkhead positions, and possibly the main di-mensions);
- preparing the model, i.e. listing all necessary simulations;
- performing the simulations;
- building the model and optimising the ship size and subdivision.
This procedure has been applied to two ships and the following are the most important conclusions I have drawn from the results:
- The metamodels have proved to be able to mimic the probabilistic damage stability. Their accuracy depends greatly on the designer's preferences. Only future experiments will tell if the models are suitable for all ships, but I have not encountered any fundamental problems which confirmed otherwise.
- A significant increase of the safety index can possibly be achieved if the ship subdivision has not been optimised informally and manually yet.
- Taking into account that A behaves rather erratically as a function of deck and bulkhead positions, one can conclude that only a global optimisation algorithm as provided by COMPACT will suffice for this type of problems.
- Sequential optimisations are promising and appear to be more effective than simple, one-stage optimisations.
- The number of decks and bulkheads is a parameter in the problem formulation. Due to the nature of the design problem it is impossible to create a parametric optimisation procedure, as this implies that the number of variables will not be constant.
- The application connecting COMPACT and PIAS should be replaced by direct communication channels between both programmes. This will not only increase the user-friendliness of the optimisation but also its efficiency.