Conclusions

Discussion
The algorithm described in this thesis has been implemented as a computer program, compiled and tested on Sun workstations with the operating system Solaris, DEC workstations with the operating system OSF1, PC of IBM-compatible type with the free operating system Linux as well as DOS/Windows. The implementation is written with portability in mind. The program uses a simple input file format that is tedius to construct ``by hand'', therefore Fredrik Berntsson wrote a graphical user interface using MATLAB. The interface uses mouse or keyboard for input of the geometry and saves the input data file for the distance function calculation as well as allows the user to start the program from a menu option. Further, menu options allow the user to call routines for visualization of the results. Thus we have a complete system for entering geometries, computing the distance function and visualizing the results. The system can be improved by e.g.

1. including more functions for editing an existing geometry,

2. visualizing non-planar geometries in 3D, and not only as a lay-flat model, and

3. adding functions to import geometry data from some existing CAD system.

These improvements are however not of academic research interest, but essential if a commercial product is to be developed or if the system is to be of practical use for industry.

The algorithm runs fast on the relatively simple geometries we have tried so far and the results from the computations seem very reasonable. Our system can of course not compete with the advanced commercial systems available from e.g. Moldflow, I-DEAS, and C-MOLD, since they offer a system for all stages in the design and not only, as we do, analyze the filling stage. However, some critical parameters can be investigated in a preliminary study by using our program. Further, the designers can, by using our program, gain a rough but very quick understanding of the problems in the mould design phase and thus hopefully avoid some later problems.

Recommendations for further study
Further investigations regarding execution time on more complex geometries and the performance of alternative algorithms for solving the WRP can be done. Methods for increasing the computational speed by improving our algorithm are also available, currently we have not found this to be urgent since the program is fast enough on the geometries we have investigated so far.

Thermal effects like viscous heating and cooling from the mould can probably be incorporated in our method, but at the cost of a more complex algorithm. The question is then: how severe complications are necessary for having a meaningful thermal analysis? As usual, it will be a trade-off between simplicity and accuracy.

More systematic comparisons with results from both experiments and simulations using commercial or research packages are also recommended. Besides the well known commercial packages for possible comparisons, there are research programs written at several European universities. One example is Hele-Shaw Fill within the Diffpack project at SINTEF, Oslo, Norway.

It seems that only the biggest research centres have resources for really systematic investigations here. We have not had such big resources.

The most important improvement would anyhow be to develop a thermal calculation that does not destroy the simplicity of the distance model.

Last modified: Wed Mar 5 15:34:13 MET 1997 by

Peter Johansson (pejoh@mai.liu.se)