Tools and Packages
Dezineforce is pleased to feature the applications of leading vendors including
in addition to the comprehensive dezineforce optimisation suite:
| Hill-climbers |
These methods improve the value of the objective function from a single starting point by moving uphill (or downhill) in the direction which improves the design, and work well where the objective function has a single maximum or minimum. |
| Gradient-based methods |
These use local gradient information to help choose the best direction to move in, and thus improve the rate at which a better solution is found. They work well where the objective function has a single maximum or minimum |
| Evolutionary methods |
These start with a population of designs and construct a sequence of populations of designs using a random (but systematically chosen) element in the search to improve the design. Genetic Algorithms are one of the better-known types of evolutionary search. Other types include simulated annealing and evolutionary programming. These methods can often avoid being caught in a local maximum (or minimum). |
| Design of Experiment and Response Surface methods. |
These are techniques which sample the objective function and construct an approximation to it - this is very effective where each of the performance simulations is expensive and might require hours of computer time and can dramatically speed up your search for a better design. This so-called "surrogate" surface is then searched using the previous methods and then further refined using additional performance simulations. At the end of the optimisation the improved designs proposed to you are always simulated using the performance simulation you have set up and these results are available to work on further as part of your design process. |
| Hybrid algorithms |
Combinations of the above, that assist the designer by initially casting a wide view over an objective function which might be expensive to calculate; helping to focus in on particular parameter values that improve the design significantly; and then zeroing in to locate the best design in that region. At each stage you gain further insight into your design and the tools help you make further choices to improve it. Such hybrid methods can also yield considerable insight into the robustness of your design where each of the dimensions might be subject to manufacturing tolerances. |