Evaluation Phase
A 'fitness' function is essentially the objective function for the genetic algorithm - it's the thing you want to maximize or minimize as you develop your design; the thing you care most about achieving overall.
A fitness function is used to evaluate how close (or far off) a given design solution is from meeting the designer’s goals.
The genetic algorithm is designed to improve the model's fitness again and again, so defining a fitness function precisely and accurately is vital.
Some examples of fitness functions that could be used in a generative design context include:
- 'Maximize hours of daylight for each desk in an office'.
- 'Maximize circulation in common areas'.
- 'Minimize number of distinct part types needed to assemble an object'.
- 'Minimize number of total parts needed to assemble an object'.
- 'Maximize the structural strength of a critical component in a product'.
- 'Minimize the fuel needed to power a vehicle'.
- 'Minimize the weight of a design'.
You can see here that these suggestions are always framed as either a maximization or minimization problem. As discussed above, this formulation is crucial for an optimization approach to be effective.
Another thing to consider is that it's desirable for fitness functions to be calculated very efficiently by a computer; that is, a good fitness function can be calculated quickly. With experience, a user comes to learn which kinds of fitness functions are likely to be particularly fast or slow.
One of the great strengths of a genetic approach is that the fitness function can be quite complicated without impacting the genetic algorithm's ability to execute.
In fact, a single genetic model can have multiple competing fitness functions for example, minimizing the weight of a design while also making it as structurally sound as possible (this is also known as 'multi-objective optimization').
Only once a fitness function has been defined can the selection phase of a genetic algorithm begin.
Last modified 3yr ago