29 May / Automatically beautifull: is it possible?

I don’t think it is possible to generate beautiful things with no guidance. This topic may refer to “Creationism vs. Evolutionism debaters”, with no success of concluding remarks I would say. And yes, this can be considered as a high-level discussion whose nobody is able to point an assertive statement with so much confidence. Given that I am not intended to go deep into this discussion, let me describe my modest experiment. It cost me some money (=time).

Everyone who works with electronic prototypes wants to have well-finished enclosures for PCBs. Engineers are inclined to categorize things, put into boxes, organize results, use as many primitive shapes as they can. This feeling was considered as an initial step for the attempt of making an algorithm which is able to: provide intuitive front panels from a given set of components/connectors. The following unexpected behavior arose from the first attempts:

  • There was more than one output which could be considered “beautiful” by the algorithm.
  • The so-called “aesthetically pleasant” ratios/proportions are at least 18 and have variable precedence rules.
  • Manufacturing constraints are not the most stringent ones.
  • As implementation problems arise, several parameters had to be added.

The complexity has grown to such level that it would not be practicable to develop the algorithm without imposing my own preferences to the user. It was possible to leave it free for him to decide, but at a cost of a huge number of configuration parameters. The parameters could be constrained too, but it would be another way of imposing my own preferences. Other issues emerged later, like how to validate the beauty? What if the generated options (assuming relaxation for the first observed behavior above) bring the final user to a Dilemma?

I realized that complex designs could generate errors for ill-defined parameters. So, I had additional issues like how to guarantee that there will be no errors? Probably, I should write a theorem. How to define the bounds of the parameters that wouldn’t combine into an error? As each of these issues was very stringent for something intended to speed panel design, we chose the Middle Path: panels.mwf-services.com.

We learned several lessons which helped us to define the concept of our panel designer. Separate components into frames, fast alignment options inside each frame, complexity growth by subdivision, instant sketch visualization and many other features which would help the designer to get a nice looking panel with minimal time and money investment.

Post a comment