4 ms·
This somewhat reminds me of the "symbolic AI" software before the first AI winter in the early 90s... It looks like the methodolgy is making progress towards a
by drcode 11y ago
This somewhat reminds me of the "symbolic AI" software before the first AI winter in the early 90s... It looks like the methodolgy is making progress towards a solution, and you are able to find answers for heavily stylized sample problems...
...but then you realize a lot of the complexity is just hidden behind your definitions for the words in the TRIZ matrix (or in the case of "symbolic AI", in the precise semantic meaning you are assuming for the symbols)
- gene-h 11y agoYou should check out the idea of function structures. Functions structures are a symbolic means of representing the functions a electromechanical product must perform in order to do its job.[0] The cool thing about function structures is that you can go from a function structure to a simplified conceptual model of a product. For example, one can replace the function convert electrical energy to mechanical energy with a motor. Unfortunately, the same function structure can translate to many different possible products, some of which don't make any real world sense. For instance, for a carrot slicing machine you might end up with a solution that uses the forward motion of the whole device in conjunction with an airfoil to create the force to cut the carrot. While plausible, this is not the best solution. If some of these problems could be fixed, we'd essentially have a hardware description language for mechanical engineering. And with this we could do the same sorts of design automation the semiconductor industry has been doing except for mechanical systems. [0] http://enterprise.mie.uic.edu/me250/files/1113/4688/4799/wood_stone_function_basis_jmd.pdf http://enterprise.mie.uic.edu/me250/files/1113/4688/4799/woo...