-3
$\begingroup$

How feasible (in practicing engineer sense) do you think is the general concept of "computational engineering"? That is, engineering design carried predominantly on computers, with as less physical hardware and "hands-on" as possible.

Sure CAD/CAE/etc. are standard tools, but their place as tools is currently lesser than what could be possible in software.

But does it hide something that someone coming from a pure software and mathematics background cannot see? What are the blind spots that you think software cannot handle? What can it handle?

Could computer simulation replace engineering tacit knowledge?

Related: How do engineers really use numerical simulation?

| improve this question | |
$\endgroup$
  • $\begingroup$ I think that you will have to much more clearly define what "computational engineering" means to you. Many different approaches to refining and optimizing systems use computer methods. But having a computer program design something completely independently is a sign of the "singularity". $\endgroup$ – hazzey♦ Aug 14 '15 at 0:58
  • $\begingroup$ Here it means all feasible computational engineering. I would not doubt that it's a philosophical possibility in the field that computers could some day replace engineering design. That is, computational engineering here means "whatever computational engineering is possible". $\endgroup$ – mavavilj Aug 14 '15 at 17:44
  • 2
    $\begingroup$ If that is your definition, then this question is too broad. $\endgroup$ – hazzey♦ Aug 14 '15 at 21:01
  • $\begingroup$ Perhaps, but perhaps not. Specifically if we consider the question in the sense of "can mathematics and computer graphics in theory tell everything there's to tell in engineering and if not, then what cannot it tell". It's a broad question, but even more a "universality" type of question. $\endgroup$ – mavavilj Aug 15 '15 at 10:45
  • $\begingroup$ There has been some work towards this. Optimizing the hell out of boundary conditions even testing variations is easy enough but the early design process is no closer than computers doing software by themselves. $\endgroup$ – joojaa Jun 29 '18 at 13:11
1
$\begingroup$

At our current level of software technology and accessible computing power, computers can't do a whole design.

Computers can automate some parts of the design process that can be well specified in a formal sense. For example, once you specify what pads have to be connected to what other pads on a PC board, minimum trace/width spacing and other rules, software can often do the actual routing of tracks.

Even this is far from fire-and-forget with current technology, but the big issue is that this is only a small part of designing a electronic device. Humans still have to decide what the device is supposed to do. Nailing down the specs is a important part of designing something, and these kind of constraints are hard to codify in a machine-understandable way. There are just too many possibilities such that software that understood all but dumbed down and contrived specifications is beyond our reach today.

Another part that would be very difficult for software to do is the actual creative engineering design part. There are a vast number of possibilities, each with its own set of tradeoffs. A computer could possibly verify that a particular design meets a particular set of specs (although, again, codifying those specs for anything but toy problems is too complex currently), but finding the optimal or reasonably optimal design is much much harder. That's where the creativity, knowledge, and experience of the engineer comes in. It will probably be quite a while yet before computers can mimic that for broad classes of real world problems.

| improve this answer | |
$\endgroup$
  • $\begingroup$ Yeah. However, as someone coming from the abstract side. I do believe that engineering design process is procedural and that physics has a fully mathematical expression. That is, there's some clearly visible set of "variables" that are based on physical laws and mathematics. And those could in theory be fully modelled in software. So one could e.g. get automatic dimension and material property designs. Which would leave engineering design to think more about how the "blocks" should be combined and "localized" to the particular design scenario. $\endgroup$ – mavavilj Aug 13 '15 at 21:55
  • 3
    $\begingroup$ @mav: Again, checking correctness or filling-in some well defined parameters is very different from the creative process of synthesizing a design. Once you've reduced the problem to picking the right parameters, most of the creating design work is already done. Computers aren't good at picking values for parameters not yet even imagined. $\endgroup$ – Olin Lathrop Aug 13 '15 at 22:12
1
$\begingroup$

One very significant thing that "someone coming from a pure software and mathematics background" might not "see" is the role that validation and verification play in the complete process.

That can be more sophisticated than simply "building a prototype to show that the computer-generated design actually works". A more effective way is to verify and validate the techniques used in creating the design, so that, based on experiments, you know when the software has actually designed something, and when it has only produced some pretty computer graphics.

You can characterize that as redefining the purpose of engineering "research and development", from "figuring out how to fix stuff that doesn't work" to "figuring out how to acquire technology that does work, within well-defined limitations".

That may be a long way from "the singularity", but it can be orders of magnitude better and faster than what an experienced team of human designers can achieve on their own.

| improve this answer | |
$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.