While optimization is an incredibly powerful technology, it is only one available method in the toolbox that is design exploration. So if optimization is only one option to proceed with design, then what is its role? Optimization methods are exceptionally good at one thing: providing a very specific answer to a highly specific question. Do you need to design your part to meet a specific set of criteria? Then optimization is the right option for you. But let me pose another question: how often do your goals remain static and unchanging? Due to modified requirements, shifting objectives, or even just plain uncertainty, the truth is that many project’s requirements remain a work in progress until they are completed. The best analogy I can imagine to explain optimization versus exploration is the difference between using your smartphone to translate a foreign phrase versus learning a language for yourself. Like optimization, the smartphone can get you that very specific answer with little effort. However, you’ll need more if you need to exist in a more dynamic environment such as a conversation.
So what has held back design exploration from having the successes seen with optimization? One limitation has certainly been hardware. Exploration requires many simulations, but modern machines and cloud computing have expanded the possibilities considerably. I think the biggest limitation has been software accessibility: learning about design exploration can very much feel like learning a foreign language. Subjects such as statistics or probability are not part of the common engineering curriculum and commercial tools typically use terminology and jargon reflecting the highly mathematical source material. Furthermore, too little is done to make it more comfortable for someone who is not a subject matter expert. But it doesn’t need to be this way! The main effects that come from a design of experiments? They are nothing more than data trend lines. The R-squared value from a regression? It purpose is to inform you what percentage of the data’s behavior is explainable from cause and effect. The concepts are not hard to grasp, they simply need to be explained.
At a more general level, the concept behind all design exploration methods is to answer the questions of “what if?”. Design of experiments ponders the question: “What if I don’t know which parameter has an effect on design performance? Which parameters increase, decrease or do not change my design?”. Response surface methods address the more specific question: “What if I need to change my design due to unforeseen circumstances? How can I investigate the effect of these changes as quickly as possible?”. Reliability methods answer question such as: “What if my design is not as reliable as it should? Let me find that out and I can optimize for reliability if it is not. Or What if my design is too sensitive to design variations? Will it still be reliable?”. And our old friend optimization gives us the answer to “What if my design needs to perform a certain way? To what values should I set the parameters?”. From these points of view, design exploration ideas are relevant to almost any project.
I think the time is now to get on the cutting edge and adopt design exploration into everyday engineering. So where do you start? Let’s go back to the language metaphor one more time. While it is possible to teach yourself, I think most experts agree the best path to success is to have a tutor and possibly even immersion. In my opinion, design exploration is no different. Grab you project, contact your local Altair support and we can help you dive right in and guide you through the process.