Things you don’t learn in physics December 3, 2010
Posted by mareserinitatis in electromagnetics, engineering, grad school, physics.Tags: engineering, physics, sea urchins, simulations
trackback
One of my fondest memories of doing my masters degree was when I decided that I was going to prove out everything I’d been doing in my simulation work. This was in the days before we had our nice anechoic chamber.
I’d been designing antennas, so we took a cartful of antiquated equipment into the middle of a field, where we would ostensibly not have a lot of electrical interference and took a bunch of patterns.
When I got back in and was able to analyze my data, I realized that aside from the fact that my pattern seemed somewhat misaligned, I had something that looked like a sea urchin.
For all the effort of dragging that equipment around, it turned out that I’d managed to ruin things by having my cell phone in my pocket: it operates in approximately the same frequency range as my antennas, and the spikes were probably the phone attempting to communicate with a nearby tower.
That was the day that I learned how difficult it can be to go from equations or simulations to making something that actually works the way you want it to. When working on my undergrad in physics, I became adept at looking at analytical solutions at things, but I don’t think I got a really good handle on what they really meant by “ideal case” until I went into engineering.
Right now, I’m starting on a new project that involves building certain package components. I was discussing the project with the engineer who will be designing and building them (where I will be modeling and possibly testing them). We were discussing the electrical properties of one of the objects, and without thinking, I reached up, grabbed a textbook, and started discussing the theoretical behavior.
He made some comment about “that’s how grad students think”.
We have this problem. He is not sure what his yield will be, so he wants to make a whole bunch of variations on the design. However, each variation for me means a lot of time in terms of building my models and then computer time running them. I’d really like to keep my modeling work limited to about a dozen designs so that I’m not sitting there, still working on things next year at this time.
But when you build these devices, you don’t know what your tolerances are and how difficult it will be to make the device to spec. There may be problems with process, and so you may have to make a lot of changes to see what works.
It’s almost certain that it’s not going to operate the way it does in the simulations.
Sometimes it makes me think that modeling is a somewhat useless exercise. On the other hand, who has time to build and test hundreds of devices to see what works? (I mean…other than people like me and the other engineer.) At least with modeling, you have an idea of how it should work, and you can also tell if something is really amiss. Once you understand how your device deviates as it goes from simulation to testing, you may have something that makes it easy to predict what device will work for a particular application. Or maybe you can incorporate more and more of the real-life dynamics that affect your device into your model to make it increasingly realistic.
The scientist in me loves to come up with predictive models. It’s just that sometimes I feel like I’m trying to develop a perfectly spherical sea urchin.
I also blog at Engineer Blogs, home away from home to some of the best engineering blogs.
Ah Cherish, it’s awesome that you seem to be at the very edge of the scientist/engineer crossover. I agree with your feelings about modeling, however saying “What’s the point” is hardly necessary. Modeling proves some very important things and disproves many others (probably more important). I think for whatever you are modeling it’d be best to get 2 or 3 protos going and getting them working in SOME way, and then going back to your modeling with these new inputs.
When I am doing SPICE simulations on a new idea, I find that building and simulating at the same time are very important. If I spend all the time up front with SPICE, I often miss something critical and waste all the time I spend modeling (“What? This op amp can’t output 10 amps? What the hell?” <– not really that bad). But still, getting it working and then pushing back to the model with new constraints seems to be the best solution to me.
That’s very much what we’re doing. The other engineer is designing the part (because he knows the manufacturing specs) with some input from me on design. Then I’m going to go model them, he’s going to build them, and we’ll test them to see how model and behavior actually compare.
I think some of it is getting out of the mentality of getting things right to many decimal places, though. It’s amazing how there is this huge focus on accuracy in school, but when it comes down to it, the accuracy we want is unachievable, with some situations being worse than others. I used to always look at experimental data as suspect, but I think it takes an equally critical eye when looking at modeling data. :-)
Boy, you guys are spoiled. As chip designers, we don’t have the luxury of modeling and building at the same time since each time we build, it costs a million dollars or so. So it’s all about modeling, simulating, more modeling, more simulating,…
But Fluxor, you’re still at an advantage. You have experience with what you’re doing. Not only do I not have experience, this is treading new ground entirely. I have never done this before, the other engineer hasn’t done this before, and there are maybe a handful of papers even addressing the topic.
And of course, in a paper, the sims always line up perfectly with the expected results.
So I have no idea where my sims will be off nor how much.
But that’s what makes it fun!
[...] about how people who don’t believe in modeling often don’t understand validation, how one needs to understand the limitations of modeling, and also why I find modeling fun. As I’ve addressed most of the main points, you might [...]