Problem Statement:
Design a 3rd generation puck that works on the table or in a basket, that requires only a few seconds and no mechanical skill to install and calibrate.
[iframe http://spreadsheets.google.com/pub?key=tgjtlUHk2rdJLnloaMxvduQ&output=html 1050 700]
I did a repeat of Phil’s specificity/sensitivity experiment. The first time through, the data was all over the place probably because of a mote time/our time mistake. Yesterday, Henry and I did it together, and so far the results look promising.
We used two rooms – a Nurse’s Station and Room One. Mote 49 was at the Nurse’s Station and mote 46 was in Room One.
We conducted this experiment by recording the mote time when Henry entered and exited each room (so that we wouldn’t get error values because our time and the clock time weren’t matching) and which room he was in. Then I used the script to label each message with the room he was in at the time it was sent.
From this graph we see that when Henry was near the Nurse’s Station, the RSSI between 49 & 233 were high and between 46 & 233 were low. And the opposite was true when he was in Room One. This was what we expected!
–post continues after page break–
I tested the spider & the spiral antenna this afternoon. Here’s the data:
SPIDER ANTENNA:
SPIRAL ANTENNA:
Neither of this give a good distinction between 1m and 1.5m.
There was a short antenna in the box, which I tried out today. I’m not sure which one it was, but here are the results:
This one also doesn’t give us the distinction between 1m and 1.5m that we’re looking for. So far the best one was the whip antenna.
I tried the whip antenna as a candidate for the anchor antenna.
Vertical:
The vertical orientation didn’t give the distinction we were looking for.
Horizontal:
The horizontal looked better with separation- but there was still no vertical line that could divide near from far. What’s strange is the RSSI for .5m and 1m- they seem to be switched in position.
After some time searching the internet I’ve compiled this list of mini CNC mills and kits that could be used for the lab at a reasonable price.
[iframe http://spreadsheets.google.com/pub?key=tiZVt-VslPxrTq0yFxE4H2Q&output=html 1100 325 ]
Two of these stand out to me as being the best deals.
The Sherline 5000 I like because it comes assembled and with a variety of accessories. Pretty much everything we need will be up and running when it arrives (supposedly). The only downside I can see is that it has a relatively small working area, especially for the y-axis.
I also like the Fireball V90, this is a kit constructed with MDF but claims to have a high resolution (0.001″ for a full step, 0.0000625″ for a sixteenth step). Combined with one of the stepper motor/driver packages available on the website it should require only a router to be as complete as the Sherline. It should also be altogether cheaper than the Sherline, with a larger working area.
The third FSR test incorporated a new test setup in which the lid freely floats on top of the sensor.
The above drawing illustrates this design. In the front view, you can see the smaller rectangular pocket in the base where the sensor sits. On the lid, there are three protrusions that serve two purposes: two prevent the lid from rotating, and to absorb lateral forces caused by a moment on the lid about its center. In other words, the general idea is that if a force were a applied in the exact center of the lid, all of the force would go directly downwards, causing the lid to translate a bit as the rubber on the FSR deforms, and transferring force to the sensor. If a force were applied on the outside of the lid, you can imagine that the lid would ’tilt’ in the direction of that force. If the clearance between the protrusions on the lid and the inside of the bottom of the apparatus is small enough, there wouldnt be room for the lid to ’tilt’; instead, some of the downward force on the lid would be resolved laterally on the protrusion (creating friction opposing downward motion), and the rest of the force would be directed onto the sensor. So we are looking for the kind of fit that reduces these two things, while not allowing the lid to ’tilt’. So there is necessarily a tight fit. The problem is solving the former problem, while not introducing too much friction and lateral force loss that we are outside of the resolute range of the sensor.
During this experiment, I first tested a range of forces (in all cases, appx 1.7-4kg) on the center of the disc as a baseline measurement. I marked test points at two radial distances (1.75cm and 3cm) in sixty degree increments. For each of these test points I applied 8 different forces in the test range, two separate times. However, I found that the outside set of points (3cm) caused the lid to tilt such that the bottom of the lid met the top inside surface of the bottom — what we are trying to avoid. Resistance measured under these conditions was in the range of 1.5-2.5 Kohm — way outside of expected results. The fit in this test setup is not precise enough. I did get data from the rest of the points, and a scatter plot is shown below.
Each data series is a different polar position from the center of the disc in 60 degree increments. The series marked ‘O’ is the center of the disc. You will notice that each of the polar positions have generally the same shape of fit line (this is confirmed by a general linear model that gives approximately the same coefficients of ohm response for each position), but are offset. It is interesting that one of the positions ‘D’ is offset more than the others, and I am not sure why (I would expect three way symmetry due to the geometry of the apparatus). Most importantly, you will see that the origin data has a different shape than the rest of the data, giving a higher slope and more separation in the low end of the force range. This suggests that, at least in this particular apparatus, that as you apply force anywhere other than the center of the puck/bottle, it will be harder to detect differences in forces.
I suspect that the fit between the lid and the rest of the puck is to blame. The solution to that problem is counter to our objective. I see two solutions: increase the thickness of the lid (which negates the purpose of the new puck design — smaller size), or have a tighter tolerance fit between the lid and base. The latter solution is essentially the same reason we abandoned the previous test design.
However, I do like the idea of this arrangement because of its simplicity. If we are able to achieve good enough dimensional accuracy, it would be fairly easy to manufacture these parts. According to the diagram, I have allowed a fit tolerance of 0.01″. This visually appears to be consistent with the model we got back from the 3D printers, but I still question the precision and accuracy of the parts they are able to produce. I will talk with Matias and see if these tolerances are something we could expect to produce from available tooling and material.
I have briefly looked for literature on this kind of mechanical system but I don’t think I know the best kind of resource or exactly what to look for. I would appreciate suggestions.
I am also starting to wonder if we should take the apparatus that gives us the cleanest sensor signal and try to produce an electrical system that gives the response we want. We really haven’t asked “how clean is clean enough?,” and it may turn out that we are able to get one of these devices to function as they are. Are we over analyzing this problem?
Chris & I did a quick experiment in the hospital. We wanted to test far & near from the anchors.
From this graph- we can see that the difference between far/near is not very pronounced in the RSSI. The highest RSSI value is equal at both distances. However, the signals that were picked up were markedly different in the two distances.
From the data that we have- this room did show a difference between near and far. The RSSI for near was much higher.
Room Two also showed a difference in the RSSI, but this difference was a lot smaller. Near was still greater than far.
Room Three data also shows a much higher RSSI for near versus far.
Overall, the trend is that the RSSI for near is greater than that for small. However, I think that we should take more data points next time.
Here is a link to the data: http://docs.google.com/leaf?id=0Bzml4qs0jKAXN2MyZDljNjMtMWI4Yi00MzNkLWJkZDEtYzE1NmVjOWMzMzVm&hl=en
Howard has been encouraging me to build a CNC machine for some time. I think we are getting close to the time when one would really be handy, since we’re using the one in the DfM lab and that lab becomes quite busy towards the end of the semester. Rogge, Matias, Brian and the rest of the solar bike crew built a large CNC machine last year. I’m thinking of something much smaller optimized for very small parts, perhaps with a 6″x6″x4″ work envelope for cutting circuit boards and making small 3D parts. There are detailed plans for such a machine on instructables.com. I’m also looking carefully at the linux motor controller that will allow us to configure the device to run g-code, among other things. The cost would be a few hundred dollars and probably a week or two of lab time, so it seems like a reasonable investment.
Here are the current schematic and board diagrams for the door minder
Powered by WordPress