The largest issue that I have faced in the puck project was schedule management. It is obvious that most design changes and troubleshooting occurred in the last week and a half of the deadline. While the project was active in the weeks leading up to the deadline, most design was selfish – in the sense that I was making prototypes, evaluating them and loosely passing them off to other team members in a staggered fashion (I didn’t deliver enough fully assembled, ’state-of-the-art’ products), and making design changes within this small bubble of evaluation. I think that a more correct process would have been to work full force on complete designs that pass on to team members that will be deploying the product (Deepti). Less prototypes made with more numerous/major changes at each step. I think that stronger direction in design (less ‘wandering’) would inherently follow. The important thing is that other team members are working with fully assembled products that have actual relevance to the project goal rather than parts of products that have little relevance to what people will actually be working with. This may be difficult to do, but I think it can be accomplished if a deadline schedule is carefully planned and communicated – the more in advance, the better.

Also, while it was important to create a new design and a new generation of the product, I think it may have been more effective to spend more time refining the original (Gray’s) design. I think that with careful consideration and the same amount of effort that was put into the new puck generation, we could have solved the underlying issues with the previous generation and would have ended up with a more robust product that is less complicated to assemble and manufacture. But I think this is more a question of overall project direction?

Specific changes, small to large – in no particular order:

FSR circuit board:

-rearrange board so that UI components (switches, USB) are close together – prettier, easier to panel

-move FSR connection closer to edge of board, move/reorient battery clip closer to edge of board, move selector header to get as many components out from under clip as possible.

-try to move IDC connector so it could be de/attached while boards are still screwed in

-order USB receptacles with mounting feet

-I realized after the proposed battery switch that a different battery with different power properties and form factor may have been better suited to this project. For example, if using the iPod batteries with the small connector, the coin battery could be ditched and the FSR board could be decreased in size by about 50%. This could substantially effect the rest of the body design – maybe sandwiching the two boards and the battery, maybe just the mote and the battery, etc…

trigger body:

-integrate ’spacers’ into body

-integrate ’spacers’ into lid to decrease the distance between counterbore and screw engagement

-add third screw to FSR board (or move second screw) closer to UI components/USB

-more carefully evaulate part thickness trends. Would be useful to increase space under FSR board and perhaps take a tradeoff in lid thickness

-consider a different fastening system for attaching the two halves of the trigger body. Experiment more with plastic clipping. Otherwise, add two more screw holes to secure each corner of the body

discs:

-consider threadless fastening options

-consider a ’skirt’ that could hide dead space and decrease overall thickness

-find more robust ways to fasten rubber and FSRs

overall:

-concentrate more on fault tolerant design. There were very low tolerances with important assembly areas such as screw holes and mating.

-organize a standard ‘tool set’ that has procedures and decision making tools for choosing mounting types, screw types, hole sizes, etc. Try to decrease the number of different types of hardware.

DISCS:

We currently have 40 working discs and 26 discs that are out of tolerance. Today, I made an order with John in the electronics shop to cut 130 (4 for each disc: two for each piece) plastic shims that will be used to put these discs back into tolerance. He is also cutting 90 additional rubber ‘dots’. Assembly and troubleshooting instructions are on an earlier post on the blog. Howard is working on one more set of 9: if you find that the rework on the OOT discs fails, let Howard know that he will have to manufacture the rest as soon as possible.

TRIGGERS:

Steps to manufacture:

Assemble motes: IDC ribbon cable must be attached to the stripped motes. Phil was working on this when I left. I have cut and stripped more than enough (to meet the qty 75 target) cables to the correct length and have crimped enough to make 70 target. Eventually, we will need to order more connectors and crimp an additional five. To crimp these cables, use the cable on a working trigger assembly as a guide. The orientation is important. I have been using the green arbor press (moved from the DFM lab to our lab) to crimp the connector, but something such as a vise should work as well. The length is measured relative to the connector so that the cable should be crimped with the end of the cable roughly flush with the outside of the connector. EG, insert the cable into the connector latch in the correct orientation and line up the end of the inserted cable with the outside of the connector. Ensure that the cable is in straight and use the press to crimp the connector. It can be kind of tricky to get the cable in the press and make sure it stays straight at the same time. Keep in mind that the once the connectors have been latched they can not be reused or recovered. The cables should be attached to the mote first by soldering the four USB leads in order. There are three other cable pins that require connection: starting with the wire furthest away from the USB leads (call this wire 1), solder wires 1 and 3 together into the Vcc header pin on the mote. Solder wire 2 into the UserInt pin on the mote.

Assemble FSR boards: we should have all but 11 of the FSR circuit boards required for the 75 qty target. Eleven will hav eto be started from step 1. The circuit boards should be located in the gray materials cabinet. Population should be pretty straightforward with two exceptions: do not solder the digital potentiometer (4-pin DIP IC) and replace the 560 Ohm resistor with a 5K Ohm resistor. I’ve found that the solder spoon is very helpful when soldering the comparator and the header pins. We should have enough header pins for the rest of the 11; they are located in the parts cabinet drawer labeled ‘0.1” headers’. You may have to clip more from the bulk headers. Gregg (and Geb) has experience in soldering these circuit boards. These eleven boards will have to be tested before deployment. This will be easier if the battery holder is not populated until testing is complete. After testing and the battery holder is attached, the solder pins and excess solder must be clipped from the underside of the board to ensure a good fit in the plastic housing. This has caused problems with the lids fitting. Make sure to clip every pin as close to the board as possible with special attention to the battery pins (which require larger holes and more solder).

Assemble triggers: the two boards can be assembled in any order, but not simultaneously due to insufficient clearance for the connector. Use two spacers under the bottom most mounting holes under the mote; the FSR circuit does not use spacers. I have been using 1/4” screws for the FSR board, but the plastic 5/16” screws work as well. The FSR board requires 3/16” screws. Assemble one board, attach the connector, and then assemble the other board. The assembled triggers then need to be tested with verified working pucks (set the trim pot to roughly half a turn).

Assemble lids: the plastic panels for the lids are in a plastic bag on top of the gray parts cabinets. I have used the blue Loctite superglue to attach them to the inside lip of the lids. Make sure that this inset is clear of plastic burrs or dirt before gluing. There are three holes on the panels: one for the switch, a circle for the button, and the largest hole for the USB receptacle. To align the panel, I place the panel laterally so that a little less than half of the material between the USB hole and the edge of the panel is covered by the lip on the puck lid. See an assembled puck for reference. You should also allow about an hour for the glue to cure. After all parts have been tested, assembly the lid to the trigger body using 9/16” screws. When I assembled 31 triggers, I encountered a >50% fit failure. This is mostly due to the trigger body flexing. Another cause is incorrect hole alignment between the one FSR circuit board mounting hole and the counterbore in the trigger body. To troubleshoot these problems, first ensure that the screw path from the counterbore to the screw hole is straight. If the mounting hole in the FSR circuit board is not aligned, remove one of the mounting screws (the one closest to the USB receptacle) on the FSR circuit to allow the board/mounting hole to float into alignment. If assembly still fails, it may be because the ‘top’ screw (that passes through the mote) is not long enough to engage its threads. I trimmed 3/4” plastic screws to be a couple millimeters longer than 9/16” (you may also try using 5/8” metal screws). Using plastic screws, you may find that the lid does not completely fasten to the trigger body, but the screw should stay in and prevent the top of the lid from rotating or flexing out too much.

PRO/E:

All current files are in S:\ProE\Current Puck 7-26. These files should be concurrent with Howard’s files and include both the MFG sequences and CAM code that he generated.

Starting from assembly, I have generally been using 5/16″ screws on the new style and 1/4″ screws on the old style. If I am screwing into the old style puck, I place my finger over the other end of the screw hole on the top and screw the discs together until I can start to feel (with my finger) the screw start to breech the surface of the top disc. This should put the tension on the sensor in a good range to test. Assembling the new style pucks is a bit easier: you can visually confirm that the screws are tight enough because of the through holes. I screw in until the screws are about flush with the top disc, and then back off a couple turns.

I then try to ensure that each screw is screwed in an equal amount: you can look through the side of the puck, between the discs, and you should notice that as you rock the disc around the sensor from all directions, the two pieces of rubber should stay in contact. If you notice that the pieces of rubber hinge on each other, you probably need to tighten the screw in that direction.

Then I test each puck: connect the sensor to an ohm meter. First ensure that the sensor connector is reliable by holding the puck down and applying force with your hand. Wiggle and rotate the connector to see if it sticks at infinite resistance in any position or other odd behavior. Note that the sensor may naturally fluctuate to open circuit momentarily (this may be a behavior of the meters, not the sensor), but if the connector is faulty it should be quite obvious. The criteria for a good puck (under the current design) that Geb was using was that the resistance should be greater than 6KOhm resting (both with and without a full bottle) and less than 4KOhm under the gentlest press that dispenses lotion. You can increase the sensitivity (if the resistance is too high) by increasing screw tension. If you must increase screw tension so much that the screws bulge out the top half of the disc, you should decrease screw length. However, if you use the screw lengths for the type of pucks as I’ve indicated above you should be okay. I’ll note that I haven’t taken data and determined whether increasing ’sensitivity’ by increasing screw tension changes the sensor response in a linear or exponential manner, or simply offsets it. This question should not be important to us now (as long as the puck meets OK criteria), but may become important in the future as different triggering mechanisms are implemented.

The biggest time consumer was the pucks. Using the velcro took a long time, but to prepare for that for the full test, we can set out the pucks the day before, or do half of the velcroing in the lab. We only put the pucks on the brackets outside the patient rooms or on the desks, we skipped any bottles inside patient rooms.

Putting the motes in the patient rooms went smoothly. They were plugged in to the panels, so we didn’t have to worry about battery life. Pyramids were placed at the nurse’s station and in the doctor’s work room. They were plugged in to the computers, and we had no problem with that.

We waited until the night nurses left to hand out pagers to the day nursing staff. Most of the doctors came to round at about 8:30, so we handed out pagers as they came in.

All in all, the parts did well in the hospital. We had some glitches with the pucks, but everything stayed in place, and nobody messed with anything. We left one bed mote in a patient room, and three pagers are missing.

The data didn’t look the way we expected it to look. The pagers heard from the pyramids, but not from the bed motes. There are very few bed mote records in the data sets, though we saw nurses walking around. We see times where a puck was hit, but then the nurse disappears until the puck is pressed again. Possible software bug?

There were ten pagers and 4 anchors.

The anchors collected the following number of records:

42358 records in 231810 seconds;

41941 records in 235157 seconds;

46818 records in 235131 seconds;

42474 records in 235237 seconds;

The pagers collected the following number of records:

105960 records in  235240 seconds;

106950 records in 235227 seconds;

105630 records in 235233 seconds;

106681 records in 235224 seconds;

105152 records in 235191 seconds;

105631 records in 235240 seconds;

107641 records in 235220 seconds;

106471 records in 235285 seconds;

106655 records in 235284 seconds;

106232 records in 235270 seconds;

GKxxxxxx     /dev/ttyUSB0      Grok Lab TelosB Mote

where xxxxxx is a unique identifier.

The new TelosB motes do not identify with a unique serial number on the USB bus. The unique ID should still show up in radio communications where it is taken directly from the id chip (I suspect). The brand new EEPROMS which should contain the serial number and identification string for USB identification purposes, however, are factory floor blank, and so the ID reported via USB is (none). With luck someone will have an initialization program which will allow the MCU to program the EEPROM. Without luck we will have to write one ourselves.

room 1 = bed mote 165
room 2 = bed mote 167
room 3 = bed mote 166
hallway = pyramid 200

exp started

2:45 – 2:47 set pager by pyramid (around mote time 3000)

2:47 – 2:52 pager in “far” positions of room 1

2:52 – 2:57 pager in “near” positions of room 1

2:57 – 2:59 set pager by pyramid

2:59 – 3:04 pager in “far” positions of room 2

3:04 – 3:09 pager in “near” positions of room 2

3:09 – 3:11 set pager by pyramid

3:11 – 3:16 pager in “far” positions of room 3

3:16 – 3:21 pager in “near” positions of room 3

3:21 – 3:23 set pager by pyramid

3:23 – 3:33 pager in hallway, out of rooms

3:33 – 3:35 set pager by pyramid

From the data collected:
In Room 1 from: 3214-3777
In Room 2 from: 3921-4507
In Room 3 from: 4658-5233
In Hallway from: 5287-5892

Analysis:

From this graph, we see a clear distinction when the pager was in room one. There are a few data points that are lower, but for the most part, it was clear that the pager was in room one from times 3214-3777.

Between the times that the pager was in the rooms, we see much higher RSSI values. However, there seems to be some odd values when the pager was in the hallway (at the end of the experiment).

Among the rooms, room three also has a clear distinction, but things get messy with the hallway signals.

From this graph, we see almost no distinction from in the hallway/ in the rooms.

The 2×2 Tables:

More Statistics:

• Analog input (from FSR battery readout) connected to ADC0
• INC from digital pot connected to GIO0
• U/D from digital pot connected to GIO1
• CS from digital pot connected to GIO2
• Vout from FSR circuit connected to UserInt
• Vcc from FSR circuit connected to AVcc