On Wednesday I went back to the hospital to employ a remedy for the puck disconnection problem. After speaking with Phil, we agreed that under the circumstances the best solution to try was to add a strip of velcro to the bottles, removing the elastic straps, and re-taping the plastic pieces to the pucks. After removing a bottle in the west wing and applying this solution, we both realized the amount of time and effort it would require to apply the solution to all of the pucks in both wings (mostly due to the preparation of the carpet tape). Furthermore, the solution didn’t seem far more effective than the current trigger configuration. Phil explained that he thought it would be best to forget the puck remedy, double check the rest of the triggers in both wings, and just try to get through the weekend. Before I left the hospital I went to each trigger in the west wing and reapplied more carpet tape to the plastic containers, and as I left I think Deepti was planning on doing the other wing. However, I wondered as I left about the effectiveness of the carpet tape after it has been soaked in alcohol.
The next day I learned that eight more pucks had failed, with 1 failure due to disconnection. After bringing the failed pucks back to the lab, I found that 4 of these eight had ‘bad’ sensors, presumably due to super glue. Most of the rest of the 4 were ‘too sensitive,’ which is consistent with a behavior Phil and I noticed in the hospital – the triggers would flash not when the bottle was pressed, but when the bottle was lifted slightly off of the puck. I hypothesize that the weight of the bottle in the basket does not cause the puck to trigger when it is idle; when someone uses the bottle, the puck triggers, but because of ‘drag’ — friction from the bottle/basket or elastic/basket — the puck does not ‘untrigger’. Instead the FSR circuit remains in the ‘triggered’ (I’m being pushed) state, the light having flashed once to indicate a trigger, and if you lift the bottle >5 seconds later, the state of the FSR circuit goes from ‘triggered’ to ‘untriggered’, signaling another flash giving the appearance of a trigger. I was able to reproduce this in a lab, even outside of a bottle cage and elastic straps, with a puck that was ‘too sensitive’. I used the old PuckToRadio program that gives a red flash when the circuit transitions from untriggered-triggered as well as when the circuit transitions from triggered-untriggered. Indeed, the trigger flashed red as the bottle was pressed — only one flash, where a ‘correct’ trigger should give two flashes with this program. As the bottle was lifted, the trigger flashed red one more time. The observation that this occured outside of the bottle holder without elastic straps is also important. I am guessing that another possible explanation for this behavior (other than ‘drag’) is that the difference in this particular puck’s sensor range between ‘unpressed/idle’ and ‘pressed/triggered’ states is within the hysterisis range of the comparator. I’m not sure how to test this. Furthermore, for a particular puck, I was able to reproduce this behavior with some triggers, but not with others. I suspect that this is due to variations in the trimpot setting between triggers.
After noticing this, I realized that our quality control method for the pucks/triggers (go/no-go on the pucks based on resistance values) may not be very effective due to variations in trigger trimpot settings. I believe that a better way to ensure quality control in this system would be to calibrate a particular puck to a particular trigger, and tether the devices as a pair.
Gregg and I wondered about the pucks that came back ‘too sensitive’ or not sensitive enough. Did they pass our quality control before they left? They all had ‘OK” or ‘passed’ stickers. Did they become more/less sensitive over time? Which batch were they produced in? We realize that because components were replaced essentially randomly as they failed in the hospital, and some failed components either got reused or returned to the lab, it would be impossible to track an individual component to determine how it failed. I think that in the future, it will be very important to tag each indiviudal component and track the component each time it changes position or use. This would have been hard to implement in this experiment because of the overwhelming number/types of failure.
After considering this, Gregg and I thought about solutions to puck failures under the consideration of the state and remaining time of the experiment. We figured that one option was to do as many individual puck/trigger calibrations as we could with the equipment in the lab, deliver these to the hospital, and continue to replace in-play triggers with calibrated puck/trigger combos in the hospital. This would take approx 4 hours and require work room in the hospital. We also very seriously considered replacing the pucks altogether with a reversion to Gray’s design. We figured that we could replace the sensor with an on/off switch, which would eliminate the need for possibly unreliable calibration. We understood that the main issue with the previous design was the ‘calibration’ of the plunger, but we figured that we have to calibrate the current pucks anyway, so we might as well try to calibrate these in the lab and we could be sure they work, completely binary. We rounded up as much of the old equipment as we could and assembled some to test. I also spoke with Phil at this point to run the idea by him. He said the he didn’t think it would be worth it to try to replace all/some of the in-play triggers at this point, but he may be interested in using 5-10 of these (if they surely work) for smaller data collection (? I didnt really understand what he was going to use them for, just that they wouldn’t replace any of the in-play triggers). After assembling a test, Gregg and I realized that, unsurprisingly, the old pucks suffered from problems that are at the root of current puck problems: the plunger could not provide enough force to “unpress” the bottles due to the ‘drag’ in the cage. We realized that with variations in bottle cage tightness and so forth, the old pucks with the plungers could be just as unreliable as the current system. I called Phil and retracted my idea.
Gregg and I continued to reflect about issues with the current design. It doesn’t make a whole lot of sense that we are now experiencing the main issue that caused us to abandon the old puck design — calibration. The whole purpose of using the FSR was that calibration could become more automated and digital. The human interface to calibrate the pucks would be controlled by software (and the digital potentiometer) rather than hardware (finnicky plungers). We would still have to calibrate each set individually, but with the software/digital pot, this could be done in less that thirty seconds, and would be very clear that the puck does/not trigger. In fact, the puck/trigger calibration tether is inherent in this design — the software/digital pot routine calibrates the trigger sensitivity to the puck that is currently connected to it. However, the pucks/triggers could be ‘divorced’ and ‘remarried’ to another puck easily, because calibration would be so easy. But this routine was never employed, so we basically ended up with a flashier version of the previous design, bugs and all. It is also noteworthy that using the digital pot/software was abandoned before the experiment started and we really got into the storm of technical failures. I think this routine was abandoned in the frantic of manufacturing, and I think we really should have thought more about that decision. That is not to say that the software calibration wouldn’t have provided its own host of technical issues, but at least it would have been consistent with design intention.
I also reverted back to notion that I had before we got serious about the FSR’s: the triggering mechanism should be on the pump of the bottles. I think another root of many of the issues we are having now is the behavior of the bottles/triggers outside of the bottle cages vs. inside the bottle cages (and the unpredictability of the latter). Putting the triggering mechanism under the bottle not only introduces the weight of the bottle, but also reactions and influence from the fit of the bottle cage into our system. I think that putting the triggering mechanism on the hand pump leaves only the force that the user puts on the bottle into the system. Because of this, our triggering criteria could be very boolean — a simple switch would suffice. The main issue is finding a way to reliable fix the switch to the hand pump, and ensuring that the switch is activated with many different user pumping ‘styles’.
An equipment tally: as we were supposed to ensure that Deepti had all the equipment needed for the experiment this week, she reported that she had all of the equipment she needed. Instead of building the remaining 19 triggers (she said she would have no use other than backup), we built what was near-complete, left the partly assembled triggers that didn’t have motes, and concentrated on delivering more pucks and reworking failed equipment that came back. There is a box of tested triggers and pucks sitting near the door (the ‘outbox’). Some of the triggers may have to be reprogrammed, numbered, and labelled. During puck assembly, we ran out of the plastic/rubber circles from the electronics shop and the rest of the production was put on hold. We have gotten these parts in from the e-shop as of yesterday, and could assemble more pucks.