Differentiating system versus user

[afm]

Yes, Steve your right. when users put in effort to aid the motor, we have a current drop. One exercise clinicians do with patients on the device is have them resist the movement of the device and hold the footplates still. When that happens we see max current draw and the current trendline "flatline" as it is being held. I can make that distinction using current, or also using the encoder sensors; if the sensor is stuck in the on/off position, we know the user is more or less holding it in place.

I will build out some logic for that differentiation as well as others' input from earlier, thank you for that thought.

 

[Steve Bailey]

Presumably the average motor current will be proportional to the weight of the person using the treadmill. For more accurate measurement you may want to establish baseline current profiles for different weight ranges of the person on the treadmill. To simulate a passive 80 kg person, you could put 40 kg on each footplate and chart the current vs time or shaft angle. A deviation from that profile with a live 80 kg person indicates effort.

 

[MaxK]

I am working on a project that needs to detect when a user is putting in effort into the system (this is a robotic gait trainer that functions similar to a elliptical).

Do you have data on the interaction of your device with the patient (for example, frequency change curves depending on the "efforts" made by the patient) (hmm... what "efforts" is?)? What range of musculoskeletal disorders are covered in your data?
Can you answer questions about how “constant” the patient’s influence on the device is (for example, if a patient has one leg damaged, will the “interaction” be the same as for a patient whose both legs are equally functional, etc.)?
I drew the curves in #44 to demonstrate how difficult it can be to identify useful information from the input data (look at the pulse duration curve and imagine what will happen if the patient's impact on the device is uneven - the curves will turn into "chaos").

 

[afm]

Hi all, sorry for the lack of response this project has been on pause as the UI team wanted to do some development on it with the alpha software requires "scans" for finding baseline.

Presumably the average motor current will be proportional to the weight of the person using the treadmill. For more accurate measurement you may want to establish baseline current profiles for different weight ranges of the person on the treadmill. To simulate a passive 80 kg person, you could put 40 kg on each footplate and chart the current vs time or shaft angle. A deviation from that profile with a live 80 kg person indicates effort.

Yes, the sensitivity of the current output was based on how I set the jumper on my 4-20 mA current transducer. Without a jumper, changes in load, speed, clutch current, etc. resulted in significant changes in the sensor value. Where I currently have my jumper, changes in load are not as significant. I went through a lengthy data collection phase to collect various scenarios of settings (change in speed, step length, assistance level, body weight support off load, etc.) in order to make a predictive model. But since the current value decays gradually with time, there is no good reference or profile for me to use to compare it to. For example, if I run the machine for 5 minutes and get a baseline current of 6.20 mA, and I leave it running at the same settings for half an hour, the current could drop to around 5.85 mA. And if I stop the machine and come back 2 hours later to run it, I could see it running at 6.45 mA, so having a baseline or profile to reference to characterize the "standard" current behavior of the motor has not been able to work for me. Hence, why I went to the "rescan" approach and am trying to find alternatives.

Do you have data on the interaction of your device with the patient (for example, frequency change curves depending on the "efforts" made by the patient) (hmm... what "efforts" is?)? What range of musculoskeletal disorders are covered in your data?
Can you answer questions about how “constant” the patient’s influence on the device is (for example, if a patient has one leg damaged, will the “interaction” be the same as for a patient whose both legs are equally functional, etc.)?
I drew the curves in #44 to demonstrate how difficult it can be to identify useful information from the input data (look at the pulse duration curve and imagine what will happen if the patient's impact on the device is uneven - the curves will turn into "chaos").

If I am looking at the raw data of the sensor output live while using the machine, slight changes whether it is unilateral or bilateral work applied can be seen. If I do slight kicks against the footplates on one leg the "spikes" in current change are seen. And similarly, if I am applying low levels of relatively uniform work over a period of time and "walking with the device" you can see the wave form origin/mean of the current wave rise slightly. However due to the oscillation and overlap of values when there is and is not work being done, is hard to make a clear distinction.

 

[afm]

On another note, but also related to this as some people have brought up this point... I am starting the process of position detection for the footplates. To reiterate, the machine does not currently have a point of reference to use for the controls to know when the footplates are together, or at either extreme of a step. There is currently 1 inductive sensor paired with a metal encoder wheel with 8 spaces (equally sized on and off segments). We cannot guarantee starting position of the footplates, so we only know when 1 revolution (or 1 stride = 2 steps) has occurred.

Some of the option I'm considering are:
1. switching to an incremental encoder to utilize its z-signal, however this is a costly swap.
2. adding a second encoder wheel and inductive sensor on the other footplate linkage
3. adding a second inductive sensor to same encoder wheel and adding a second "ring" of segments.

#3 is what I have been ideating at the moment as much of the logic in the controls could remain the same. My current thought is that the outer ring will remain the same with the 8 segments, and I will add an inner ring with only 1 open segment to reference the location for when the 2 footplates are together (when InductiveSensor2==FALSE). From this point, (if I am only using one footplate as my reference), I can use the pulse counter from the InductiveSensor1 to say:
2 pulses= front extreme of step
4 pulses= footplate are together (1 step occurred)
6 pulses= back extreme of step
8 pulses= footplates are together (1 revolution = 2 steps occured)

Is there another simpler approach that I am missing, thoughts are appreciated, thanks!