Acceleration Data Analysis Guidance for MSc

[Khym France]

Hi I'm currently on the data analysis section of my MSc and I'm after some guidance if possible please. Basically I've managed to figure out many variables from force plate data including displacement/net force/velocity/power etc however I'm trying to do the same with data collected from a wearable that's just gives me acceleration...so I have time and acceleration magnitude from 3 axes but when I input the same formulas into excel to get the other variables I seem to be going incredibly wrong. I basically need to compare the data from the force plate to that of the wearable. If someone wouldn't mind assisting to start me off that would be great as I'm running out of time and I have over 1450 reps to analyse...any help would be greatly appreciated

 

[berkeman]

Hi I'm currently on the data analysis section of my MSc and I'm after some guidance if possible please. Basically I've managed to figure out many variables from force plate data including displacement/net force/velocity/power etc however I'm trying to do the same with data collected from a wearable that's just gives me acceleration...so I have time and acceleration magnitude from 3 axes but when I input the same formulas into excel to get the other variables I seem to be going incredibly wrong. I basically need to compare the data from the force plate to that of the wearable. If someone wouldn't mind assisting to start me off that would be great as I'm running out of time and I have over 1450 reps to analyse...any help would be greatly appreciated

Welcome to the PF.

Can you post more information on the test setups with diagrams, and say how the data seem not to be agreeing? Can you post an example of the disagreement?

 

[Khym France]

Welcome to the PF.

Can you post more information on the test setups with diagrams, and say how the data seem not to be agreeing? Can you post an example of the disagreement?

thank you for your response ill take some screen shots and send them to you now if that's ok...the first ones will be of the the force plate break down which I think is correct as I'm getting jump height and peak power etc which looks about right...it's the second tab I'm struggling with to calculate all the variables from acceleration...you'll notice in this one I'm not calculating peak power or jump height at all...both are one rep of a squat taken simultaneously. Thank you so much I've wasted about 3 weeks now going round in circles.

 

[Khym France]

Here is my spreadsheet, I'm currently working on one rep of the squat and counter movement jump so you should see 4 tabs:
Squat fp think this is correct
Squat w needs work as these are the calc I'm trying to figure out from acceleration magnitude alone
CMJ fp think this is correct
CMJ w as above

I'm hoping the acceleration from the wearable is similar to that of the FP

Many thanks...

 

[Khym France]

Thank you for any guidance x

 

[JBA]

Without any of your equations it is not possible for anyone to analyze your procedure.

 

[Khym France]

Without any of your equations it is not possible for anyone to analyze your procedure.

Hi I attached my spreadsheet with all the equations within. Here you can see how I've worked out every variable and like before I trust the force plate data is correct it's the wearable acceleration data that I'm struggling to covert

 

[JBA]

OK, I see the actual workbook now.In trying to evaluate the second sheet, one thing I find that disturbs me is that there is a very large variance in the weights when calculated based upon the imported acceleration values. See the below screen shot column N of the few examples I calculated.

 

[Khym France]

OK, I see the actual workbook now.In trying to evaluate the second sheet, one thing I find that disturbs me is that there is a very large variance in the weights when calculated based upon the imported acceleration values. See the below screen shot column N of the few examples I calculated.

View attachment 195406

Thank you so much, can I ask what you mean by the weights? The force plate was measuring at 1000hz and the wearable at 62hz so when I did the body mass calculations and any that required time intervals I used the body mass calculated from the force plate for the wearable data and I changed the time intervals from I believe 0.001 to 0.006. I need to normalise all the data yet to body weight but would do that at the end once I've calculated all 59 subjects reps...hope that makes sense Thank you so much for your help.

 

[JBA]

It was difficult for me to judge how much accelerometer output "noise" existed during what should have been a relatively static period leading to to the actual squat cycle. (A constant 1.000 value representing the acceleration the accelerometer experiences due strictly to gravity). As a result, I decided to try converting those outputs and your reference mass to calculate an equivalent weight for each accelerometer value. Where was the wearable accelerometer mounted during the test? If it was mounted somewhere on your body (like your arm) that could move independently of the actual motion of the squat then that could explain why there is a lot of variations in my reference weight calculation values.

There is an additional issues I suggest you consider:
Based upon the above, it appears to me that, as opposed to your method of calculating your impulse based upon the total accelerometer output, your actual input acceleration value for your squat should be based upon the difference between the accelerometer value and the static gravity value of 1.000 that the accelerometer experiences due strictly to gravity even when you are not moving at all.

 

[JBA]

After an additional bit of review, another general issue related to your calculations of velocity for both sheets is that you use a*t to calculate the total velocity at the end of each time period when I believe the correct formulas for the final velocity and total distance traveled at the end of each time segment n should be: v@n = v@(n-1) + t@n*(a@n - g) and then d total @n = 1/2*v@n*t@n + d@(n-1)

Edit: Followup on the velocity calculation, it works OK on the short term period of a cycle but over a sustained time period there is a bit drift in the accelerometer's readings that cause an accumulated 0 point error in a long term plot that I cannot figure out how eliminate other than by trucating the curve and resetting the reference "1" g value.

 

[Khym France]

Hi and thank you very much for your time invested looking at this. The wearable was mounted on the wrist and during the squat the hands were placed on the shoulders crossed and during the CMJ they were placed on the hips. There will be some obscure data where movement occurred before and after the main exercise but my intentions are to only extract the data using the force time curve during the upward and concentric parts of the movement.

 

[JBA]

That makes sense. On the equivalence of the accelerometer and force plate readings the best way to deal with that is to use the Fz N for both by converting the accelerometer values to an equivalent effective weight using Fz N = g reading x Weight (N) rather than Fz N = g reading x Mass. Once you do that, and focus on the 6 to 8 second period of the actual squat then you will see that the resulting accelerometer curve generally matches the profile of your force plate curve but for some reason exhibits a variance in the squat duration time with the time intervals of 0.001 sec for the force plate and 0.006 sec for the accelerometer you specified and; unfortunately has an increased range of Fz N values as seen in the below graphs copies
Note: The start time for the second graph has been normalized to match the start point time of the first graph

 

[Khym France]

Hey thanks so much I'm taking a look at this now and for the next few hours so I'll see what I can do. Would it be ok to convert the accelerometer values though would this not skew my results?

 

[JBA]

Just for reference, I due to some encouragement from my Engineering Mgr at the time (1985) starting using a spreadsheet called "Visicalc" for engineering analysis and test data reduction and have written literally 100's for such programs with Excel the remaining period until now so I have developed a number of techniques for normalizing, manipulating and filtering data during that period.

As a result, I have been taking some time to see what I could do with this in Excel (by modifications to a copy of your Excel workbook) just for my own enjoyment and on both the force plate and the accelerometer recordings. On the accelerometer recording, one technique I have used to smooth out some of the noise in the squat data curve is to first calculate and F N using: F N = accel x (your reference wt) and then create a successive F N set of values by doing a running 5 pt (two values below through two values above) averaging on each original F N value . You can use the same technique directly on your G values if you prefer.

Next, since only the > 1 G values are important during the squat for data you want I then used an F net = if( F N >= (your ref wt), F N - (your ref weight) ,0) statement to eliminate all of the negative F net values. (PS You can also use the if statement on your force plate and since that curve is must cleaner, that process automatically isolates exactly those positive force values you need for the further analysis.)

Below are the resulting successive plots using those procedures.

I am happy to give you some assitance with all of this but there is going to b delay until Sunday April 30 before I can do that because I am going to be offline from starting tomorrow morning for all of the days until that date.

 

[Khym France]

Wow this is amazing...what I'm actually trying to do is compose the template in order that I can review 3 reps of a squat and 3 reps of a counter movement jump per subject on 2 consecutive testing days so it's clearly a reliability and validity study.

My intentions are to drop all the raw data into this template and extract only the data from the concentric part of the movement which is mostly the upward phases.

I'm looking to see if the wearable is consistent to that of the force plate. I therefore need to set up the sheet in order to extract this data and then after each Rep of analysis I'll move to my calculations sheet (first tab) then onto my master workbook in a separate file.
I'm looking at peak accelerations and both leak and average power hence why I'm getting frustrated that my force plate readings and all the variables look correct but my wearable ones don't...I had a look at it earlier and all the formulas to me make sense because if I have acceleration and mass I should be able to calculate the force which I've done and then the net force and all the other variables by doing exactly the same but it's not worked...any help is greatly appreciated so thank you so much for your time. I'm in talks with my tutors too and I also need to normalise the data to Mass so need to set up a tab somewhere in order to do this also...but I'm assuming I would do this after all the tests on the data I've extracted per subject so for example sub 1 squat FP peak acceleration is...CMJ FP peak acceleration is ...and is compare across subjects per variable so peak acceleration, peak power and average power...?

 

[JBA]

As stated, for the force plate, I am able to isolate the force curve for any single squat thrust using the method I described and therefore the data you desire just as you apparently can do; unfortunately, when it comes to the wearable there is considerable pre-thrust acceleration greater than G = 1 noise, so I have not figured out so far any method that will allow excel to automatically isolate the desired thrust region.

One issue that concerns me that I have wanted to ask you about is that on the two relative plots is the difference in the apparent time interval for the squat thrust. For the force plate record the time for the positive F net is approximately .96 sec, but for the wearable positive acceleration it is only approximately 0.4 seconds. Do you have any explanation for this time period differential in the two records?

 

[JBA]

Just for your information below are the successive F net, Velocity and Displacement curves from my Force Plate analysis.

 

[Khym France]

Hi further to this and after several days I've realized that the accelerometer was measuring in units of g therefore after now extracting all of the data my acceleration readings having now converted this is m/s2 are looking pretty positive. I do however have another issue. I'm trying to calculate peak power which I've completed by the initial process of using the accel mag in g. Calculated force by x weight then net force by - bw then velocity from the accel x time which was 0.016 and finally power by force x velocity. I have figures way off as clearing I've used it all in units of g and not accel m/s2. Can I just convert my peak power reading in g to watts? or do i need to start all over again? Any guidance again would be greatfully received.

 

[JBA]

As I see it the problem getting the power from your available data from the wearable is that it gives you only max force and time @ max force but not the distance or velocity required to use either of the following equations:

P = (g max) x (body w) x (distance @ g max) / (time @ g max) or P = (g max) x (body w) x (velocity @ g max); and, I am having a problem seeing how your above described calculation extracts from the wearable's g max and time @ g max data the distance for the first equation or the velocity for the second one.

Can you clarify that for me?

 

[Khym France]

Hi further to this and after several days I've realized that the accelerometer was measuring in units of g therefore after now extracting all of the data my acceleration readings having now converted this is m/s2 are looking pretty positive. I do however have another issue. I'm trying to calculate peak power which I've completed by the initial process of using the accel mag in g. Calculated force by x weight then net force by - bw then velocity from the accel x time which was 0.016 and finally power by force x velocity. I have figures way off as clearing I've used it all in units of g and not accel m/s2. Can I just convert my peak power reading in g to watts? or do i need to start all over again? Any guidance again would be greatfully received.

As I see it the problem getting the power from your available data from the wearable is that it gives you only max force and time @ max force but not the distance or velocity required to use either of the following equations:

P = (g max) x (body w) x (distance @ g max) / (time @ g max) or P = (g max) x (body w) x (velocity @ g max); and, I am having a problem seeing how your above described calculation extracts from the wearable's g max and time @ g max data the distance for the first equation or the velocity for the second one.

Can you clarify that for me?

So I'd use accel x mass to get force but clearly I've done all this using g and not m/s2, is then use force - mass to get net force and then to calculate velocity it's accel x time interval which was 0.016 etc I'm thinking I now need to start all over again and convert the g to acceleration from the start and then add all the formulas to calculate the rest!

 

[JBA]

I now understand how you are getting the velocity if you are using the g net for the velocity calculation As far as the time period, that depends on whether you want the "peak power" for which you would only use the time at the peak g level; or, you want the total power over the g =1 to g max lifting segment of the squat. If you want the the total power, then a good approximation, depending upon the curve linearity during that period, using the average g value and the total time for that section might give you a good approximation. Just a suggestion.