Measuring pulse times

[bhowell]

Hello,

The data I am going though is velocity and looks like:

__/\___/\___/\___/\___/\___/\___/\___/\___/\___/\___/\___/\___/\___/\__

Every spike is a piston moving forward (technically there are down spikes between each of these too, but I was worried about ascii formatting so I left them out). The time from 0 - 0 velocity should be the length of time it takes the piston to move forward.

I am hoping to find a way to measure the length of time of each spike and then take the average. So some kind of formula that would look a 5% change from nominal then when it's within 5% of nominal again (to account for noise in our data).

Essentially this is to determine the length of time it takes to complete motion.

Thanks for the help!

[martindwilson]

how about you specify the formula then we try to apply it to excel?

[bhowell]

I'm not sure I follow you. The formula is what I am looking for here.

I would like to measure the time from the start of the pulse to the end of the pulse, then take the average of those times for all the pulses in my data set.

[ChemistB]

You gave us a visual representation. Do you have some actual numeric data that we can look at?

[bhowell]

What I did was position measurements of a piston that I then derived into speed. The file I am working with is over 100mB, we were sampling it at 2000Hz for about a 5 minute period. I don't see a feasible way to get the data here, I'm also worried my company might see it as a leak of information if I did (IT tends to watch everything, especially large uploads and are quite cautious)

The measurements though are generally zero until the piston moves (with some slight "static" from the sensor noise, this is why I suggested waiting until it's moved beyond a certain value, say 1mm/s out of its max: 100mm/s). When it does move it accelerates from 0 to its max value, at halfway it reaches its peak, then decelerates down to zero.

Essentially a formula that would look for the speed to reach 1mm/s, then find when it crosses past 1mm/s on its slowdown. Then to calculate the time (x-axis) it took to do so.

This happens 25-100 times within our sample data and I want to calculate the average of all of those time.

The data again will look something like:

---(zero)---(spike up to 100mm/s /\)---(zero)---(spike down to -100mm/s \/)---

Repeated throughout the data.

Let me know if you need any more clarification.

Thanks!

[shg]

I think a matched filter is the way to do this.

Identify (manually) the signature of a dozen or so of each type of event. Average and normalize the values to create a filter kernel for each, and then use a formula to convolve the kernel against the data. Set a threshold at a value that captures most or all events with few or no false positives.

An example is attached that finds the spikes in an EKG.