I tried something new today. Specifically, I calculated and plotted the noise of a signal by referencing the phase shifted signal point to the same waveform from a non-shifted signal point. In the Plot Det E section above there is a spreadsheet that shows you the calculation. I did this with the hopes that it would help show the non-smooth character of each waveform. The result I got startled me and if my initial conclusion is correct delights me.
Specifically, you will see in the figure and Plot Set E that the noise, as I call it, of the dynamometer is very low whereas the noise on the system load cell is quite pronounced. Is it possible that it is showing us that all of the acceleration induced noise is on the system load cell and not on the dynamometer? It needs more investigation. However, if correct this might provide a relatively straightforward way for a user to check general noise level on the system load signal during cycling operation.
I performed the E467 calculations on the +/- 2500 lb, 100 kN frame, 0.25 in diameter bar at 5 Hz. With the peaks and valleys identified for both the load cell and dynamometer, the largest valley error is 6.77 lbf and the maximum peak error is 7.28 lbf. I understand that when you look at the entire plot of error there are expected acquisition errors away from the peak and valley, but I don't see that a 0.135% and a 0.146% maximum valley and peak error shows bad correlation between the data. I am looking for my baseline noise on the accelerometer file, but can't find it. I also realize that I noted all of the accelerometer units wrong. They are in Volts not mV.
Notice how the high frequency noise is much lower for this part compared to the two previous ones. I think this one was run on a different system. I wonder how the dither setting changed between those two systems.
Looking at the phase shift between the acceleration and the dynamometer based error, I wonder if there was a phase shift there that needs to be backed out. The two signals do seem to share a similar fundamental, but they’re out of phase slightly. Not sure yet how to check that out and correctly adjust the phasing.
Is phasing an issue that would need to be dealt with? The standard is concerned with dynamometer errors at the peak and valley and the accelerometer with the amplitude of the signal? Just trying to be sure we don't spend time digging down holes that are not of the greatest concern. I would expect them to be 90 degrees out of phase if the data collection was perfect.
Accelerometer phasing is very important since it defines how much inertial load is added to the peak load, and its polarity.
Why would you expect acceleration and load to be 90 degrees out of phase? It has been my experience that it can range from acceleration leading the load peak by 90 degrees to acceleration lagging the load peak by 90 degrees, or anywhere in between.
Yeah, I thought about that a little more and was not thinking of it correctly. If data acquisition was perfect than the maximum on the accelerometer would be at the peaks and valleys. I know my acquisition was wonky with what I had to pass through to get more than two signals together. I could next time take just each signal versus dynamometer data to see what the phase lag looks like there, I would need less electronics involved for that. But again, the phasing doesn't impact the calculations in the standard.
The standard only looks at amplitude of the accelerometer signal because in our initial implementation where it was just a quick check ignoring phase meant the person assumes the worst case for acceleration phasing. That is, they assumed that maximum acceleration errors occur at the peak and Valley.
Now that we are considering this method of accelerometer checking as a more common method I think phasing will become important. Maybe not but it certainly seems to me that now is the right time to look at the question and figure out what options we have.
This afternoon I looked at the same test as the one shown in the figures above. This time, however I phase shifted the dynamometer to align with the test machine load cell, and then took a difference to look at the error based on the dynamometer.
I have a simple graphic, which Shows how they correlate. Or more precisely how they don’t correlate. Tomorrow, I hope to put some text around it and add that to this document for us to discuss. This does seem to throw a major wrench in our works. Perhaps it’s a fluke!
Please look at a plot of the the error against a plot of the loading curve. I expect that at the peaks and valleys where the load rate attains a close to zero speed that the errors will be close to zero and when the curve crosses zero when the load rate is at its maximum we will see greater errors. I did plot this for the last item in the stack 100 kN, 0.25 in, 5 Hz, +/- 5 Hz and I see very much what I expect. The error is 90 degrees out of phase and when the loads are stable we see minimal error and when the load is moving fast we get larger errors. This data was collected at 1612 Hz, If I am doing my math correctly, as we cross zero we should see 48 lbs between points near zero. Since we know the sampling can't be perfectly synchronized, I see errors between the two near zero as an expected sampling vestige and not an issue with correlation.
I believe that is one of the reasons that this standard is only concerned with the errors at the peak and valley and not throughout the entire cycle.
I did as you suggested with some of the plots. Then I adjusted my phase shift protocol to be able to shift waveforms a fraction of one sample point. By making that change, I was able to make the error signals somewhat flatter. Also, it was easy now to note, as you suggested, the errors at the peaks of the Test way For remained relatively stable as I adjusted the phasing.
For a quick reference, I got a bunch of the plots into another Facebook post.
PLOT SET E
********
I tried something new today. Specifically, I calculated and plotted the noise of a signal by referencing the phase shifted signal point to the same waveform from a non-shifted signal point. In the Plot Det E section above there is a spreadsheet that shows you the calculation. I did this with the hopes that it would help show the non-smooth character of each waveform. The result I got startled me and if my initial conclusion is correct delights me.
Specifically, you will see in the figure and Plot Set E that the noise, as I call it, of the dynamometer is very low whereas the noise on the system load cell is quite pronounced. Is it possible that it is showing us that all of the acceleration induced noise is on the system load cell and not on the dynamometer? It needs more investigation. However, if correct this might provide a relatively straightforward way for a user to check general noise level on the system load signal during cycling operation.
I performed the E467 calculations on the +/- 2500 lb, 100 kN frame, 0.25 in diameter bar at 5 Hz. With the peaks and valleys identified for both the load cell and dynamometer, the largest valley error is 6.77 lbf and the maximum peak error is 7.28 lbf. I understand that when you look at the entire plot of error there are expected acquisition errors away from the peak and valley, but I don't see that a 0.135% and a 0.146% maximum valley and peak error shows bad correlation between the data. I am looking for my baseline noise on the accelerometer file, but can't find it. I also realize that I noted all of the accelerometer units wrong. They are in Volts not mV.
Increased load and stiffness:
Notice how the high frequency noise is much lower for this part compared to the two previous ones. I think this one was run on a different system. I wonder how the dither setting changed between those two systems.
Looking at the phase shift between the acceleration and the dynamometer based error, I wonder if there was a phase shift there that needs to be backed out. The two signals do seem to share a similar fundamental, but they’re out of phase slightly. Not sure yet how to check that out and correctly adjust the phasing.
Is phasing an issue that would need to be dealt with? The standard is concerned with dynamometer errors at the peak and valley and the accelerometer with the amplitude of the signal? Just trying to be sure we don't spend time digging down holes that are not of the greatest concern. I would expect them to be 90 degrees out of phase if the data collection was perfect.
Accelerometer phasing is very important since it defines how much inertial load is added to the peak load, and its polarity.
Why would you expect acceleration and load to be 90 degrees out of phase? It has been my experience that it can range from acceleration leading the load peak by 90 degrees to acceleration lagging the load peak by 90 degrees, or anywhere in between.
Yeah, I thought about that a little more and was not thinking of it correctly. If data acquisition was perfect than the maximum on the accelerometer would be at the peaks and valleys. I know my acquisition was wonky with what I had to pass through to get more than two signals together. I could next time take just each signal versus dynamometer data to see what the phase lag looks like there, I would need less electronics involved for that. But again, the phasing doesn't impact the calculations in the standard.
The standard only looks at amplitude of the accelerometer signal because in our initial implementation where it was just a quick check ignoring phase meant the person assumes the worst case for acceleration phasing. That is, they assumed that maximum acceleration errors occur at the peak and Valley.
Now that we are considering this method of accelerometer checking as a more common method I think phasing will become important. Maybe not but it certainly seems to me that now is the right time to look at the question and figure out what options we have.
This afternoon I looked at the same test as the one shown in the figures above. This time, however I phase shifted the dynamometer to align with the test machine load cell, and then took a difference to look at the error based on the dynamometer.
I have a simple graphic, which Shows how they correlate. Or more precisely how they don’t correlate. Tomorrow, I hope to put some text around it and add that to this document for us to discuss. This does seem to throw a major wrench in our works. Perhaps it’s a fluke!
Joe,
Please look at a plot of the the error against a plot of the loading curve. I expect that at the peaks and valleys where the load rate attains a close to zero speed that the errors will be close to zero and when the curve crosses zero when the load rate is at its maximum we will see greater errors. I did plot this for the last item in the stack 100 kN, 0.25 in, 5 Hz, +/- 5 Hz and I see very much what I expect. The error is 90 degrees out of phase and when the loads are stable we see minimal error and when the load is moving fast we get larger errors. This data was collected at 1612 Hz, If I am doing my math correctly, as we cross zero we should see 48 lbs between points near zero. Since we know the sampling can't be perfectly synchronized, I see errors between the two near zero as an expected sampling vestige and not an issue with correlation.
I believe that is one of the reasons that this standard is only concerned with the errors at the peak and valley and not throughout the entire cycle.
I did as you suggested with some of the plots. Then I adjusted my phase shift protocol to be able to shift waveforms a fraction of one sample point. By making that change, I was able to make the error signals somewhat flatter. Also, it was easy now to note, as you suggested, the errors at the peaks of the Test way For remained relatively stable as I adjusted the phasing.
For a quick reference, I got a bunch of the plots into another Facebook post.
https://www.facebook.com/groups/fjaa140856/permalink/675732981905161/
Facebook image annotations.
********
https://www.facebook.com/groups/fjaa140856/posts/674180435393749/