LION-AG Experiment

Hi can thanks a lot for these plots. I’m still not sure what to make of them. There is a lot going in there.

I’m curious about the radiation behavior. The significant and persistent drop in value after the device is hot in the active run. And recall you have recently been doing quite a lot investigation background radiation etc.

Do you have any ideas what could cause the change in the graph here? Are there outside causes of indirect causes from the experiment that could cause the observed change

I’m also curious if any one Has ideas about what is happening with the current values.

StephenC

So far the only way to consistently induce my Geiger counter a change in behavior is by adding/removing shielding or moving it into a different location (also within the same room).

I've had a periodic signal, but that would occur on a 24h basis. Until a few days ago I've ran the GC a few meters away from the testing equipment, however. I don't know what LION is using. Other counters might be more sensitive to temperature, EM interference (what's going on with power measurements there?), etc.

Alan Smith

A close-up view does indeed show that the on-off action is regular and not occurring at a very fast rate. Still, there's something strange with the actual values.

Quote from can

A close-up view does indeed show that the on-off action is regular and not occurring at a very fast rate. Still, there's something strange with the actual values.

Weird stuff. The downward transients in the current trace usually come once every 23 seconds. Each one is brief -- lasting 1 second or less -- and they appear to have no important effect on reactor temperature. Could the 23 second periodicity actually reflect a much higher PID duty cycle that is aliased down to a super low frequency by the 1 Hz acquisition rate?

can can I make a strange request?

Could you make a similar scale plot of the active run when there is no strange current effect.

Also one if the same scale during the calibration run?

I have a couple of reasons for asking the first normal the second a bit strange.

The first is it might be interesting to see if we see similar 23s in some smaller structure may be bumps in the plots elsewhere that might point to a sampling effect.

The second is... I was looking at the radiation curve above . But to my inexpert eyes there was a small what could be regular oscillation the broader variations. That could match that 23s frequency. I think maybe Each time we may have a little oscillation peak we have a spike down in current but it doesn’t look always true though. Now I appreciate it is noisy and eyes can play optical tricks when comparisons with under lying patterns are there. I guess it most likely that. But I thought it was curious.

Could the oscillations if really there be an artifact of the current? Or is there a sample rate effect impacting both sets of data? Or is this different detector some how being similarly effected?

It could be interesting to also see a zoom of the full data set over the time frame of just a few current spikes. It’s interesting that each spike does oud not just one value. Although not completely regular or identical they do seem to have something regular about them.

Another question do we know whether the various sensors are sending digital data using connectors close to the device or sending analog data that is digitized in a separate more distant processor? I suppose the current is measured at input though...

I think that can is using a library of timeseries analysis routines for his graphing and analysis. If they are up to the task, then the way to see if the transients in the current trace are reflected in the radiation trace is to use event-triggered averaging. This consists of using the current transients to trigger acquisition of, say, 10 second snippets of the radiation data. If you then average together all of these snippets, the random components will decrease as the square root of the number of acquisitions and any signal correlated with the transients will pop out.

I can accomplish this too, eventually. But I'm not sure that going down this particular rabbit hole is worth it. Can should also probably pass on this unless it is easy for him.

Hi Bruce__H I think it is also most likely an artifact or visual impression biased by the current graph.

If it looks similar outside the range or during the calibration I think there is not to much point going further. But if it is only visible at these times then could be worth considering.

It’s Interesting that there are mathematical tools to look for these kinds of correlations though.

@Stephen_C

Bruce__H

I'm not sure I understood what I should do with the snippets but check if the data attached corresponds to them.

aggregate-10s.zip

can thanks well I guess the behavior looks typical. Was just my eyes after all. Thanks for clarifying it though.

No small 23s glitches in the current either.

Quote from can

I'm not sure I understood what I should do with the snippets but check if the data attached corresponds to them.

Hi can.

Yes. You have acquired 10-second data segments each of which ends on a leading edge of one of the transients. What I would now do is average together these data segments such that you end up with a single 10 second trace. So, to take the current trace first ... if you have n snippets, the first point in the averaged current trace will be the average of n initial data points from the n snippets, the the second point in the averaged current trace will be the average of second data points from the n snippets, and so on. The random components in each snippet will cancel out in the averaging but the signal that you used to trigger the acquisitions will not, it will pop out more and more as you make n larger. Now use the same snippets but this time look at, say, the averaged radiation trace. If there is any signal in the radiation trace that is somehow locked to the transients in the current trace then you should see them pop out too. Random or uncorrelated components will be averaged away.

It is actually more usual to acquire snippets with the triggering stimulus in the middle. For instance, for the transient that occurred on Nov 11 at 0:00:13 you would acquire everything from 0:00:08 to 0:00:18. Also, with respect to these particular data where there are clearly different eras during which transients occur, I would just work on one particular era at a time and not average together information from different eras.

Bruce__H

Thanks for the detailed explanation; unfortunately it appears to be less straightforward to do than I initially assumed so I'll have to pass on this one given that it might also not lead to a significantly better understanding of the results.

Here is a plot of the current and ThermoA from a short portion of the run 3 raw data spreadsheet:

A couple of odd things. Most of the time the current oscillates around zero from +.19 to -.30 (Amps presumably). I assume there should never be negative current, so the zero point of the hall current sensor is not set correctly. When I have used that type of current sensor, I always had to have a calibration step to zero out the sensor. The zero can also drift with temperature and voltage.

The big spikes in current (over 1 A) make it look like the PID is not tuned correctly. It is acting like a bang-bang control system, or maybe the current level cannot be set to intermediate values.

The period when the current is off is interesting. It stays well above room temperature (32-35C) for over a minute with zero power input. There is much lower variance in temperature when the current is off. When the current is cycling, the variations do not exactly follow the changes in current, so something interesting may also be happening then .

Here is another plot from later. It is ramping to a higher temperature set point and the current stays on at 1 A.

Quote from Robert Horst

Here is a plot of the current and ThermoA from a short portion of the run 3 raw data spreadsheet:

The fine structure of the Current and Thermocouple A traces do seem correlated ...

In these plots I have not imputed any data values at points where the thermocouple A trace contained "Error" messages. So this may look a bit different from can's data. The current trace contains no such errors.

I note that the reduction in the variance of the thermocouple readings beginning at 7:32:10 is very sudden -- much quicker than I think would be allowed for by the thermal time constants of either the heating chamber or the thermocouple -- so I suspect that there is an extrinsic signal that is being impressed on both the current and thermocouple channels from somewhere. I am also puzzled by the double-line nature of the thermocouple readings at this time. This is not due to straightforward discretation errors because there are occasional intermediate values.

Quote from Robert Horst

Here is another plot from later. It is ramping to a higher temperature set point and the current stays on at 1 A.

I note that something like this happened on thermocouple B, but not thermocouple A, during one of the calibration runs (see file LOG00001). In this run there were no reactors in the heating chamber.

Quote from Bruce__H

I am also puzzled by the double-line nature of the thermocouple readings at this time. This is not due to straightforward discretation errors because there are occasional intermediate values.

The double line structure looks like an artifact of the way you plotted it (points without connecting lines). The temperature just has small variations around 33C. See my plot of the same region below.

The way I read the current line is to think of it as set points of the PID.

Negative = too hot, cut all current

Positive = too cool, add current to heat up

Zero = keep at this temperature by cycling current on and off

So it makes more sense if you think of the current plot as the commanded current, not measured current. I don't know if the spreadsheet really has commanded current, or if the zero point of the current sensor is wrong. That would be critical to know in interpreting the data.

Quote from Robert Horst

The double line structure looks like an artifact of the way you plotted it (points without connecting lines). The temperature just has small variations around 33C. See my plot of the same region below.

I don't think I am looking at an artefact of plotting. You and I seem to be dealing with datasets that are actually different. For instance, in the 53 seconds between 7:52:73 and 7:53:36 my datasheet only shows 24 valid ThermoA data values (the rest show as "Error"). In your plot, on the other hand, there are 53 separate values. This leads me to think that your plot is showing data where interpolated values have been imputed for about half of the data points.

Quote from Robert Horst

The way I read the current line is to think of it as set points of the PID.

Negative = too hot, cut all current

Positive = too cool, add current to heat up

Zero = keep at this temperature by cycling current on and off

So it makes more sense if you think of the current plot as the commanded current, not measured current. I don't know if the spreadsheet really has commanded current, or if the zero point of the current sensor is wrong. That would be critical to know in interpreting the data.

I mostly agree. i think that LION has attempted not to use the PID controller at all here. Instead he has set out to control the heating coil current using the rheostat on the PCM in the setup. I don't know how he has programmed the PID parameters -- whatever settings he has used may have left some unintentional residual activity which is what we are seeing. I have asked LION.