MIZUNO REPLICATION AND MATERIALS ONLY

Quote from j9381

look at the gray line (no correction for heat loss) in the graph and see that it is higher than the red line (output, corrected for heat loss, I assume) for the first 30 minutes of the 5 hour run. At one point in that first 30 minutes, the uncorrected value (gray line) is about 130 watts while the corrected value (red line) is 50 watts.

Ah, I see what you mean. The red line goes up at the same slope for a while, then it suddenly drops, then recovers. It looks like it drops around minute 20 and recovers to rejoin the red line about 10 minutes later.

I don't know what that could be. My guess is that ambient temperature changes caused it. Things like that cause the biggest problems at low power levels. Interference from something like rapidly changing temperatures might last for 10 minutes. I guess it has to be some factor in the spreadsheet not used to compute the "no correction" curve. I guess the perturbation is introduced by the adjustment for heat losses from the insulated box, which of course are not included in the gray line. I don't recall seeing a spreadsheet with "no correction" so I don't know what factors are included in it. For the red line, the equation for heat adjusted for losses from the box is:

W_out/W = (W_n)/(5.76 × exp(−(K_n)/1512) − 4.78 + 0.00314 × (K_n)) [Kn is the reactor body temperature]

The equation for the temperature is pretty complicated:

(T_{out −} T_in)/deg = ((I_n) + (J_n)) × 0.5 − (H_N) − (−0.302 × exp (− (O_n)/1.829)) − 0.376+ (M_n)

Maybe there is a different version of the temperature calculation for the gray line?

Quote from JedRothwell

For the red line, the equation for heat adjusted for losses from the box is:

W_out/W = (W_n)/(5.76 × exp(−(K_n)/1512) − 4.78 + 0.00314 × (K_n)) [Kn is the reactor body temperature]

The equation for the temperature is pretty complicated:

(T_{out −} T_in)/deg = ((I_n) + (J_n)) × 0.5 − (H_N) − (−0.302 × exp (− (O_n)/1.829)) − 0.376+ (M_n)

Maybe there is a different version of the temperature calculation for the gray line?

Display More

My expectation is that the gray line is W_n and is calculated from T_out - T_in. I then further expect that the red line is W_out and is calculated from Wn as per the equation you list.

My problem now is this ... for the particular expression for W_out used here, the denominator is less than 1 at all working temperatures. So how does W_out (red line) ever end up smaller than W_n (grey line) if it is nothing but W_n divided by a number less than 1?

Something is wrong and it has nothing to do with temperature changes (because the denominator is less than 1 for all temperatures you use) .

It could be something like the data smoothing. A different rolling average applied to the two different curves. The gray one looks pretty smooth to me. Sm-o-o-o-th.

There is always a reason for things like this, but it is seldom an interesting reason. Or significant. It is prosaic.

Quote from JedRothwell

It could be something like the data smoothing. A different rolling average applied to the two different curves. The gray one looks pretty smooth to me. Sm-o-o-o-th.

There is always a reason for things like this, but it is seldom an interesting reason. Or significant. It is prosaic.

I think the answer to that is maybe ... but maybe not. It could be something serious too. And besides, if the grey trace is more heavily smoothed due to averaging then why does it rise faster than the red trace?

By the way, the heat recovery rate (the denominator for W_out) in use here is an oddball. Not like your Figure 2 of the first 2019 paper.

Quote from Bruce__H

And besides, if the grey trace is more heavily smoothed due to averaging then why does it rise faster than the red trace?

I do not understand what you mean by that. It will rise faster if the smoothing papers over some outlier data points that pulled down the red line. It is like the smooth 7-day rolling average lines we see every day in COVID-19, which leap right over the weekend deficits. (https://www.ajc.com/news/coron…d/jvoLBozRtBSVSNQDDAuZxH/) There are ~120 data points in the 10 minutes this happens. Not many. One or two odd ones will pull down the red line if it is not smoothed as much as the gray line is. Stray points could just be computer data errors, or hardware errors. Unless he has new computers, Mizuno still uses equipment is from the 1980s with floppy disks. I have seen some outlier points that I assumed must be noise from the thermocouples being jogged or something like that. Not every data point represents a real event.

Mind you, I am just speculating about the smoothing. Stuff like that happens. That's all I mean to say.

Quote from Alan Smith

a few words from the cook are more useful than many words from the dish-washer..

A few more words from the cook would be very helpful to understand why.

Quote from JedRothwell

Stuff like that happens. That's all I mean to say.

I agree completely. But it could also be some sort of coding error that has more far-reaching implications.

There does seem to be a stray point in the red trace. If this point is also originally in the grey trace and has been smoothed away then it could be pulling up the trace in that region. Since the crossover point is at about the 20 minute mark, the rolling average would have to employ a window at least 10 minutes wide (5 minutes on each side) to produce such an effect. The problem with this scenario is that windowed averaging is equivalent to a low-pass filter and the corner of the grey trace where the input is shut off looks very sharp .... it is nothing like a signal that has been averaged with a window that large.

Quote from Bruce__H

I don't understand the question.

I was in a bit of a rush this morning.
What I basically meant is an adjustable single factor that matches input power to the steady state part of the power trace, but I can see that might be trickier when it comes to excess heat cases.

It seems to me that a calorimeter loss correction factor formula should be based on the outlet air temperature rather than the reactor temperature. (If one is to be used at all). The temperature of the calorimeter inside air should have a more direct effect on calorimeter losses than surface temperature point(s) on the heat source. If the reactor and control are externally identical, and the temperature measurements have been done many times and always at the same place, I suppose it is almost the same thing. But if the active reactor gets a hot spot near the surface measurement point, then the correction factor based on reactor temperature may be too high. The outlet air temperature Is also independent of the reactor/control configuration (at steady state), so reactor design changes wouldn’t require extensive re-training of the outside surface to loss correction formula.

Quote from Bruce__H

The correction factor is 100% calculated from steady-state data. Whatever this problem is, it has nothing to do with the reactor warming up.

Yes, but the warm up period is not steady state at all, so a correction to the warm up period based on steady state data is probably not applicable.

A 800 W calibration compared to the 800 W excess, now that could be close.

Quote from Bruce__H

I don't think so. Look at Figure 10 of Mizuno and Rothwell J Cond Matt Nucl Sci 29:1-12 (2019), then equation 2 of the same paper. The equation they fit to their calibration data is fractional power capture = O/I = 0.98 - [5.0811E-4 x T] where T is "the reactor temperature"

水野：

このときは炉温度で補正式を作りました。同じく校正データーも同じ形状、重さの炉で補正式を作りましたが、精度が悪く、正確ではなかった。そのために箱からの熱逃散を補正に使った。その結果、極めて精度が良くなった。

私が出すデーターは新しいものほど精度、正確性が良い。さらにどのような形の、重さの炉であっても補正を入れた一般式で表すことが出来るようになった。

Mizuno:

Before 2017, the correction formula was made with the reactor temperature. Similarly, for the calibration data, a correction formula was made with a reactor of the same shape and weight, but the accuracy was poor and it was not accurate. Therefore, the heat dissipation from the box was used for the correction. As a result, the accuracy became extremely good.

The newer the data I give, the better the accuracy and precision. Furthermore, it became possible to express any type and weight of furnace by a general formula with correction.

Quote from mizunotadahiko

水野：

このときは炉温度で補正式を作りました。同じく校正データーも同じ形状、重さの炉で補正式を作りましたが、精度が悪く、正確ではなかった。そのために箱からの熱逃散を補正に使った。その結果、極めて精度が良くなった。

私が出すデーターは新しいものほど精度、正確性が良い。さらにどのような形の、重さの炉であっても補正を入れた一般式で表すことが出来るようになった。

Mizuno:

Before 2017, the correction formula was made with the reactor temperature. Similarly, for the calibration data, a correction formula was made with a reactor of the same shape and weight, but the accuracy was poor and it was not accurate. Therefore, the heat dissipation from the box was used for the correction. As a result, the accuracy became extremely good.

The newer the data I give, the better the accuracy and precision. Furthermore, it became possible to express any type and weight of furnace by a general formula with correction.

Display More

Thank you very much. That was an excellent answer.

Quote from Ascoli65

A few more words from the cook would be very helpful to understand why.

水野：良いところに気付きましたね。出力計算は平均値を使っていたので、元ファイルを添付します。

いつでも校正後の値が大きいです。

Mizuno: You've noticed some good points. Since the output calculation used the average value, attach the original file.

The value after calibration is always large.

Quote from Paradigmnoia

Yes, but the warm up period is not steady state at all, so a correction to the warm up period based on steady state data is probably not applicable.

I am surprised that multiple people are failing not to grasp the point I made about the crossing lines in Mizuno's 800W plot. It doesn't matter how far from steady state the system is, if you take an output power time series and correct it by multiplying by a factor that is always greater than 1, the result should be bigger. It shouldn't be sometimes smaller and sometimes bigger, it should just be bigger.

This has nothing to do with transients or complex physical phenomena. I mean, just draw out any squiggly line as a mock up of the uncorrected output (make it as complex as you want!) ... that is your grey trace. Now "correct" it by multiplying, pointwise, by some function that is always greater than 1. The result will be your red line. Those two lines will not cross. They will not cross. That is the sum total of the argument here

Paradigmnoia

I completely agree with your desire to see an 800 Watt calibration. It's a different issue though.

Quote from mizunotadahiko

Mizuno:

Before 2017, the correction formula was made with the reactor temperature. Similarly, for the calibration data, a correction formula was made with a reactor of the same shape and weight, but the accuracy was poor and it was not accurate. Therefore, the heat dissipation from the box was used for the correction. As a result, the accuracy became extremely good.

The newer the data I give, the better the accuracy and precision. Furthermore, it became possible to express any type and weight of furnace by a general formula with correction.

When Mizuno says "Therefore, the heat dissipation from the box was used for the correction. " Does he mean there are thermocouples (or similar) on the outside of the box that are being used to estimate the heat emitted by the box? Or is he using the temperature of the flowing air (from the air flow calorimetry) to estimate the heat emitted by the walls of the box to the environment. Either way would work but the former would be more accurate (and be a more complex experimental set up).

Quote from mizunotadahiko

Before 2017, the correction formula was made with the reactor temperature. Similarly, for the calibration data, a correction formula was made with a reactor of the same shape and weight, but the accuracy was poor and it was not accurate. Therefore, the heat dissipation from the box was used for the correction. As a result, the accuracy became extremely good.

The newer the data I give, the better the accuracy and precision. Furthermore, it became possible to express any type and weight of furnace by a general formula with correction.

Paradigmnoia

I don't understand Dr Mizuno's answer. Can you explain your understanding?

Quote from j9381

When Mizuno says "Therefore, the heat dissipation from the box was used for the correction. " Does he mean there are thermocouples (or similar) on the outside of the box that are being used to estimate the heat emitted by the box? Or is he using the temperature of the flowing air (from the air flow calorimetry) to estimate the heat emitted by the walls of the box to the environment. Either way would work but the former would be more accurate (and be a more complex experimental set up).

I am confused. On the one hand your suggestions make sense. On the other hand all the data in the first 2019 Mizuno and Rothwell paper are for experiments conducted in 2017 and 2018, and in that paper they clearly stay that the reactor temperature is used for calibration.

Dr Mizuno's recent clarifications fail to address the issue of the crossing grey and red lines in one of his plots. I am glad to see, however, the arrival of new data that do not have this flaw.

Quote from Bruce__H

I am surprised that multiple people are failing not to grasp the point I made about the crossing lines in Mizuno's 800W plot. It doesn't matter how far from steady state the system is, if you take an output power time series and correct it by multiplying by a factor that is always greater than 1, the result should be bigger. It shouldn't be sometimes smaller and sometimes bigger, it should just be bigger.

This has nothing to do with transients or complex physical phenomena. I mean, just draw out any squiggly line as a mock up of the uncorrected output (make it as complex as you want!) ... that is your grey trace. Now "correct" it by multiplying, pointwise, by some function that is always greater than 1. The result will be your red line. Those two lines will not cross. They will not cross. That is the sum total of the argument here

Paradigmnoia

I completely agree with your desire to see an 800 Watt calibration. It's a different issue though.

Display More

I know what you mean. The only answer is that the temperature-related correction formula, or the data it depends on can produce a result less than 1 of the grey line value. The reactor temperature is not plotted. If both the grey power trace and red power trace were calculated independently from each other, from the recorded data, rather than the red line plotted as a direct function of the grey line, they might not match step for step.

An easy way to cause a big momentary drop in the delta T, which would propagate to the power calculation, is to touch the inlet thermocouple, raising its temperature, reducing the delta T until the thermocouple returns to the air temperature. During start up, something might have been not quite right, then quickly adjusted, creating a bounce in the recorded data which might have been smoothed over for presentation, but not the same way for both sets of data. Unless one is calculating the entire experiment energy in-out, these sorts of things make no difference to the overall results.

Just a guess.