Error bounds for Mizuno R19 results

Reference

Jed's view on calorimetry errors:

Apparently you are conflating two separate conditions: high input power, and low anomalous excess heat.

High input power makes almost no difference to the calorimetry. It does not interfere in the measurements. It is easy to measure. Indeed, it is the easiest physical force to measure.

Low excess heat is difficult to measure. The error margin is large.

Thus:

300 W input, 330 W output, 30 W excess is very easy to measure. The 300 W of input power produce very little noise. The 10:1 ratio makes no different to the measurements.

1 W input, 3 W output, 2 W excess would be difficult to measure with this instrument. The low input power does not help. The ratio of 1:3 does not help.

I have a lot of actual data showing this is the case.

Jed's assertion that the Mizuno calorimeter error bound is +/- 2W

I calculated it from the spreadsheet, and posted a graph showing the source of my calculation. The noise is about 0.1 deg C which translates into 2 W. Take it or leave it. But you are evading the issue. "People more generally" means nothing. The top tier of cold fusion papers covers this in great detail. Despite that, you and Shanahan claim that every single study in this field is invalid. You have no reason. You point to Mizuno to show that Storms and McKubre are wrong. That's pathological skepticism.

This thread. Consider what we know about the R19 Mizuno calorimeter and results. How can we work out an error bound for those results? Is is 2W as Jed thinks?

Why bother? Anyone getting smaller excess heat indications needs to know whether those are definitely excess heat, or whether they could be something else. Convincing others is also easier with a complete error bound analysis (commonly done in experimental work). Use R19 as an example of how to do this.

So Jed's post above introduces the first question to answer. What are error bounds?

Jed is detailing the (good) reasons he has for bounding the noise and repeatability of the measurements. Obviously, short-term noise (electrical, sensor-based, in this case perhaps turbulence-based) limits resolution and therefore also accuracy. Also, repeatability (related to differences between characteristics on different cycles, possibly caused by environmental influences, or system change on heating/cooling) limits accuracy, but not resolution. In fact short-term (high frequency) noise can be smoothed and potentially the low fequency accuracy (what we actually want, usually) can be a good deal better than the single measurement resolution.

Jed also points out that the resolution of the thermocouples (0.01C) limits accuracy. He points out however that thermocouples are accurate, and given this 0.01C uncertainty we can extrapolate the uncertainty in power measurement.

That is true.

Taking Jed's first principles data which I'm happy to deal with first, noting only that there may be additional constraints on output, with lower bounds, that can be got from calibration with control data.

Pout = Tdelta * SHC * MF

Tdelta - temperature difference - Jed has detailed the sort of temperature accuracy we can expect above. Let us call the variability in Tdelta +/- Tx, with the measured value Tdelta'.

SHC - specific heat capacity. Not quite a constant because it changes with temperature and pressure, but these are small changes. Being lazy, we can represent the possible change a variability +/- SHCx on a middle value of SHC'

MF - massflow (air mass per second) this has a variability as well, more complex to estimate. it will be mainly multiplicative, e.g. MF'(1 +/- Me). (MF' = measurement value)

Finally to determine the excess heat bound we have to know the input power Pin. If this is measured by standard equipment, power meter or voltmeter and ammeter, it will have accuracy determined by the equipment. Normally this is also multiplicative:

Pin = Pin'(1 +/- Pe) (Pin' = measured value)

The range of possible values of excess power is then:

(Tdelta' - Tx)*(SHC' - Sx)*MF'(1-Me) -Pin'*Pe < Pxs < (Tdelta' +Tx)*(SHC' + Sx)*MF'(1 + Me) + Pin'*Pe

Caveats:

• Before you jump on this and say that must of these error factors are so small they can be ignored - the point of doing this algebraically first is so we can get some feeling for how the error changes with parameters.
• Before you jump on this and say that control calibration can reduce these other errors - yes, I agree, assuming we can bound the difference between control and active system efficiency. We always have two separate calculations we can do here, first principles absolute, and relative to control. They have different analysis. But, Jed's 2W would require control and active systems to have very very close efficiency. We have no data that tells us this.

We can immediately say, taking Jed's example, that the error measuring 30W excess on 300W input will be higher than the error measuring 30W excess on 1W input.

In this error bound the Tx part (which Jed quantified above) stays the same. The other errors are all multiplied by Tdelta' which is proportional to (Pin' + Pxs) and therefore 10X higher for 300W input than for 1W input.

The error in measuring first principles excess heat is (mostly) proportional to the input power, so COP is a good measurement of how likely measurement will be larger than excess.

Of course there are also additive errors; if we were measuring a few 100mW here the additive Tx would dominate and COP would no longer be the relevant criteria.

We don't yet know how large are Sx, Me, Pe.

But, I've now done what I said I'd do quickly in this post. Because most of the error terms in the error bound analysis are multiplicative on power, a large fixed input power increases errors.

To go deeper into the R19 error analysis it would be very helpful for somone to answer this post. It makes a 25% difference in the R19 COP data which way it is answered (e.g. COP=1.5 -> COP=1.25 or COP = 1.875). I'm not trying to be difficult here, and I've probably missed somewhere it clearly indicates what is given as Pout in Table 1.

Dealing just (for now) with the absolute, first principles, data, we can see that the measured COP from the calorimeter is very constant with temperature, increasing slightly at higher temperatures. the maximum value (ignoring outliers) is 1.35, at a power in of 200W, corresponding to 70W excess.

The negative error bound on this measured Pout (rearranging the equation above) =

270*(1-Tx/Tdelta')*(1-Sx/SHC')*(1-Me/MF') -200 - 200*Pe = 70 -200*(Pe/Pin') - 270*(Me/MF' + Sx/SHC' + Tx/TDelta' + second order terms ....)

each of the error terms here is now arranged as a percentage of the corresponding quantity. We expect each to be small, which means that although the complete lower bound must include second order terms as well, these should be quite a bit smaller than the 1st order terms here. Omitting the second order and higher terms means that are lower error bound for excess power is a little bit higher than it should really be.

THH

Quote from Ascoli65

Yes, why bother?

You obviously are entitled to propose any thread you wish, but let me say that this specific topic is quite surreal. The "Error bounds for Mizuno R19 results" – as well as for many other famous and sensational CF/LENR results - don't depend upon the accuracy of the instrumentation or the adequacy of the experimental set-up only, they mostly depend upon the reliability of the reported data. In other words, the upper bound is determined by the human factor.

For example, what is the upper bound of an error caused by entering a wrong value for the shunt resistance in a data log system (*)? A factor of 1.01, 10, … 1000? How much an inaccuracy of 2 W (or even more) in the measuring of the output power weighs upon a basic error caused by a human error which affects the input power for a 2 or 6 factor?

A scientific experiment includes not only the equipment (specimens, set-up, instrumentation, …) but also the experimenters and, in some cases, separate reporters. An evaluation of the plausibility of the claimed results - especially when they are so sensational as those in the LENR field - should start with an assessment of the reliability of the numerical data and other basic information reported by the human components of the experiment, otherwise you risk to produced a huge quantity of words, as well as thousands of papers, based on nothing.

(*) Mizuno reports increased excess heat

Display More

Ascoli - as is often the case i agree with half of your point but not the other.

True, Mizuno's results depend on the human factor, and from evidence (that you unearthed) we know that here to be flakey.

Nevertheless proper accounting for errors is worth doing, and LENR experiments tend not to do this. They should.

For results to be solid you need both no mistakes (the human factor) and also experimental errors < measured results.

One additional correction we need to make to the table 1 data (reproduced above) is due to the blower power.

The blower is operated at 6.5 W. The outlet air temperature is measured with two RTDs. They are installed in the

center of the pipe, one in the stream of air before it reaches the blower, and one after it, to measure any heat added to the
stream of air by the blower motor. The difference between the two is less than 0.1◦C. However, there are indications
that heat from the blower motor is affecting both of them. A calibration with no input power to the reactors shows that
when the blower power is stepped from 1.5 to 5 W, the outlet RTDs are ∼0.35◦C warmer than inlet (Fig. 8). This is a
much larger temperature difference than the moving air in the box alone could produce.
Blowers are inefficient, so most of the input power to the blower converts to waste heat in the motor. It seems likely

some of this heat is conducted by the pipe to raise the temperature of the two outlet RTDs

In addition, Figure 8, which shows the output RTD temperature increase for different blower powers has a curve fitted

over 0W - 5W which is not very accurate, so we cannot simply extrapolate this to the actual 6.5W.

A reasonable bound to the output RTD uplift from this effect would be 0.4C.

However, the data here is troubling. The careful measurement before and after the blower shows that this cannot be direct

heating of air from the blower (as might be expected) because that would affect the after RTD but not the before one. The only

other heating mechanism is via the RTD mounting. The blower heats up, that heat is conducted to the RTD mount, and affects the RTD itself.

Jed states that output power is calculated as 16.9W/K from the calorimeter.

To check this we have power calculated as per Jed's equation: 1.006 kJ/kgK * 0.00342m^2 * 5.4m/s * 1.184kg/m^3

Note that although SHCp (1.006 kJ/kgK) does not depend significantly on temperature, density (1.184kg/m^3 @ 25C and atmospheric pressure) does,

so this must be included, later on, in the error bound, because hotter air is less dense.

From the above, power is calculated at 22W/K. I'm not sure what is the discrepancy here. The power will decrease by 0.5% per degree K above 25C. At say 40C outlet

we would have 20.5W/K. My guess is that the 16.9W/K figure comes from a different air velocity.

Using the correct 22W/K, and the reasonable 0.4C uplift in temperature due to blower heat conduction direct to output RTD, we get an estimated 8.8W uplift in

measured power over that calculated from the RTD temperature difference. Correcting for this we get a corrected COP of 1.32.

It is possible that the fan correction has been done in the Table 1 data? I think we cannot continue with this analysis till we know whether the Table 1 output powers include a fan temperature uplift correction or not.

The fan temperature correction is troubling because we know that it is needed, but we do not know how it changes as the reactor temperature goes up.

The known correction from "not understood fan heating of RTD" is 8.8W or 12% of the corrected excess at 0 power. However Mizuno, according to Jed, calculates this as something lower: around 6.5W.

If the whole calorimeter heats up during a higher power run this could increase this correction, we do not know.

Quote from Ascoli65

I also agree partly with you: both terms are important, but they have to be treated in the correct order. When you have a signal influenced by two noises of very different magnitudes, you must necessarily remove the loudest in order to identify and evaluate the weakest.

In the case of the "dramatic" Mizuno's results brought to our attention by JR in June (1), the most urgent issue to clarify is: can you trust all those data? Which of those data are reliable?

In this respect, I have already asked JR to explain why the 2 spreadsheets of the 120 W tests carried out in May 2016 show different quantities in the "Input power" column (2). He didn't answer to my question, but in the meanwhile he rebutted any remark on secondary aspects of the Mizuno's calorimetry up to exhausting his most resolute interlocutors (3). Why he didn't answer my simple question?

Everyone on L-F, in particular the "resident skeptics", knows that he is very skilled in the art of rhetoric. The language factor prevents me from competing with him at the same level. What I'm proposing to you, and to other L-F members interested in the Mizuno's results, is to help me to urge JR to explain why the 2 spreadsheets are different, before dealing with other less priority, although equally important, issues.

(1) Mizuno reports increased excess heat

(2) Mizuno reports increased excess heat

(3) Mizuno Airflow Calorimetry

Display More

I do agree that your concern about this 2017 result is valid. Basically, the calibration and active run datasheets, as i understand this, use a different method to calculate input power. One uses direct measurement from a power meter. The other uses V * I as calculated from a data logger and i measured via the voltage across a shunt resistor.

There is a lot of possibility for mistake in this. Perhaps it indicates that calibrations are used from some significant time before the active runs, and that introduces uncertainty about what other equipment or setup or environmental condition changes exist between the two things. However, those 2016 results are not very large in any case, and all I take from this is that Mizuno is lax in his methodology and therefore the exact measurements he makes, or even perhaps equipment he uses, cannot be trusted to be what we expect. That is a big caveat on all his results, I agree.

THH

https://docs.google.com/spread…EMipY/edit#gid=1856684673 (Mizuno 2017 calorimeter results)

Looking at the two output air temperature RTD measurements in this spreadsheet, they vary between 0.35C and 0C difference. It is not clear what controls this variation. Therefore it seems that there is something not understood about the output temperature measurement.

THH

Quote from Ascoli65

This is not my concern, because it is excluded that two different methods were used to measure the input power.

The two spreadsheets come from the same data system, which included (1):

- the power input analyzer (Yokogawa, PZ 4000),

- the data logger (Agilent, 34970A),

- the PC for data acquisition;

- and the HP A/D converter (2) (maybe the previous PC or a separate instrument).

The active and control tests were performed in two consecutive days (May 19 and 20, 2016) by using the same instrumentation. Therefore, the methods of measuring the electrical parameters, as well as the information recorded by the data logger and all subsequent calculations, were identical. So the two spreadsheet should have both reported the input power measured by the Yokogawa analyzer, which is presented as a reliable 16,000 $ instrument. But, the data from this last instrument appear only on the spreadsheet of the control test, not in that of the active test which was run the day before!

What happened to those data? When was they removed from the "excess heat" spreadsheet? And where (Japan or US)? And Why?

These are the questions that need to be answered, before continuing the discussion on other less critical issues.

Those 2016 data were the most sensational results presented at the ICCF21 (3). For the most powerful test, an excess heat of almost 250 W was claimed, with a COP close to 2. These results look even better than the last ones reported for the R19 device (4).

That's what I say. It makes no sense to discuss about the (instrument) error bounds, if we can't trust the basic data.

(1) https://www.lenr-canr.org/acrobat/MizunoTpreprintob.pdf

(2) Mizuno reports increased excess heat

(3) https://www.lenr-canr.org/acrobat/MizunoTexcessheat.pdf

(4) Mizuno Airflow Calorimetry

Display More

OK - well i agree that is more difficult to explain but i still hold for mistake rather than deliberate malfeasance. Jed perhaps can work it out.

Quote from Wyttenbach

The logic is wrong: Electro motors have a high efficiency thus they transport most of the energy into the air movement (T depends on average speed too) . This is the reason why you have to add! the energy % of the efficiency to the airflow and not the other way round.

But the blower speed is constant! As long as we assume the inlet air has the same density the change will be marginal. Thus only calibration must be done with a constant delta (= blower power) subtracted!

To go deeper we would need he blower spec! With 30% excess energy a total relative error of 5% in the calorimeter is no problem.

Wyttenbach - you are outside you realm of expertise here. I agree the blower efficiency would need to be checked, but small motors tend to be not very efficient.

p442 https://sda.sanyodenki.us/data/cooling/catalog/Blower.pdf

on the load graph taking the point of max efficiency (and the low back pressure used here may not meet it) we have:

7.2W (rated power)

0.4 m^3/min

180Pa

hence air power out < 0.4*180 / 60 = 1.2W

The actual power in at this point will maybe be less than the rated power (it depends on load) but not much less.

Hence, for this blower, most of the electrical power goes into heat. I have not shown this is motor heat not turbulence in the air, but that is true. The motor may get cooled by the moving air but according to the statement in the paper this does not much happen.

Why does this change with reactor temperature? Blower power stays the same but if as claimed in the paper the heat in conveyed direct to the RTD via a mounting bracket, if the whole casing including blower heats up this increment might increase. Or, it might decrease. we do not know.

Jed - perhaps you can think of some reason for the change in power measurement over consecutive days, between calibration and active runs.

My suggestion would be that both calibration and active reactors can be operated independently, and they might therefore use independent power inputs both always connected to the logger?

It does decrease the integrity of the control - you want if possible to keep everything the same.

Yes that typo has already been noted