Energy meter with fast cycling load

[Skogsgurra]

I have on my bench a kWh meter that is supposed to have a systematic (all meters of that type seem to have the problem) error.

Calibration works out fine, but when you cycle the load with a few seconds on and a few seconds off, the meter registers more energy than is actually used by the load.

The reason this deviation was detected is that a guy with a (I think) badly designed thermostat in an around 2 kW application thought he paid too much for electricity. Utility did a check and found that the meter actually registered too much energy. Not when testing with steady load, but when checking with parallel meter from other manufacturer.

I have some theories about this. But feel genuinly unsure. My first thought is that some accumulating or filtering software is asymmetrical so that 'attack' is faster than 'decay' - if I am allowed to use musical language.

Anyone has any experience with this? Heard about a similar thing? I am open for all possibilities - the weirder the better (well, within limits).

Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[electricpete]

There is one small error - as written, the code does not see a "1" if it is located in the very first row of the data, and the search is upwards. To correct this:
"thisrow = myrange.Rows.Count - rowcounter"
needs to be changed to
"thisrow = myrange.Rows.Count - rowcounter - 1"

Attached is a spreadsheet I used to test the corrected function. Sorry for the error.


=====================================
Eng-tips forums: The best place on the web for engineering discussions.

 

[electricpete]

I was thinking about that 1-2 second delay that is apparent. It seems unusual... surely there is no intentional delay. Makes me thing maybe there is a microprocessor busy taking care of some housekeeping details required to shift from monitoring mode to pulse transmission mode, handling communications, switching back to monitoring mode and somewhere along the line reset the integrator=accumulator back to zero watts. One wouldn't think that would take a long time, but there has got to be some reason for that delay, most likely during the communication part.

So let's say during that busy period after reaching to limit trigger, the microprocessor is in fact busy... so busy that it cannot continue to monitor the incoming power to update the totals.

One presumes that the unit tries to compensate by adding an amount of energy based on elapsed time (I presume there is a clock) as well as last known power level. That would be an accurate compensation provided there is not a transition (off-to-on or on-to-off) occuring during the critical period while last-know power is used as a substitue for actual power.

At first glance, that would seem like a random error, since the probability of hitting off-to-on is same as hitting on-to-off, isn't it? No, actually not. An off-to-on transition can only occur when the heater was initially off, and an on-to-off transition can only occur when the heater was initially on. Considering the duty cycle and difference in power levels, the majority of pulses are sent (or at least the accumulator limits reached) during the on-state. So if we monitor for 1 day, we might have 1000 pulses sent from on-state and 100 pulses sent from off-state (just numbers for talking purposes). Since duration of off state and on-state are roughly the same, the probability of hitting a transition during that time interval is the same for both types.... let's say 10%. So 10% of the 1000 pulses while on experience an on-to-off transition during the critical time period and 10% of the 100 pulses while off experience an off-to-on transition during the critical time period. That is a lot more on-to-off type errors during a critical transition period than off-to-on type errors during a critical transition period.... ends up overcharging.

Or, maybe not ;-)


=====================================
Eng-tips forums: The best place on the web for engineering discussions.

 

[Skogsgurra]

Very good. That is something I am beginning to think also. I have done tests with very fast cycling and the apparent energy/pulse is dropping from nominal 3600 Ws/pulse to around 2400 Ws/pulse when cycling with 500 ms period and 20 % duty cycle. Resulting in a 50 % extra charge for the energy.

I admit that this is an abuse of the meters. But the intent of this exercise is to provoke bad behaviour - and I think we have just that. The customer's load also cycles, but with a lower rate and over months and years. Customer says that he thinks he has payed around 15 % too much, but data from parallel meter says that it could also be more.

I attach Excel sheets (sorry, I am in a hurry so I will use your code when not so pressed for time). I have reduced row heights so you can only see the first three - four rows, and after that only the rows where pulses are output.

Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[Skogsgurra]

Should be 'paid', I think?

Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[electricpete]

At 300 watts continuous, you would have 1 pulse/12 seconds.
At 2900 watts continuous, you would have 1 pulse/1.24 seconds.

It says 500 sec cycle on top, but below it says 2hz. I assume it is 500 millisecond cycle (?).
Than for 50% duty cycle, you have only .25 sec in each state.....so you have many heater on-off cycles between each pulse.

If I stick with the “theory” that I had suggested above, it would require that every single one of the pulse “triggers” (when integrator reaches the limit) occurred during on states (otherwise if pulse trigger occurred during offstate, we would have had a high watt-second reading).

For 50% duty cycle that is plausible. There is 2900/300 ~ 10x as much probability that each pulse trigger is reached during on state compared to off state.

For 20% duty cycle, it is not as much. 4x as much time in off state as on state. Only 2.5x as much probability that each pulse trigger is reached during on state compared to off state. I would have thought at least one of those 7 pulses during the 20% duty cycle would have shown a total which was at or above the 3600. So it does not seem very consistent with my theory from that standpoint. Given that there is some unknown delay in the neighborhood 1-2 seconds, there is no hope of going back to the data to figure out the condition when the trigger occurred.

Another thing to note: the average power is 1600 watts in 50% duty cycle and 820 watts in 20% duty cycle. If the trigger occurred during on state, the difference between “assumed” (under my theory) and average power would be 2900-1600 = 1300 watts at 50% duty cycle or 2900-820= 2080 watts at 20% duty cycle. Assuming the time for which the error occurs is the same, the ratio of errors for the 20 to 50 duty cycle should be 2080 to 1300... the actual ratio would be higher.

It would be interesting to see results if the cycle time were much slower such that you get let’s say average of two or three pulse triggers in an on period and one pulse trigger in an off period. Maybe 3 seconds on, 12 seconds off. Then if timing of pulse relative to transition (or to state at time of pulse) is important, you should get some normal pulses (3600 watt seconds) and some pulses with lower watt seconds, and perhaps pulses with higher higher watts seconds. Focusing on the timing characteristics which are unique to the abnormal pulses and what preceded them might help define the mechanism better. Just a thought.

Of course that experiment probably means data collection over longer period of time which could become difficult to manage. And whether it’s worth the effort depends in part on judgement of the likelihood of uncovering useful data. And I’m not sure if your goal is to recreate the behavior and that is all that is needed, or to how much additional effort would be warranted to understand the mechanism.


=====================================
Eng-tips forums: The best place on the web for engineering discussions.

 

[Skogsgurra]

Working on it. I am rechecking all calibration now.

I can handle 60 second's worth of data. And there is no problem recreating a longer cycle - like the one we saw at the customer's place. Stand by! I am still confused. Want to make 100 % sure that this isn't a red herring.

Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[Skogsgurra]

Yes, of course, 500 milliseconds.

I have checked the data acquisition and accumulation in my measuring device. It delivers very consistent data when I run without switching. All pulses are then worth between 3498 and 3505 Ws. Instantaneous power is in line with expectations and also with an electrodynamic wattmeter with class 0.2, so I feel confident about the measurements. What I think could be a problem is if the sampling interval isn't correctly specified. Then, the accumulated energy could be off with a few percent - which it perhaps is. Or, since the utility guys say that the meters can have errors up to 5 percent (which is hard to believe), it could also be that I am spot on - or close.

Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[waross]

Congratulations on the Tip Master award, E-Pete.
Yours
Bill

Bill
--------------------
"Why not the best?"
Jimmy Carter

 

[Skogsgurra]

That's 'just in time!' Well deserved.

Sorry to such a bore. I have checked lots of combinations and this one is definitely sick. What the heck is going on?

Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[electricpete]

Thanks. I am in good company.

I think stretching out the duration a little more may make the pattern clearer.

=====================================
Eng-tips forums: The best place on the web for engineering discussions.

 

[Skogsgurra]

I made a 'macro' recording with all voltages and currents removed and over 60 seconds. Duty cycle is still 50 % but period is 10 seconds. The delayed reaction is clearly visible and I think that the meter is favoring short pulse duration, also when level is low.

Actually, that is what we also saw on customer's site and what we have noticed in other recordings.

Do you know anything useful about sampling theory? Does this have anything with the retired railway man and the trains, that Bill mentioned in his 26 Feb 10 2:22 post, to do?

BTW, I am not so sure about the exact power levels in previous recordings. But, in this one, the levels are 180 W and 2700 W.

Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[Skogsgurra]

Finally! I see what the problem is. When those meters (I now have three of them) are feeding a load that is dimmed (using thyristors), I get very high watt-hours. The excerpt from the report may be of interest.


Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[electricpete]

So, the most recent data is similar to the earlier data, except that now the cycling time is very fast (dimming), correct?

What is the new insight that it gives?

=====================================
Eng-tips forums: The best place on the web for engineering discussions.

 

[Skogsgurra]

I have no real insight. But I think (I may have to eat this) that the algorithm somehow makes a mistake. It seems that the manufacturer looks for the negative zero crossing and then assumes that conduction angle always is 180 degrees.

That strategy works in most circumstances but not when conduction angle is small. We will probably never know. The whole thing is now subject to discussions on a national level. Big money involved. And probably lawsuits.

Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[electricpete]

OK, I understand now... the finding that it overcharges for dimmers helps define the scope of the problem (how many customers might be affected), but not necessarily the mechanism.

=====================================
Eng-tips forums: The best place on the web for engineering discussions.

 

[electricpete]

... although it's another clue in the puzzle.

=====================================
Eng-tips forums: The best place on the web for engineering discussions.

 

[itsmoked]

Clearly the meters aren't doing what I'd expect for a revenue grade watt-meter. That's measuring the instantaneous current and voltage and multiplying them either continually or a thousand times a second. Instead they're using some Mickey Mouse zero crossing detection multiplied by assumptions.

No-doubt they got the bid to supply the bazillion meters because their POS was much cheaper than meters that actually work.

Nice work Gunnar.

Keith Cress
kcress -

http://www.flaminsystems.com

 

[stevenal]

I don't know how the Euro standards deal with this, but ANSI tests all the way to + or - 2% rated frequency. I'm sure most manufacturers do better, but until there are some appropriate standards we cannot expect very much.