emonTx accuracy test
The test detailed below performed on one emontx comes to the following conclusion:
The error below 40W could not be measured reliably and below 100W it was worse than 10%. Above 100W it was better than 10%, above 150W better than 6%, above 250W better than 4% and above 500W better than 2%.
It may be that your emontx will surpass or fail to achieve these accuracy levels depending on:
1) Calibration - as demonstrated in this test, it may be necessary to apply an additional calibration, the precision of the different components in the circuit means that the voltage calibration can be anywhere between 203.1 – 265.4 while the current sensor calibration can be anywhere between 109.44 – 112.78 see CT and AC power adapter installation and calibration theory
2) Bias position will have a large effect on accuracy at low powers see: http://openenergymonitor.org/emon/buildingblocks/measurement-implications-of-adc-resolution-at-low-current-values
3) Level of electro-magnetic interference in your environment. Some laptops and switch mode power supplies can be a large source of electro-magnetic interference. Whether your circuit is shielded in any way i.e inside a metal box. A certain level of noise is actually a good thing as it helps overcome the limitations of trying to measure power where the current waveform is close to the ADC level magnitude, by causing the levels to flip higher and lower which after averaging gives a more accurate result.
If you could also repeat this test and post up your results, it would be great to get some comparative data of different emontx's in real world use.
If your also a non-emontx user but have build the emontx circuit on a breadboard, stripboard or arduino shield protoboard it would also be interesting to compare results.
The test setup consists of an emontx with the standard Mascot AC-AC voltage adapter for voltage measurement and the standard current sensor, a YHDC sct-013-000 CT. The emonTx is powered via this usb power supply. The EmonTx sends the realpower values wirelessly to a jeelink that is connected to the computer.
The normal emonTx_CT123_Voltage sketch was used, the measurement rate was increased from once every 5 seconds to twice a second.
The CT is clipped around a modified cable that goes to a fixed load (a 20W and 60W incandescent lightbulb was used in the test). The cable modification was to separate the live and neutral wires and getting enough cable length to wrap around the CT up to 14 times in the case of this test.
A 20W and 60W incandescent lightbulb was used to ensure that the current waveform would be a near-sinusoidal waveform. It is expected that a low energy lightbulb may show higher accuracy at lower powers as its non-linear current draw, a current draw that looks more like a spike at the peak of the mains voltage waveform will utilise more ADC levels, see Measurement implications of ADC resolution at low current values.
The test was performed in two parts: first 1-14 turns with a 20W lightbulb and second 1-14 turns with a 60W lightbulb. Care was taken each time to make sure that the CT was clipped firmly, that the contact between the split cores where good and that there was no observable effect in trying to push the cores closer together - indicating good contact.
The measurements were logged to a text file and opened and graphed in realtime using KST, an open source data viewing program. KST was set to a data range of 20 datapoints and a label was created to show the Min, Max, Mean and Standard deviation (Sigma) of the power data.
20 datapoints at 2 datapoints a second covers a time window of 10 seconds for each Min, Max, Mean and Standard deviation measurement. Care was taken to make sure that the mains voltage did not vary significantly during the sample duration.
Min: the minimum power value in the sample of 20 datapoints.
Max: the maximum power value in the sample of 20 datapoints.
Mean: the mean value of the 20 datapoints.
Standard deviation: the standard deviation of the 20 datapoints
These values where logged alongside the actual power as given by the plug meter multiplied by the number of turns in a spreadsheet.
These are the raw results, its clear that there is a significant calibration error as the error appears to increase in a particular direction linearly:
The plug in meter column on the left shows the number of watts as measured on the plug in meter: 21W relates to the actual power consumption of the 20W incandescent lightbulb, 60W relates to the actual power consumption of the 60W lightbulb.
The number of turn column is the number of turns of the live or neutral wire in the CT. The number of turns multiplied by the plug in meter power gives the power (or actually current) level as seen by the CT.
|plug in meter||no of turns||CT power||Max (A)||Min (B)||Mean||Standard Deviation||error (A)||error (B)||% error (A)||% error (B)||MAX % error|
Applying a calibration
The calibration factor was calculated by dividing the plug in power 854W by the mean emontx power of 801W.
Calibration factor: 1.066
The next question is whether this error is down to voltage measurement only or is also down to current measurement or some other source. Comparing the voltage measured on the plug meter: 251.2V and the emontx 242.5 it looks like the difference here accounts for just over half the error.
Applying the calibration:
|plug in meter||no of turns||CT power||Max||Min||Max CAL (A)||Min CAL (B)||Mean||
|% error (A)||% error (B)||MAX % error|
Plot of Maximum error at power
Removing the first and largest error value so that we can better see the other datapoints:
cagabi has repeated the tests and his results after calibration:
- For measurements below 50W: error >34%
- For measurements 50W - 100W: error >5%
- For measurements 100W - 240 W: 2% < error < 5%
- For measurements above the 300W: error < 1% (more less)
The results in a graph: