Current sensor: accuracy vs error

[red_alert]

Current sensor: accuracy vs reading error

Hello,

I need to replace an in-circuit current sensor with a non-invasive one.

I came to this LEM product (HTFS 200-P) but I have doubts regarding the actual reading error.

In the datasheet, there is a mention about accuracy at Ipn (1% of Ipn, where Ipn is the rated nominal current - 200A in this case). That is, the accuracy is 2A when reading 200A.

But what is the reading error for a smaller current? What will be the error when measuring a current of 10A?

On a different product datasheet (Honeywell CSNF6661), the accuracy is relative (0.5%). That means the reading error is 1A when measuring 200A and 5mA when measuring 1A?

Thanks in advance for any clue.

 

[schmitt trigger]

On the datasheet, do you see the 12 different parameters that are listed after "Accuracy-Dynamic Performance Data"?

Long story short...you have to take all of those parameter into account. Some contributions may be negligible in your application, but at least you have to consider them.

But if you want a first-order estimate of your accuracy, in addition to the parameter you mention, at least take into account the zero current offset voltage.

 

[D.A.(Tony)Stewart]

Re: Current sensor: accuracy vs reading error

Hello,

I need to replace an in-circuit current sensor with a non-invasive one.

I came to this LEM product (HTFS 200-P) but I have doubts regarding the actual reading error.

In the datasheet, there is a mention about accuracy at Ipn (1% of Ipn, where Ipn is the rated nominal current - 200A in this case). That is, the accuracy is 2A when reading 200A.

But what is the reading error for a smaller current? What will be the error when measuring a current of 10A?

On a different product datasheet (Honeywell CSNF6661), the accuracy is relative (0.5%). That means the reading error is 1A when measuring 200A and 5mA when measuring 1A?

Thanks in advance for any clue.

1st spec
X Accuracy 2) @ I PN , T A = 25°C ≤ ± 1 % of I PN ( 1% of full scale)
ε L Linearity error (0 .. 1.5 x I PN ) ≤ ± 0.5 % of I PN (0.5% of full scale)
(footnote) Excluding offset and Magnetic offset voltage.
V OM Magnetic offset voltage @ I P = 0,
after an overload of 3 x I PN DC < ± 0.5 % of I PN

2nd spec
Accuracy ± 0.5 % ( of full scale)
Offset Current < ± 0.2 mA
Offset Current Drift < ± 0.5 mA


You are best to consult with the OEM supplier for more details but errors from Offset , Gain, Drift and overdrive all affect accuracy and the minimum current to operate at. Some are open loop, closed loop and chopper types with extra windings to get even better accuracy due to asymmetry and saturation but cost more.

Consider these too. https://www.lem.com/index.php?optio...erie&serie=HLSR Series&limit=30&limitstart=30

You may have to perform a periodic null adjustment if you need better accuracy or a degaussing if overcurrent has occured.

 

[red_alert]

I've seen all those temperature/linearity/magnetic coeficients but it's still unclear if that 1% it's an absolute or relative error.

That is, if I measure a 10A current using that sensor (200A, 1% full scale) what will be the reading error: 0.1A (1% of 10A) or 2A (1% of 200A)?

Btw, on the second datasheet there's no mention about "full scale" but only "0.5%" accuracy. Could it be 0.5% of the actually measured current?

Could someone recommend a good non-invasive current sensor (DC) in 150A-200A range? I need to monitor my off-grid energy consumption, solar/wind generated power and the battery charging current.

I need a good accuracy (<1% of the actual value) mainly for lower currents (0 - 50A) but I need the sensor to withstands the higher current peaks (200A).

 

[FvM]

I've seen all those temperature/linearity/magnetic coeficients but it's still unclear if that 1% it's an absolute or relative error.

1% @IPN means a scale factor error, in other words relative.
linearity error 0.5% of IPN is an absolute error, as well as the 0.5 % magnetic offset

HTFS sensors have also considerable noise which matters in wide band applications.

 

[red_alert]

HTFS sensors have also considerable noise which matters in wide band applications.

Thanks a lot for the explanations. Could you recommend another LEM/Honeywell product series? I do prefer a single (5V) power supply operation but I've seen that all high-end sensors need +-12(15)V power supplies.

 

[D.A.(Tony)Stewart]

Linearity refers to gain errors.
Offset is absolute of full scale and depends on current peak. Over current is worst case.

Accuracy at small levels is not specified.

 

[FvM]

Linearity refers to gain errors.

I would explain it slightly different. A linear characteristic can be represented by a straight line, linearity error is deviation of the actual characteristic from this line, either a curvature or discontinuities, e.g. steps.

There are different ways to specify linearity errors, one possible way is to define a +/- error band around the ideal line. With data converters, this specification is well-known as INL (integral non-linearity), either given as fraction of full scale range or related to the LSB resolution.

The HTFS dataheet also specifies an INL quantity, as percentage of full scale range.

HTFS sensors are typical for what you can expect from a direct current sensor. For obvious reasons, all 5V supplied sensors are using this principle. Higher accuracy, e.g. <= 0.1 % linearity usually requires a compensation sensor along with a high voltage supply, e.g. +/- 12V.

 

[red_alert]

Thanks for your input! To summarize the above comments, what's the actual reading error (including linearity) of this HTFS sensor (and similar)?

Is it around 1% (or 2-3-5%, including those linearity/temperature coefficients) of the actual value? If I want to measure a 10A current, the sensor would return anything between 9.9A and 10.1A (+/-1% of 10A) or anything between 8A and 12A (+/-1% of 200A)?

 

[FvM]

Following the specification, total error could be up to +/- 2A. (0.5 % linearity, 0.5 % offset, each of 200 A). Plus 1 % of actual reading scaling factor error.

In practice, there won't be a noticeable linearity error at small currents, I would also expect that the system is initially zero adjusted, so offset drift of maximal +/-25 mA/K (0.3 mV/K/2.5V*200A) and hysteresis after possible overload are the dominant errors.

 

[red_alert]

Thanks, now I really get it. I think I'll order some higher accuracy (and price) current sensors as I have much more expensive equipments relying on their readings.

Moreover, I need to make further operations with different sensor reading values (add/substract) and if the Murphy's laws are still in effect, those errors will sum up in the worst direction.

 

[red_alert]

I have another question regarding the accuracy/reading error.

For a specific measured current, is the reading error the same or is it a random value within linearity error range?

I wonder if making multiple sensor readings (100 measurements in a 100us burst, by example) then calculating the average result could lead to a better accuracy.

Also, could it be a good idea to measure the current using 2-3 identical sensors then averaging the results?

I need a better accuracy at lower currents (0 - 10A) but I have to measure currents up to 100 - 150A (DC). Single supply (5V) open loop sensors are more suitable for my application but their accuracy is not the best.

 

[FvM]

The discussed error terms (scaling error, nonlinearity, offset) can't be reduced by averaging multiple readings. This is only the case for sensor noise.

Some applications allow auto-zero, probably the most effective way to achieve accurate low range measurements.

 

[red_alert]

Could you elaborate about "application-dependant auto-zero", please? It's something related to auto-zero amplifiers? Does it have to be implemented at sensor level (built-in)?

In my application, I could only provide a zero current calibration. Beside, I could monitor the sensor Vref to check it against uC internal ADC reference then adjust the readings accordingly.

Anyway, I'm going to use two identical sensors for measurement redundancy thus I could average the readings during normal operations.

 

[KlausST]

Hi,

I used the HTFS-200 many times.

I find they are relatively good. You can trust the datasheet. I´ve seen they are better (at room temperature) than specified.

As mentioned before, an overcurrent causes an offfset error. (caused by remanence in the ferrite core)
We used two strategiesto cancel it out.
1) on a device where we neeed to calculate RMS only of AC, we used a DC blocking (high pass) filter.
2) on a device where DC should be measured: We did a dynamic DC correction as long as the load is switched OFF (known state of zero current). When device is activated we freeze the DC correction value.
A new DC correction process is also possile to be initialized by a PLC.

Both work without problems and give very reliable and precise results for a chemical plant.

Klaus

 

[red_alert]

Many thanks, Klaus! HTFS-200 was on my short list, too (btw, it has a convenient mounting design) thus I've just placed an order.

I've read about some demagnetisation procedures in the datasheet (using a special pattern recovery current). I hope there won't be the case (too often) though.

 

[red_alert]

I'm trying to calibrate some current sensors (**broken link removed** but I don't have any other calibrated device (ampmeter).

I just have an Arduino-like microcontroller board which has some 12-bit ADC (3.3V internal reference).

According to datasheet, I have the following relations:

(1) Vzero = Vref +/- 0.025V (Vzero = output voltage for zero input current)

(2) Vout = Vref + 1.25 * Ip/Ipn

My strategy is to measure both Vout and Vref then, after substracting Vref from Vout I could get the Ip, the measured ("primary") current. Ipn is the nominal current (200A in this particular case).

The problem is the two relations don't match for Ip = 0. Using the second relation I got Vout(zero) = Vref while from the first relation Vout(zero) = Vref +/-25mV.

How to deal with this situation? Do I have to measure the Vout at Ip = 0 then use this value as Vref in relation (3)?

As you can see, the output is increasing with 6.25mV for every 1A of measured current but the zero current drift (+/-25mV) it's quite large (4A).

 

[FvM]

+/- 25 mV is not drift, it's initial (not calibrated) offset error.

I don't understand your calculation problems. Actual zero is Vout with zero primary current.

 

[red_alert]

+/- 25 mV is not drift, it's initial (not calibrated) offset error.

So do I need to measure this offset error at zero primary current then I could substract it from any further measurements? It has a constant value for the entire measurement range? May I use the following relation:

Ip = (Vout - Vref - Voffset)*200/1.25 ?

 

[FvM]

So do I need to measure this offset error at zero primary current then I could substract it from any further measurements? It has a constant value for the entire measurement range?

That's the nature of an offset quantity.

Can be simplified to Ip = (Vout - Vzero)*200/1.25