Level sensor accuracy problem

[Lancie1]

Attached Images

I think that your formula is too complicated.

It can be simplified if you divide your horizontal cylinder into two separate problems: (1)Level = 1/2 tank radius or less, and (2)Level Greater than 1 radius. These two levels can be easily computed separately at the proper times by using comparison limits.

(1)Level = 1 radius or less (half full or less):
Volume = Area x Length, where Area = R^2[cos^-1(R-h/R)], cos^-1= Inverse Cosine, R = constant Tank Radius, h = measured height of liquid.

(2)Level = Greater than 1 radius (more than half full):
Volume = Area x Length, where Area = R^2[PI-cos^-1(h-R/R)].

Using this formula, you will see that the max error (as liquid is added or removed from tank, waves are produced) will occur at liquid height = tank radius +/- 10%. You could apply your filter at that level only.

A programmed filter routine simply averages reading over some time period, or averages a New Reading with some sum of Old Readings. This will help to reduce errors caused by actual liquid level fluctuations (waves in tank), errors in level sensor, and electrical noise.

 

[(8{)} ( .)]

1st of all 10 bit resolution is pretty low.
2nd of all a good SCADA package will allow you to define a dead-band for analog points.

 

[kallileo]

@shooter
As I mentioned above the tank is a inclined cylinder so the aforementioned formula should be used to calculate the table.
I have attached an excerpt from the table of the 3000l tank.

@mickey
I did some tests in excel and it seems that the simple filter you propose could be a good solution.

 

[Peter Nachtwey]

Since when does a analog device have resolution? A higher resolution analog input card is what is needed. The MDT rod will be good but it normally a MDT rod can easily get 5 micron resolution if using SSI. The problem is not with the rod, it is the analog input card.

I would also determine the noise on the wires and make sure they are properly shielded. I would remove the wires from the current analog level sensing device and put a resistor across the wires. If the analog input is changing then there is noised induced on the wire. If the input at the card is shorted out and there is still noise then the analog input card is faulty.

What is r, x and phi? Only one is the level. Are you sure this formula can't be simplified?

The formula can be used to generate a spline or cam table that can be quickly indexed into to get the volume much faster.

 

[kallileo]

The manufacturer of the current sensor told me that if for example the measurement range of the rod is 1500 mm and with 5 mm resolution the distinct values on the rod will be 1500/5=300.

The formula is already simplified you can see the original formula in link below:
http://www.lmnoeng.com/Volume/InclinedCyl.htm

It's the Case 4 - Bottom and top ends partially covered

 

[Lancie1]

As I mentioned above the tank is a inclined cylinder so the aforementioned formula should be used to calculate the table.

No, I don't think you did mention that it was an "inclined" cylinder. You said it was horizontal. If it is horizontal, then the calculation can be simplified greatly. If it is at some angle to horizontal, then you are stuck with the complicated formula.

"The level sensor is installed on horizontal tank of cylindrical shape."

 

[RocketTester]

As you found, the average of a set of fluctuating readings can have better resolution than the individual readings. This is well-known in electrical engineering. It was widely used in things like cameras.

Indeed, purposely adding noise (or "dither") can increase the resolution. If the signal is perfectly quiet, it can sit on one A/D count, so you don't know exactly where it is. By adding a little random noise, you can get the signal to span a few A/D counts which gives more information to determine where the exact center lies. Of course, the disadvantage is that time-averaging gives a lagged response, but probably of little concern for you.

 

[curlyandshemp]

hoo boy, the old float level sensor, probably one of the worse level sensors around. The assumption is this ball will move linearly from top to bottom. Your best bet is to use a temperature compensated level sensor such as the Anderson SL series.
http://andinst.com/PDFs/5006.pdf

 

[shooter]

peter is correct these levelsensors are very bad for your goal, they do give a simple indication, but you can not use it as calculation how much bertha123 has given today .
After installation of a tank, you must make a table (BY HAND) of each tank and put it in the PLC.
Use interpolation in between the points. to get rid of the resolutions.
Together with a temp and a density (of milk, take the fatcontent in account) and a good analog sensor (not these with a lot of resistors in a tube with a magnet floating around some reedswitches).

 

[kallileo]

The cause of the fluctuations was very bad design of the power supply.
The sensor was supplied 24Vdc directly by rectifier bridge without the filtering capacitor and voltage regulator. This way the due to the rectified AC voltage the sensor was switched On-Off all the time.
I have confirmed that the resolution of the rod of the current sensor is 5 mm. This way for example the 1500 mm rod is divided in 300 sections. The reading is not changing unless the float is moved by 5 mm.
On big tanks 5 mm could be as much as 50 liters so the displayed result isn't accurate at all.
We have ordered new Magnetostrictive sensors with resolution < 1 mm. I think this way we will be able to get more accurate results.
This sensor has its own controller with embedded LCD display. Some points of the calibration table are inserted into the controller and the rest of them are calculated using linear interpolation.