Floating Point number

Maia · Sep 4, 2007

Hi PLC experts,

Using CFC, I configured a logic to totalize a certain parameter which is flow. The totalizer is functioning well except that, when the value reaches a million, the totalized value being displayed seems not accurate. The increment becomes linear. I'm thinking maybe it has something to do with the 32-bit thing of the floating point data type but I'm not sure what exactly is the explanation for that. Can somebody give me an explanation for this?

Please help.

Thanks,
maia

s7xp.com · Sep 4, 2007

A floting point or real use a sign bit, 8 bit exponent and 23 bit mantisse.
Sign: + or -
Exponent: +E001 or -E001
Mantisse: 1.23

Integer 10 is as float 1.0+e001
1.0 x 10 power of 1 = 10

To convert an integer to real use:

L MW10
ITD
DTR
T MD12

Maia · Sep 4, 2007

Can you include some explanations? Sorry, I didn't catch it..

dchartier · Sep 4, 2007

Hello Maia;

Here is a link to a review of IEEE 754, single-precision 32-bit floating-point values, such as it is used in most PLCs today (with examples):
http://en.wikipedia.org/wiki/IEEE_754

Since floating-point values can attain huge values, i don't think this is your issue here. A value of 1 million is hardly a problem for REAL values.
Please produce an screen capture of your CFC blocks, or better yet an illustration of the logic you are using (along with an exact description or examples of the problems you face), so we can see what we can do to help you.
Hope this helps,
Daniel Chartier

TConnolly · Sep 4, 2007

The problem that you run into with floats as they get larger is precision. For example, with a 32 bit float, you cannot add 1.0 to 16,777,216 and get 16,777,217 because of the loss of precision as the number gets larger. If you are adding a very small increment, like something around .032, then I suspect that is your problem, as you approach 1 million you begin to loose precision. What are you totalizing and how large is each increment?

Maia · Sep 6, 2007

Thanks Daniel! Thanks Alaric!

Will try to get a screenshot of the block and the logic itself by tomorrow and will post it as soon as I get it.

Alaric - The increment is about 0.284 per second. And I noticed that the value becomes inaccurate when it reaches millions, say 3 million. But when the totalized value is smaller, I don't see any problem. I think you're right, the precision might be the problem. Is there any way to correct it?

maia

JesperMP · Sep 6, 2007

Is each increment really in the order of 0.284 ?
How big a total value do you need to count to ?

If you are adding a value with increments of 0.284 then you will run into rounding errors long before you reach 3,000,000.0
The largest mantissa value that a REAL can contain is +3402823. That is indeed about 3 million, but the fractional part has begun to become rounded off at a much smaller value.

A DINT can count to +2147483647
So you could interpret it as +2147483.647, in other words as there is a fractional part with 3 decimals. But that will only let you reach 2 million.

It looks like you have to program your own counter by combining several variables.

PeterW · Sep 6, 2007

Could you come to a programming get-around solution,like have two additional floating point stores, count in one up to a million, then add the million to the second store and zero the first. The third being your total value, which adds the other two together.

Peter Nachtwey · Sep 6, 2007

Don't use floats for totalizing. ALWAYS use DINTs.

Count the number of items you are totalizing using DINT. Multiply the DINT by 0.284 to get desired result. Simple.

TConnolly · Sep 6, 2007

You can add the sample (~.284) to one number and periodically sum that to a second floating point number and zero the first, as Peter suggested. You will loose precision, but the overall value will keep incrementing and will still be accurate to whatever significant digits are meaningful for a particualar number in IEEE754 format

You may also want to evaluate the units you are using. For example, if .284 represents liters then you might want to use CCs, or if its cubic-feet then try gallons.

JesperMP · Sep 6, 2007

Peter Nachtwey said:
Count the number of items you are totalizing using DINT. Multiply the DINT by 0.284 to get desired result. Simple.

Yes, if the increment is constant at 0.284.
But if the increment varies around this value, then you have to count up with a fractional part with at least 3 decimals.

TConnolly · Sep 6, 2007

But one could totalize the input to a certain value and then count those values, eg, if totalizing cubic meters, totalize the samples until you have more than 1 cubic meter, then increment the count for cubic meters using a DINT and then subtract 1 cubic meter from the running intermediate total. Any convenient unit can be used, even quarter cubic meters, and then multiplied by whatever is needed to scale it to the final units.

Floating Point number

Maia

Member

s7xp.com

Member

Maia

Member

dchartier

Member

TConnolly

Lifetime Supporting Member

Maia

Member

JesperMP

Lifetime Supporting Member + Moderator

PeterW

Member

Peter Nachtwey

Member

TConnolly

Lifetime Supporting Member

JesperMP

Lifetime Supporting Member + Moderator

TConnolly

Lifetime Supporting Member

Similar Topics