Hello,
when you start with 19 000 and you add 0.1 you have 6 digits. Between 6 and 7 (in decimal system) digits the floating numbers beginn to be rounded.
Without having calculated, I think it is possible you have this effect.

Have a look on the definition of REEL:
You have 23bit mantissa and 8bit for the exponent and a signum bit (all in dual system), this is in decimal system a matissa with 6 or 7 digits.

On a S7-300/400you should use DINT, here you have no rounding.
Just start with 190000 (for 19 000) and add 1(for 0.1) each time. DINT can go up to 2 147 483 648 (which is 214 748 364.8 for you)

If you have a S7-1200 you should use LREAL (long real with 52bit matissa) instead of REAL.

you can use oscat library to implement quasi LREAL variable:

TYPE REAL2 :
STRUCT
R1 : REAL;
RX : REAL;
END_STRUCT
END_TYPE

FUNCTION R2_ADD : REAL2
TITLE = 'R2_ADD'
//
//R2_add adds a real to a double real which extends the accuracy of a real to twice as many digits
//
//uses: REAL2 [UDT11]
//
VERSION : '1.0'
AUTHOR : hugo
NAME : ADD
FAMILY : R2

VAR_INPUT
X : Real2;
Y : REAL;
END_VAR
VAR
temp: REAL;
tR2_ADD: REAL2;
END_VAR

BEGIN

temp := X.RX;
tR2_ADD.RX := Y + X.R1 + X.RX;
tR2_ADD.R1 := temp - tR2_ADD.RX + Y + X.R1;

R2_ADD := tR2_ADD;

(* revision history
hm 20.3.2008 rev 1.0
original version

*)
END_FUNCTION

Look in OSCAT lib manual, I think you can use a part of this UDT (Rx or R1) as "most significant" so you can rapresent a totalizer with only one real number (for example on panel), but you have double precision when accumalting numbers.

However is not recomended to use real numbers for totalizer, except if you have fragment (non integer) parts to totalize

Siemens REAL precision fault:

Same problem on a 1518 fw2.1, and its not that I am reaching the last significant number.

Problem starts right from 0.0.

Try to run on each cycle:

fRealNumber := fRealNumber + 0.1; 

You will get numbers like

28.70007

100.399

149.6999

229.8048

1768.015

and so on.

Same problem with LReal, seems like last bit of the number is random. 

LuisPim

Siemens REAL precision fault:

It is time you make a study of how floating point numbers (REAL) are represented by digital computers. Wikipedia's description should help but this calculator makes it more visual:
http://www.exploringbinary.com/floating-point-converter/

Perhaps it is best for you to use 64-bit integer with a whole number divider.

Hello experts,

With reference to below query and discussion,

Shall we consider it the software issue in Siemens PLC?

I have  also faced same issue for totalizer but instead of adding 0.1 in real , 1 added in DINT , converted DINT to real and then divide by 10.0.

But in some cases if we need to calculate deltaP it creates issue

If  P1= 10.0 &  P2=0.9 then P1-P2=10.0-10.9= 0.9 but sometimes it gives result as 1.0.

Shankar

LuisPim

Siemens REAL precision fault:

Same problem on a 1518 fw2.1, and its not that I am reaching the last significant number.

Problem starts right from 0.0.

Try to run on each cycle:

fRealNumber := fRealNumber + 0.1; 

You will get numbers like

28.70007

100.399

149.6999

229.8048

1768.015

and so on.

Same problem with LReal, seems like last bit of the number is random.