Created Date: August 04,2008

Created by: AZRoger

vmcanard,

The link below shows how IEEE floating point assigns meaning to it's bits. (Modbus uses this standard format.)

http://www.psc.edu/general/software/packages/ieee/ieee.html

It's probably word-swapped in the PLC. It's not trivial to decode, but can be done with a few accumulator operations. However, from what I 've read on other posts, if you ever actually get a negative number from your meter (co-generation, etc.) it could be tough sledding. You also need to know whether 32 bit or 64 bit numbers are arriving as modbus doesn't provide descriptors of what's actually in the data.

If you can force to meter to produce results with a fixed scaling factor, then things are much easier. Can this be done?

Roger

Created Date: August 04,2008

Created by: franji1

IEEE floating point is optimized for a large range, but limited significant digits, i.e. you can have really big numbers,

1,234,567,890,000,000,000,000

or really small numbers

0.000 000 000 000 123 456 789

So what range do you expect for the REAL values? 0..100? can they be negative? How precise do you have to be, i.e. is 3.01 much different than 3.02?

Created Date: August 05,2008

Created by: franji1

That's good. If at all possible, if the meter can be configured in "tenths of kW " in BCD, that would be PERFECT. However, a lot of field devices come preconfigured and you cannot change the scaling.

If they had that, then if the actual reading on the device was 29.8 kW, then you would have the value 298 BCD in your V memory register, which works natively.

If they do not have the capability of configuring the scaling of the value, then you'll have to do it yourself using the algorithm AZRoger pointed out, which is quite intense for a little PLC. Maybe someone out there has done such a thing in ladder logic (convert IEEE 754 to a binary or BCD value w/1 implied decimal point)?!?

Created Date: August 05,2008

Created by: AZRoger

I'm working on the PLC logic to do this conversion. The IEEE floating point numbers have a 24 bit number (the fraction) and an 8 bit number as the exponent. The exponent determines where the binary-point goes in the value.

The following list of values illustrates an interesting binary-point effect:

Binary 10.1 = decimal 2.5 = 2-1/2

Binary 10.01 = decimal 2.25 = 2-1/4

Binary 10.001 = decimal 2.125 = 2-1/8

Binary 10.0001 = decimal 2.0625 = 2-1/16

As you can see, you can't get exact tenths out of the numbers to the right of the binary-point.

So for the conversion to BCD with 1 digit after the decimal-point we need to keep at least 4 binary digits after the binary-point (1/16ths of a watt). There would then be rounding to get to 10ths.

Is accuracy to 1/16th of a watt from the meter rounded to 10ths for display OK?

Created Date: August 06,2008

Created by: AZRoger

OK. Here's a solution. I think it could be tightened up a bit by remembering what has already been pushed into the stack.

This takes a 32 bit IEEE floating point number in V10100 and V10101 and produces a 7.1 digit BCD number in V10116 and V10117.

The rounding isn't perfect but it's close. Binary and Decimal don't really get along that well. http://forum1.automationdirect.com/board/smile.gif

EDIT: This morning I tested this from 0 to 131,071.9 Watts. The binary calculations overflow above that.

// Rung 1779

// Address 6327

#BEGIN COMMENT

"Floating Point to BCD conversion Sample "

"Takes 32 bit IEEE float and make decimal "

"value with one implied decimal point. NN.N "

#END

STRN B10101.15 //Only do Positive Values

LDD "Float " //Get the whole IEEE value

ANDD K7fffff //Mask off sign and exponent

ORD K800000 //Add in the implied 1 bit

OUTD "Fraction " //save as 32 bit binary

LD "FloatTop " //get the Top 16 bits

ANDD K7f80 //Mask or sign and fraction bits

SHFR K7 //Slide exponent to low orders bits

OUT "ExponentBin " //Save as 16 bit number

SUBB K78 //Subtract 120 (which multiplies by 128)

OUT "BinaryPointPos " //Save binary point position

INV //negate

ADDB K18 //And adjust for amount to shift fraction

OUT "NumBits2Shift " //save number for later shift instruction

LDD "Fraction " //get the fraction from before

SHFR "NumBits2Shift " //shift out the low order bits we don't need

OUTD "BinaryValX128 " //Now we have, in binary, the value x 128

SHFR K7 //divide by 128

BCD //make the integerpart of the number BCD

SHFL K8 //Multiply by 100 (can't use MUL instruction)

OUTD "BCDValX100 " //save this number

LD K7f //get a mask to pick up only fractional bits

AND "BinaryValX128 " //And in to get the non-integer part

MULB Kc800 //This is 6553600/128 which produces 0 to 99

//in the TOP HALF of the accumulator

SHFR K16 //move the answer to the bottom half

BCD //and make it BCD

OUT "DecHundredths " //save the hundredths

ADDD K5 //add 5 to do rounding to tenths

ADDD "BCDValX100 " //add in the 100X integer from before

SHFR K4 //drop one digit to make it tenths

OUTD "BCDValX10 " //save the BCD with one implied decimal

#BEGIN ELEMENT_DOC

"V10100 ", "Float ", "DWORD IEEE Float ", "Low 16 Bits Fraction "

"V10101 ", "FloatTop ", "Sign 8 Bits Exp ", "7 Bits Fraction "

"V10102 ", "BCDValX100 ", "DWORD 8 Digits ", " "

"V10103 ", "BCDValX100Top ", " ", " "

"V10104 ", "BinaryValX128 ", "DWORD 32 bits ", " "

"V10105 ", "BinaryValX128Top ", " ", " "

"V10106 ", "BinaryPointPos ", "Exponent - 120 ", "(Automatic * 128) "

"V10107 ", "ExponentBin ", "WORD 16 Bit Bin ", "Exponent "

"V10110 ", "Fraction ", "DWORD 32 Bit Bin ", "Bottom 16 "

"V10111 ", "FractionTop ", " ", "Top 7 "

"V10112 ", "NumBits2Shift ", " ", " "

"V10114 ", "DecHundredths ", "DWORD 8 Digits ", " "

"V10115 ", "DecHundTop ", " ", " "

"V10116 ", "BCDValX10 ", "DWORD 8 Digits ", " "

"V10117 ", "BCDValX10Top ", " ", " "

#END

Roger