richard_swain154

I am looking for the best way to find the average of load cell, I need to sample every 10ms for a variable amount of time to a maximum of 120seconds (1200 samples). I cant use a rolling average becausethe test time/number of samples will vary, when the sample/test is complete i need to ignore the first and last 10% of the data taken and use the mid 80% for the calculation. Is there a way to store more than 100 samples of REAL data in an Array? is datalogging the way to look at it? or Use SCL to calculate. Any help would be greatly appreciated. The application is for a tensile tester that complies to ASTM F88 if anyone has come across this problem before.

Taking a sample every 10ms for 120s will not give 1200 samples. You make no mention which software or hardware you are using.

You could use a cyclic (time) interrupt to give the timing your require. Recording the first 10% and last 10% does not make sense since the running average calculation does not depend on it. So perhaps delay the recording / calculation and stop early.

Working with REAL is a problem. Better to work with the raw (INT?) value. It is less processing and best accuracy to work with the raw value.

Getting a running average does not need to use an array. Assume the raw value is an INT and 1200 samples are taken. 
1) Inside the cyclic OB:  DINT total += raw INT value
2) After the 1200 samples: DINT average := DINT total / 1200
3) Calculate the REAL engineering value from the DINT average

qwazee

Taking a sample every 10ms for 120s will not give 1200 samples. 

Hold my beer.

10 x 120 = 1200

:D

?

jacek d

Hold my beer.

10 x 120 = 1200

:D

You are such a joker most of the time, it is hard to tell if you are making a joke now.

1. {1000ms per 1sec} / {1 sample per 10ms} = 100 samples / sec
2. {100 samples / sec} * {120 sec} = 12,000 samples

Keep your beer.   ;-)

PS: I am not sure if the measuring device and PLC analog input (?) can deliver a measurement every 10ms.

I still do not know for sure what your sampling rate is. It is  also unknown if you are using an analog input or receive data by communication.

Assuming analog recorded every 10ms for 120s needs the following DB entry that does compile.

I have given it a quick try on a S7-1212 G2, just out of curiosity. The PLC does not even notice the code being there:

I have a 10000us cyclic interrupt that does the sampling conditionally:

Just adding some considerations, doing this type of math in real values is not recommended. While adding all the numbers, specially in the last cycles of the for loop, you will have a very big number being added to a small number. In this case it will be fine i guess, 10000x is not that big of a difference in a 32 bit float, but it is better to entirely circumnvent the problem and go to a int sum.

as an example, if you do 16777216.0 + 1.0 it will equal 16777216.0

16777216 is the first case of a full integer error, but you are missing precision way before this.

With an integer sum you need to keep in mind that they are not very big, but in modern PLCs this can be easily solved with long integers. in my example the variable sum is a LINT and i am adding INT, worse case scenario i can add 281’483’566’907’400 INTs before overflowing.

In my mind it is better to use integer values because it is much easier to estimate the limits of the system.

alanTI

Just adding some considerations, doing this type of math in real values is not recommended. While adding all the numbers, specially in the last cycles of the for loop, you will have a very big number being added to a small number. In this case it will be fine i guess, 10000x is not that big of a difference in a 32 bit float, but it is better to entirely circumnvent the problem and go to a int sum.

as an example, if you do 16777216.0 + 1.0 it will equal 16777216.0

16777216 is the first case of a full integer error, but you are missing precision way before this.

With an integer sum you need to keep in mind that they are not very big, but in modern PLCs this can be easily solved with long integers. in my example the variable sum is a LINT and i am adding INT, worse case scenario i can add 281’483’566’907’400 INTs before overflowing.

In my mind it is better to use integer values because it is much easier to estimate the limits of the system.

Only the S7-1500 has 64-bit integers (LINT / ULINT). 

As far as i have tested the S7-1200 G2 supports all the types and operations with longs (lreal, lint, ltime), i have ran the code above in physical hardware.

Also in the standard S71200 you can declare and use Lreal, maybe also lint but i am not sure.

It seems in the latest firmwares this has been silently implemented.