Process instruments testing- Practical approach to define Pass/Fail measurement errors 1

[DjazAutomation]

Some process instruments such as PT (pressure transmitters) demonstrate an accuracy of 0.1% or 0.2% (depends on the technology of course). I understand that to calculate the overall measurement error of the measurement system (measurement system: PT + impulse line + Valves + PLC input module + Pressure calibrator), we need to take into account all sources of measurement errors.

We need to set pass/fail criteria for the measurement error to help the technician during calibration/testing of a particular instrument to tell if the measurement errors he records are acceptable or not. (the technician uses a certified Druck pressure calibrator)

Now considering the number of instruments we got, it would be a huge and tedious task to calculate the overall measurement error for each instrument using the usual error formula.

I would like to know if there is a rule of thumb to define the acceptable measurement error by type of instrument.
An example:
- The PTs we got on site have 0.2% accuracy
- The Druck calibrators we got have 0.025% accuracy
- The PLC input module has an accuracy of 0.1%
- Add a safe margin to take into account the remaining source of errors (Impulse line, elevation, etc.)

In this case, I would say that I can use 1 or 2% error as a good and practical pass/fail criteria for the measurement error.
What do you think ? Is my reasoning correct ?

Djaz

 

[IRstuff]

That should depend on what your accuracy requirements are for your reported measurements. 1% may be too loose if your measurements need to ±0.5%

Depending on the application and the customer, they may require a 4:1 or 10: test accuracy ratio (TAR), i.e., your measurement accuracy would need to be 1/4 of the customer's accuracy requirement. For difficult measurements, TARs of 2:1 or 1:1 are sometimes acceptable.

TTFN
faq731-376

Need help writing a question or understanding a reply? forum1529

 

[DjazAutomation]

Many thanks IRstuff and sorry for the late reply.
It's seems that I was heading in the wrong direction...
It's true that it's the Process accuracy requirement (Customer requirement as you call it) that matters to define the tolerable measurement errors.

I was trying to start from an existing measurement system performance (based on the Manufacturers specs)and define the tolerable measurement errors based on that, on the assumption that the EPC contractor has selected those instruments carefully to meet the process (customer) requirements.

I would like to express my problem differently based on your explanations:
1- At the engineering phase of the Plant construction project (in 2003), the customer requirement was 0.8% accuracy for a particular instrument loop.
2- The EPC contractor provides a measurement system that provide 0.2% accuracy (4:1 ration in this case) to meet the customer requirement.
3- As an operator of the plant (in 2013!), I don't have any idea about what the customer requirement was (in this example 0.8%). The only data I have is the System performance that I can evaluate based on the instruments specs I have.

I would appreciate if you or anyone can give me an example about how to evaluate the accuracy requirement for a control loop or safety loop (even a basic example would be okay for me).
Thanks

Djaz

Thanks

 

[IRstuff]

Barring any information from the customer, there is no "rule of thumb" since it's process dependent, as it should be. Your facility ought to have process control data from which you can determine where the process goes out of bed and how accurately that needs to be known.

A more fundamental thing would be to re-look at your OP and the numbers you represented. Assuming they are all normally distributed/independent variables, RSS'ing them would result in 0.3% net accuracy, but you need also to determine what your "accuracy" represents, i.e., is it a 1-σ value or something else? How many false positives are you willing to tolerate? How many false alarms, etc? Given that, you can determine using Student's T distribution or similar to determine how large a margin is required, which would then set your requirement.

TTFN
faq731-376

Need help writing a question or understanding a reply? forum1529