May 2006

In this issue

	Classes and seminars
	Article
	Trick of the month: Sampling signals, ideal and practical sampling rates
	To contact us

Coming classes and seminars

Analytical methods for troubleshooting and Control loop tuning
May 16-18, Dalhousie, NB, English course(Question period bilingual)
May 23-25, Cincinati, OH

Detailed Loop Analysis with ExperTune
June 12-14, Milwaukee, WI

ExperTune Advanced Tools
June 14-16, Milwaukee, WI

See our calendar for 2006

Article

In Control Design, an article by Martin Emond on why we should rarely use ramps in process control;
« How to tame unruly ramp function behavior »
Click here to read the article.

This month's tip : Sampling signals, ideal and practical sampling rates

Data analysis form archived data or from a HMI
When using signals to analyze performance and to diagnose problems, data sampling must be fast and without signal alteration. If not, the observed signals will be different and misleading.

Example: Flow loop
A flow loop is fast and one must capture all movements in the signal. If the sampling is not fast enough, the observed signal will be different. Also, the PID controller must execute its function every second and sometimes each 100 ms. It would be unrealistic to manipulate the valve (or variable speed drive) every 10 s since between two successive executions, the flow would not be under control and many events could occur in between the moves. Hence, to analyze or observe the flow signal (and controller output), we have to sample fast enough.

If we observe such signals every 20s or 30s, the analysis will be corrupted. For example, trying to analyze oscillations, valve stiction, process model, performance, etc. will result in the wrong diagnostics.

Very simply, we have to sample all signals faster than the phenomenon to be observed. If not, not only we miss information but also what we see is wrong! For example, not sampling fast enough will result in a false oscillation into the observed signal:

The real signal is red.The blue signal is what will be displayed.
Many systems will report a change to the data acquisition system only when it exceeds a threshold; this will also modify the signal.

Case 1 : Noisy flow loop with oscillations.
Sampling= 1 s	Oscillation, 20s	Noise, 0.3%

First graphic displays real data. On second graphic, the report by exception modifies the signal. Third graphic shows that the sampled signal is almost the same as real values.

Case 2 : Noisy flow loop with oscillations; slow sampling.
Sampling= 15 s	Oscillation, 20s	Noise, 0.3%

First graphic displays real data. On second graphic, the report by exception modifies the signal and observed oscillations are different. Third graphic shows that the sampled signal is really different and even a new oscillation appears.

Case 3 : Noisy flow loop with oscillations; very slow sampling.
Sampling= 30 s	Oscillation, 20s	Noise, 0.3%

First graphic displays real data. On second graphic and third graphic, the oscillations disappeared.

Conclusion :
To use sampled data :

Sampling must be fast enough Report by exception should be small Filtering and compression algorithms should be minimal

General recommendations :

Pressure and flow loops : Level and temperature loops : Others :

1s 5 à 10 s sampling rate faster than what we want to observe

Rule of thumb : sampling time smaller than loop dead time.

• ts < td

To contact us

	Lévis's office	+1 (418)834-4321
	Pointe-Claire's office	+1 (514)695-3492
	Hartland's office	(877)867-6473
	Toll free	(877)867-6473
	Email	info@topcontrol.com