Emerson Exchange 365

The AI function block provides a "first order low-pass" filter functionality with its time constant being the AI block parameter "PV_FILTER" in seconds. This will help attenuate "noise" on the PV signal. Note that setting the PV_FILTER too large will create a slow response of the indicated variable compared to the true PV change. If the OUT of the AI block goes to a PID block, there is also the same type of filter on the PID block. In general the signal should not be filtered in both places if the AI block in inside the same module as the PID block. Lot's more on this topic but this is the basics!

As James points out, the AI1 filter should be used sparingly. I'd been taught to address process noise as close to the source as possible: first in the process (better location of the sensor), or adjusting the transmitter's signal filter if needed (much more accessible these days with HART). This way, you reduce the effects of aliasing based on the AI signal sampling rate.

DeltaV has anti aliasing filters in its AI input channels, with a configurable filter intended to match the control module execution rate. These filter are often not set correctly. Finally, the AI filter is your last option to reduce noise, and is effectively another tuning variable to the PID, along with Gain, Reset and Rate parameters.

If you over filter, the PV may look good, but the actual process behavior to the Close loop response may not be what is needed for overall production rates. On the other hand, adding the filter can reduce wear on the valve. Use the minimum amount of DCS filtering to achieve your goal and avoid masking a "problem".

Andre Dicaire

In reply to Andre Dicaire:

As usual, I agree with everything Andre says on this topic! Excellent info! By the way, always check each of these potential locations of signal filtering, especially if it seems as though there is excessive filtering (transmitter, AI channel filtering under I/O, AI block in an indication module, AI or pseudo AI in a module with the PID block, a separate FILTER block and finally the PID input filter.