Booster station here, CompactLogix 1769-L24ER-QBFC1B with a 1769-IF4 in slot 2, mag meter on channel 0, 4 to 20mA. Two 50 hp pumps on PowerFlex 525s share the panel with it.
The flow value bounces about plus or minus 3% every scan. 410, 398, 421, 405 gpm. The low flow alarm at 400 chatters every few seconds whenever we sit near it.
I widened the alarm out to 380 first and it still chatters on a slow day, so then I tried averaging the last 100 samples, which does calm it down but it lags a pump start by ten seconds and the alarm fires late. No luck either way really.
Any suggestions on what people use to filter an analog in the PLC? I don't want the operators losing the real flow.
Tag and range first. Channel 0 is scaled 0 to what, and what does the meter itself show on its display while the PLC shows 410 then 398? Is the bounce there with both drives stopped?
0 to 800 gpm on the channel, and the meter display sits at 407 and barely moves while the PLC value jumps around. Good question on the drives. Both stopped, the value sits still. One drive at 40 Hz and it bounces. With the meter cable off at the card and a 12mA source on the terminals its rock steady, drives running or not.
Then the drive's on your cable, not on your meter. Fix that first, filter second.
Moved the meter cable out of the tray it was sharing with the drive output leads, and the shield was grounded at both ends so that's at the panel bar only now. Bounce is down to about 1%. Still enough to chatter at 400 when we sit near it.
Tried the 100 sample average again on top of that, same ten second lag. Do I just live with it?
No. The average is the wrong tool. Every sample in the window carries the same weight, so the newest one is 1% of the answer and a step takes the whole window to come through.
A first order filter weights the newest sample more. One line, in a periodic task at 100 ms so the rate's fixed:
```
Flow_Filt := Flow_Filt + K * (Flow_Raw - Flow_Filt);
```
At 100 ms, K of 0.1 gives you a time constant of about a second, 0.05 two seconds, 0.2 half a second. That's the knob. Then the alarm gets its own deadband, on below 400, off above 410, so the last 1% can't chatter it. Someone did an article that has the filter in it: https://plctr.com/utilizing-plc-for-condition-monitoring-and-predictive-maintenance-in-manufacturing/
Filter's been in a 100 ms periodic task with K = 0.1 since yesterday afternoon. The number's steady to a gpm and it follows a pump start in a second or two, which the operators can live with. Deadband on the alarm, 400 on and 410 off. No chatter overnight.
Two things stacked up. The drive leads were coupling into the meter cable and there was no filtering on the value at all, so rerouting it with the shield grounded one end took most of it out and the one line filter took the rest. The 100 sample average is gone.
One thing I haven't decided is the historian, it gets the filtered value now and I'm not sure I want a smoothed number in there for the pump curve work. Thanks marekw and valh7.
Log both. Raw for the curves, filtered for the alarm.