I have a valve I have linearized on a pipeline inlet into our sour gas
plant. Typically this 8" butterfly valve runs wide open, and when it
does, the pipeline pressure upstream of the valve is just under 100 kpa
above the downstream pressure (plant pressure). When the valve begins
to close the pipeline pressure goes up, and then obviously the valve
gain goes up. I have a H2S slug controller configured in the DCS to
close the valve to 20% when the slug is detected, then control the rate
in of the slug, and then disable, and go back to flow control when the
slug is over. Depending on the length of the slug, there can be 1000+
kpa differential across the valve when the slug is over, so what was
the flowrate with the valve wide open before the slug is now flowing
with the valve around 50% open until the pipeline pressure is brought
So anyway, I figured the way to linearize the valve was to have the
valve wide open, with minimum dp across the valve, and move the valve
to 10%, let the flow stabilize somewhat and then move to 20%, and keep
going that way. By the time I finish, I'm flowing a lot more than at
first because I've created line pack in the pipeline, so I just have to
kind of figure the best fit for the top numbers.
So my question is, is there a better way to linearize this valve? I
did get the valve profile I expected out of a butterfly, and it seems
OK. The flow controller is kind of tuned for about a 500 kpa
differential, so I spit the difference and get adequate control.
What you need is a linear INSTALLED flow characteristic; i.e. dM/dx (or dQ/dx)
gain, or slope of the curve relating flow to valve opening) to be approximately
the valves opening range under installed conditions.
a) A situation in which pressure differential across the valve is constant at
all flows calls
for a valve with linear INHERENT flow characteristic, to produce a linear
b) A situation in which the pressure differential across the valve decreases
flow and valve opening calls for an equal-percentage INHERENT flow
characteristic, so as to
compensate and produce a linear INSTALLED flow characteristic.
I may have misunderstood; but this latter situation sounds like yours. Butterfly
valves have an
approximately equal-percentage INHERENT flow characteristic, so I'd expect the
characteristic of you valve to be approximately linear. Use plant data to make
phase plots of
normalised flow (0-100%) vs. stem position (0-100%) over time. How close to a
45-degree line is
Kelvin B. Hales
Kelvin Hales Associates Limited
Consulting Process Control Engineers
I haven't worked on the big industrial stuff, but here's what I think:
Assuming roughly atmospheric pressure on the downstream side of the
valve put a pressure sensor on the upstream side. You'll have to do
some PLC programming but at reduced valve openings your flow should be
roughly proportional to the valve opening * sqrt(pressure). At wide
openings the flow will be controlled by the pipeline; it'll be up to you
to figure out where the valve stops making a difference and the pipeline
head drop takes over.
If you can't assume atmospheric pressure on the downstream side then
think of a differential pressure sensor, or two sensors.
An approach that is adequate in theory and will range anywhere from
perfectly adequate to grossly malfunctioning in practice is to build a
little dynamic model of the pipe pressure as a function of flow and
time. If you can't make it work with a simple low-pass filter with
perhaps a squared term in there for flow then it's probably too complex.
Use this model to predict the pipe pressure given the valve position
and use the valve position to predict flow. Then servo your valve
position to get the flow that you want.
This is a WAG. YMMV. UAYOR. IMHMHUMA. Etc.
There's obviously something I don't get. Is there a reason not to just
use feedback? Sense the flow with a prop, Pitot-static, venturi, or
other flow gauge and adjust the butterfly valve to achieve the desired
value? I know it works for sewage; why not gas?
James Watt didn't invent steam engines. They had been around for
generations when he produced his first one. He certainly contributed
many refinements, such as extracting energy from both pressure and
condensation in the same space, and the eccentric-driven slide valve.
His great invention, the one that attached his name to the mechanism and
relegated Newcomen* to a footnote, isn't really part of the engine at
all: the flyball governor. It's principle is simple: use the engine's
speed to close a butterfly valve called a throttle. Without a governor
to control the throttle, an automatically-valved steam engine is
unstable. With a governor, it's speed is steadier than a horse's.
It is significant that, although steam engines were used in France and
Germany as well as England, the governor is an English invention. A case
can be made that it wouldn't have happened in a country without a
Consider the poultry incubator. These were in use with moderate success
in several countries, marvels of clockwork. In order to maintain
reasonably constant temperature after stoking, dampers were operated by
clockwork and the firewood size, dryness, and quantity was carefully
controlled. With properly shaped cams and careful clockwork regulation,
the system succeeded more often than not. From a social perspective, the
scheme was monarchical. Intelligent instructions, scrupulously carried
out, was the recipe for success.
In England, with its limited rather than absolute monarchy, the social
paradigm of using analysis of events to guide future action led to a
different technical approach. The temperature in the incubation chamber
was directly controlled the dampers. The firebox could be restoked any
any time with a variety of reasonable fuels. Indeed, doing that depended
on the invention of the bimetallic strip, but many historians of
technology contend that the *idea* of feedback called the invention
forth, just as it later called forth the flyball governor. The idea of
feedback is not a natural one in an authoritarian society. With an
authoritarian mindset, a controller *controls*, it doesn't accept input.