Beruflich Dokumente
Kultur Dokumente
Introduction
I have a short case study to present. With hindsight the solution is pretty obvious but I
think its interesting as a demonstration that simple problem can be hidden by a complex
simulation.
A PID pressure controller on a seal oil system would periodically oscillate and the cause
was not understood. It was not clear if it was the PID loop causing the problem or if it
was reacting to outside disturbances.
dP
Flare
History
Simulate 2 1
pumps
PC
DPC
Oil Capacity 2
SEAL
Oil Capacity 1
210 psi
Vo
. V .dt
Bpipe
1
C
P = pressure (psi)
V = volume flow into system ( m3/sec)
C = system capacitance
V0 = total volume of system (m3)
B = bulk modulus (psi) =
Boil
Bpipe
70 , 000
psi/m3
d = wall thickness
D = pipe diameter
E = modulus of elasticity = 29500000 psi (cold rolled steel )
CV . K .V .
dP
sg
F = flow (m3/hr)
dP = pressure drop across valve (psi)
CV = valve CV
K = 0.227 , conversion for CV in US gallons per minute
Sg = 0.83 = specific gravity
V = fractional valve opening ( 0 to 1 )
Seal resistance modelled as valve and CV chosen to match plant oil flow
plant
data
Simulation
Recovery time
from pump trip
120 secs
140 secs
Minimum
pressure
reached
160 psi
160 psi
33 %
38 %
Valve
movement
But is it
430 PSI
250 PSI
160 PSI
Remove interpolation
But is it
Pressure falls in 1 scan period of historian
Tells us nothing about the response time of the system except its < 10secs
Recovery time
Function of controller tuning
430 PSI
250 PSI
160 PSI
Integral time
Ti = 0.03 min
Integral time
Ti = 0.03 min
= 0.9/K * ( T/Td )
= 0.04 % / %
Integral time
Ti = 0.03 min
In fact for dominant dead time processes they are usually tuned with
much more integral and lower gain.
Typically
Ti = Td/3
Kp = 0.18 / K
This would imply dead time of about 5 seconds and time constant
of about 0.1 seconds.
This is pretty reasonable the hydraulic oil is virtually
incompressible and we can easily get to a few seconds or so of dead
time
Adding two seconds of dead time to the simulation was enough to
prevent us raising the controller gain beyond that seen on the plant
Conclusions so far
The oil is so incompressible the time constants are pretty insignificant ( so why spend
any time modelling each individual pipe segment in detail )
The loop dynamics are dominated by seconds of (apparent) dead time these are an
accumulation of effects, some known but some unknown ( so we cant model them
accurately anyway ). Some typical values
Simulation
Added a ramp rate limit to the simulation and a couple of
seconds of dead time
Tuning values from the plant now limiting values for
simulation
Some comments in the original simulation report
Transients are not accurately simulated but the behaviour
of the system can be qualitatively predicted
It is not proposed to give controller settings here past
results would suggest they do not pass well from model to
plant with a great deal of success
Quick simulation
100
90
80
70
60
50
40
30
20
10
0
0
Quick simulation
100
90
80
70
60
50
40
30
20
10
0
0
10
12
14
16
18
20
Simulation is below the trip setting for 70 seconds. To reduce this to the trip
delay period of 10secs requires us to reduce our dead time/time constant to
About 2 .
Taking the positioner off the valve, may help but again not
significant enough to fix the problem
Pipe changes , valve characteristic changes , adjusting
positioner characteristic limited value
Only solution is to soften the system characteristic and install
some hydro pneumatic accumulators