Nonlinear Control Problem
Vivek R. Dabholkar
12 May 2025
PID Controller Tuning
• Old beaten down subject
• Many Textbooks are full of Tuning rules, while very few address
nonlinearities encountered in the industrial setting.
• Most Textbooks say pH control is highly non-linear but do not offer
a remedy that actually works
• Approach presented here is based on Qualitative understanding
first and then implementing appropriate algo. used by experienced
practitioners in the industry.
Examples of Non-linear Process Gain
• Flow Controller Valve Output may vary nonlinearly near 0% and
near high output. A standard PID Algo with fixed controller gain
may work in one region but will be sluggish in other region or
worse can go unstable. Controller action needs to be fast where
change in Valve OP does not appreciably change the flow and vice
a versa. Most control valve do not do much beyond 70% OP.
• Damper OP for Draft Control on Furnaces
• Recycle valve on Compressor
• pH Control
Central Idea
• If the Process Gain is changing then the PID Controller Gain must
also change (one way or the other) to achieve consistently optimal
dynamic control-loop performance across the entire range of
operation and not in a small subset of operating region.
Three Approaches
• Method-1 Output Characterization in the DCS
• Pros:
• Do not require building additional points in the DCS
• It is relatively straight-forward for Air-to-Open Valves
• Cons:
• Need to inactivate live loop and down-load with Output Characterization
• For Air-to-Close configuration is bit tricky (I have witnessed Furnace Shut Down due
to incorrect Damper Characterization)
• Operator cannot relate PID.OP to Valve OP unless special attempt is made to display
additional info. This is a problem during manual stroking of the valve, while it is on
bypass
*Most Experienced Control Engineers are familiar with it and few with pitfalls.
• Central Idea of Method-2 Continuously Change Controller Gain
• Move the PID output quickly in the regions where the process variable, (flow,
draft, pH) changes negligibly.
• Move the PID output very slowly in the regions where the process variable,
(flow, draft, pH) changes appreciably.
• One must change controller gain as a continuous function to avoid jerky
action at cross-over points. This can be done by piece-wise linear function
available as a standard function block in most DCS Systems
*Author personally has had a lot of success implementing this method for
last 20 years, starting with 2004 in Exxon Baytown Cracker Plant using BPL
on PMX System on Furnace Damper Ops.
• Method-2 Variable PID Gain
• Pros:
• Don’t need to inactivate, change, and download existing loop with new Output
Characterization which requires Operator buy-in and interactions
• Don’t need to worry about Air-to-Open v/s Air-to-Close
• Operator is comfortable with PID.OP matching Valve OP outside.
• Can commission piece-wise linear function incrementally if full-range data is not
available.
• Cons:
• Need GENLIN block for the piece-wise linear function created based on Excel Calcs
for the Controller Gain with PVSRC = MAN/AUTO options
• Need Logic Block to push out the Gain after validation into Ext-Gain or K parameter of
PID
• New Engineer may be confused with continuously changing gain
• Method-3 PV Characterization (for Highly non-linear pH Control)
• Pros:
• No need for logic block to push out gain
• Intuitive in nature
• Cons:
• Both SP and PV need Characterization so operators need to be trained to interact with
PID with separate holding block for raw pH Setpoint before transforming
Ref: F. G. Shinskey Characterizers For Control Loops (May 1999),
ExperTune Inc.
Taken From F. Shinskey
• Clearly Process Gain is Highly Non-linear
• Controller Gain can be computed for Variable
Process gain in each interval and then use Method-2
Alternatively
• Invert the curve
(single-variate) by
switching y → x and
x→y
How does it work in Layman Language?
• In the region of steep pH change, PID controller thinks there is not-
much offset between SP – PV (while the sign is maintained) since
transformed SP and PV are both nearly horizontal, so the PID
output connected to reagent flow controller changes the reagent
flow very gradually.
• While in the region of gradual pH Change, PID controller thinks the
is a bigger deviation between SP and PV so PID.OP changes rapidly
to manipulate the reagent flow.
