Rate Predictive Process
Control
By: Sedona Rockwood
With Dr. David M. Bagley
Background
Process control is the control of chemical processes for quality and
safety.
Since the 1980s Model Predictive Control (MPC) technology has been
widely used for industrial process control.
The use of MPC is widely published, however there are certain
disadvantages.
Background (cont.)
MPC is based on detailed process models and error-minimization.
Mr. Kernβs work concluded that these assumptions are actually root
causes behind MPCβs limitations Specifically:
1. Process models change frequently requiring retuning
2. In industry process stability is more important than error-
minimization
RPC and XMC
Mr. Kern developed rate predictive control (RPC) algorithm to address the
limitations of MPC.
RPC adjusts the controller output based on a preset move rate, the ongoing
rate of change of the controlled variable, and a future predicted value based
on process response time.
RPC is perhaps industryβs first and only inherently adaptive process control
algorithm.
Mr. Kern also developed a model-less method of multivariable control (XMC),
which utilizes the RPC method internally on a matrix control basis.
Goals
Determine the control theory behind RPC
Test the theoretical performance of RPC
Compare the theoretical performance of RPC to the operational
performance of RPC
Procedure
Simulate RPC in Excel and Simulink
Determine overall control loop and transfer functions
Simulate RPC in MATLAB using control loop and transfer functions
Control Loop Process
Control Process
Transfer
RPC Band Transfer Gain Controlled
Move Time Delay FunctionFunction Variable
Set Point
Rate Predictive
Transfer
Function
Transfer Functions
1
πΊ = In Control Theory, transfer π π functions are used to
understand the limitations
1
πΊ = of the controller.π Ο1π + 1
πΊ = πβπ‘ππ π‘ππ Ο1 = Process response time
td = Time delay
πΊπ
ππΆ = ΞΈπ + 1 ΞΈ = Predicted response time
Transfer Functions (cont.)
Standard feedback control finds error through the following equation:
ππ π β π = πΈ
RPC finds error through the following equation:
ππ π β πΊπ
ππΆπ = πΈπππππππ‘ππ
Overall Transfer Function
From the overall control loop given previously the following overall
transfer function is defined:
π πΎ3πΎ2πΊππΎ1πΊπ
=
ππ π 1 + πΎ3πΎ2πΊππΎ1πΊππΊ1
Open Loop Transfer Function
Based on the overall transfer function the open loop transfer function
is given as:
πΊππΏ = πΎ3πΎ2πΊππΎ1πΊππΊ1
By substituting the transfer functions in the overall transfer function
can be simplified to give:
ΞΈπ + 1
πΊ = πΎ πΎ πΎ πβπ‘ππ ππΏ 3 2 1 π (Ο1π + 1)
βProofβ of Transfer Functions
Target Y
Conditions:
Excel Y β’ First-order
Process
Simulink Y β’ td = 0
β’ ΞΈ = 10
β’ Ο1 = 10
1
β’ = 10
K3
βProofβ of Transfer Functions (cont.)
Target Y
Conditions:
Excel Y β’ First-order
Process
Simulink Y β’ td = 5
β’ ΞΈ = 15
β’ Ο1 = 10
1
β’ = 15
K3
Stability without Time Delay
For a first order system when ΞΈ = Ο1 without time delay, the open loop
transfer function simplifies to give the following:
πΎ1πΎ2πΎ3
πΊππΏ = π
Stability without Time Delay(cont.)
Conditions:
β’ First-order
Process
β’ td = 0
β’ ΞΈ = Ο1 Phase under
these
conditions
-135
degree line
Stability without Time Delay(cont.)
For a first order system where ΞΈ β Ο1 and there is no time delay, the
open loop transfer function is simplified to give:
ΞΈπ + 1
πΊππΏ = πΎ3πΎ2πΎ1 π (Ο1π + 1)
Stability without Time Delay(cont.)
Conditions:
β’ First-order
Process
β’ td = 0
Phase under
β’ ΞΈ = 5
these
β’ Ο1 = 10
conditions
-135
degree line
Stability without Time Delay(cont.)
Conditions:
β’ First-order
Process
β’ td = 0
β’ ΞΈ = 10
Phase under
β’ Ο1 = 5
these
conditions
-90 degree
line
Stability with Time Delay
For a first order function with a time delay when ΞΈ = Ο1, the open loop
transfer functions simplifies to the given equation:
πΎ1πΎ πΎπΊ = 2 3 πβπ‘ππ ππΏ π
Stability with Time Delay (cont.)
Conditions:
β’ First-order Phase under
Process these
β’ td = 5 conditions
β’ ΞΈ = Ο1
Point of
Instability
β’ K1K2K3 = 3
-180
degree line
Stability with Time Delay (cont.)
Target
Conditions:
Y β’ First-order
Process
β’ td = 5
β’ ΞΈ = Ο1
β’ K1K2K3 =
0.0667
Stability with Time Delay (cont.)
Target
Conditions:
Y β’ First-order
Process
β’ td = 5
β’ ΞΈ = Ο1
β’ K1K2K3 = 10
Stability Constraints with Time Delay
Conditions:
β’ First-order
Process
β’ ΞΈ = Ο1
β’ First-order
Process
β’ ΞΈ = 5
β’ Ο1 = 10
β’ First-order
Process
β’ ΞΈ = 10
β’ Ο1 = 5
Maximum Allowable Constants with Respect
to Time Delay
Operational Issues
Mr. Kern set operational guidelines for defining K3 and ΞΈ shown below:
ΞΈ β₯ π‘π + Ο1
1
πΎ3 β€ πΎ1πΎ2ΞΈ
Operational Issues (cont.)
Target
MV = Conditions:
Manipulated β’ First-order
Variable Y Process
β’ td = 5
β’ ΞΈ = 15
MV β’ Ο1 = 10
1
β’ = 15
πΎ3
Operational Issues (cont.)
Target
Conditions:
β’ First-order
Process
Y
β’ td = 5
β’ ΞΈ = 15
MV β’ Ο1 = 10
1
β’ = 10
πΎ3
Operational Issues (cont.)
Target Conditions:
β’ First-order
Process
Y β’ td = 5
β’ ΞΈ = 15
β’ Ο1 = 10
MV 1β’ = 1
πΎ3
Operational Issues (cont.)
Target Conditions:
β’ First-order
Process
Y β’ td = 5
β’ ΞΈ = 10
β’ Ο1 = 10
MV 1
β’ = 15
πΎ3
Operational Issues (cont.)
Target Conditions:
β’ First-order
Process
Y
β’ td = 5
β’ ΞΈ = 1
β’ Ο1 = 10
MV 1
β’ = 15
πΎ3
Conclusions
This analysis confirms the overall control loop and transfer functions
determined previously for RPC.
For a first order process without time delay, RPC is inherently stable, even for
errors in prediction time versus actual process response time.
For a first order process with time delay, RPC is stable within a range of
constants.
Mr. Kernβs published guidelines for βoperational stabilityβ (no overshoot for
either the Y or MV) are much more conservative than conventional process
control stability criteria (no increasing oscillations).
Questions?