Principles of Adaptive Control
The Control Design Scheme Modeling
(1) Mathematical
representation of
a system
Satisfactory Design
objective
Unsatisfactory
Adaptive Control =
Design
Validation
(partial) (2)
automatic realization
Achieved
of the design practical
performance
Controller
Implementation
(3)
Robin DE KEYSER EeSA Department Ghent University / Belgium 1
Principles of Adaptive Control
Predictable Unpredictable
Constant but
changes changes
unknown
in in
dynamics
dynamics dynamics
Real-Time
Auto-tuning Auto-tuning Auto-tuning
Identification
Gain
Adaptation
scheduling
Constant Predictable Unpredictable
controller parameter parameter
parameters changes changes
AUTO-TUNING ADAPTIVE SELF-LEARNING
Robin DE KEYSER EeSA Department Ghent University / Belgium 2
Principles of Adaptive Control
1. Auto-Tuning: ref. part I
2. Adaptive: ref. beside
Vd
C in (t-)
Model: C in (t)
Q
dC (t )
Mass balance: Vm = QCin (t ) QC (t )
dt
Vm
C ( s) e s
Transfer function: = P( s) = C C(t)
Cin ( s ) 1 + Ts
Time constant T = Vm Q and time-delay = Vd Q :
both varying if the flow Q varies (e.g. 0.4 Q 2)
Solution: tune at Q = 0.6; Q = 1.0; Q = 1.4; Q = 1.8;
Schedule: for 0.4 Q < 0.8 use parameters for Q = 0.6; ...
3. Self-Learning (Self-Adaptive): ref. next pages
Robin DE KEYSER EeSA Department Ghent University / Belgium 3
Principles of Adaptive Control
Principle of Principle of INDIRECT
(SELF-)Adaptive Control Adaptation (MODEL-Based)
Adjustment mechanism
Adjustment
mechanism
Synthesis
Identifier
rule
Setpoint or
reference Output
r + u y
Setpoint or
Controller System reference Output
- r + u y
Controller System
-
Robin DE KEYSER EeSA Department Ghent University / Belgium 4
Principles of Adaptive Control
Principle of DIRECT Adaptation Principle of EXPERT
(MODEL-Based) Adaptive Control (RULE-Based)
Reference Model output Qualitative performance
model criteria
Adjustment mechanism
Adjustment mechanism
Adaption rule
Expert system
Output Setpoint or
reference Output
r + u y
r + u y
Controller System Controller System
Setpoint or
reference - -
Robin DE KEYSER EeSA Department Ghent University / Belgium 5
Principles of Adaptive Control
Principle of an INDIRECT Adaptive Controller (explicit identification)
^
Objective: Identify
Kv
Kv
- process K Kv
1 + Ts Kc = ------
Kv
Process
Kv: varying, unknown 1 S
Input
X Kv ---------------
T : constant, known r(t)
1+sT
Kc
- compensate the process, ^
Kv
such that it behaves like K :
1 + Ts K
K : specified
Robin DE KEYSER EeSA Department Ghent University / Belgium 6
Principles of Adaptive Control
Principle of a DIRECT Adaptive Controller (implicit identification)
dK c e m
= k *e Model K
dt K c ---------------
+
e
X
MIT - rule Input
1+sT
-
min e2 r(t) Process
Kc Multiplier
2
1 s
e X Kv ---------------
gradient: 1+sT
K c
Kc
k Integrator
K Kv ----
e= r Kcr s
1 + sT 1 + sT
e K 1
= r = m = , k = k *
K c 1 + sT Kc
Robin DE KEYSER EeSA Department Ghent University / Belgium 7