ADAPTIVE NEURO CONTROL SYSTEM

INTRODUCTION

Adaptive control:
 Adaptive control is the control method used by
a controller which must adapt to a controlled system
with parameters which vary, or are initially uncertain.
 Adaptive controller is a dynamic system with on-line
parameter estimation.
 inherently nonlinear
 and design rely on the Stability Theory
Why Adaptive Control?
while dealing with complex systems that have
unpredictable parameter deviations and uncertainties.

Basic Objective
 Maintain consistent performance of a system in the
presence of uncertainty and variations in plant parameters.

 Adaptive control is superior to robust control in dealing

with uncertainties in constant or slow-varying parameters.

 Robust control has advantages in dealing with

disturbances, quickly varying parameters, and un modelled
dynamics
In order to design and tune a good
controller, one need to:

(1) Specify the desired control loop performances.

(2) Know the dynamic model of the plant to be controlled.

(3) Possess a suitable controller design method making it

possible to achieve the desired performance for the
corresponding plant model.
Self-Tuning Controllers (STC)

• Combines a controller with an on-line (recursive) plant

parameter estimator
• Reference model can be added
• Performs simultaneous parameter identification and
control
•Uses Certainty Equivalence Principle
controller parameters are computed from the estimates of
the plant parameters as if they were the true ones.
Model-Reference Adaptive Control
(MRAC)

Plant has a known structure but the parameters are

unknown
• Reference model specifies the ideal (desired)
response y to the external command r
• Controller is parameterized and provides tracking
• Adaptation law is used to adjust parameters in the
control law m
Types of adaptive control system

 Direct adaptive control system

 Indirect adaptive control system
Direct ACS:
Direct adaptive control - what we do normally in direct
adaptive control is, given a plant and a controller;

Controller has certain parameters and the parameters are

directly updated using the error signal here.That is, between the
plant output and the output of a model, reference model or if it
is not reference model it can be also the desired trajectory.

The plant output has to follow some command signal y d or it

has to follow some output reference model the output given by
a reference model you give a command signal R here.

The error generated here is used to directly tune the controller

parameters in such a way the overall system dynamics is stable.
So, the parameters of controller are directly adjusted to
reduce some norm of the output error and this is called
adaptive control
direct adaptive control. Here we do not do any kind of
identification of plant parameters directly.
•The design of the controller is done now in order that:

(1) the error between the output of the plant and the output
of the reference model is identically zero for identical
initial conditions;
(2) an initial error will vanish with a certain dynamic.

•When the plant parameters are unknown or change in

time, in order to achieve and to maintain the desired
performance.

•An adaptive control approach has to be considered and

such a scheme known as Model Reference Adaptive
Control (MRAC).
This scheme is based on the observation that the
difference between the output of the plant and the output
of the reference model (called subsequently plant-model
error) is a measure of the difference between the real and
the desired performance.
Limitations

 Despite its elegance, the use of direct adaptive control

schemes is limited by the hypotheses related to the
underlying linear design in the case of known
parameters.
 One of the basic limitations is that one has to assume
that the plant model has in all the situations stable zeros,
which in the discrete-time case is quite restrictive.
 The problem becomes even more difficult in the multi-
input multi-output case. While different solutions have
been proposed to overcome some of the limitations of
this approach direct adaptive control cannot always be
used.
Introduction to Indirect ACS
 The basic idea is that a suitable controller can be
designed on line if a model of the plant is estimated on
line from the available input-output measurements.
 The scheme is termed indirect because the adaptation
of the controller parameters is done in two stages:
(1) on-line estimation of the plant parameters;
(2) on-line computation of the controller parameters
based on the current estimated plant model.
 This scheme uses current plant model parameter
estimates as if they are equal to the true ones in order to
compute the controller parameters.
 This is called the ad-hoc certainty equivalence
principle
•The basic scheme for the on-line estimation of plant
model parameters is shown in Fig.
•The basic idea is to build an adjustable predictor for
the plant output which may or may not use previous
plant output measurements and to compare the
predicted output with the measured output.
•The error between the plant output and the predicted output
is used by a parameter adaptation algorithm which at
each sampling instant will adjust the parameters of the
adjustable predictor in order to minimize the prediction error
in the sense of a certain criterion.
•This type of scheme is primarily an adaptive predictor
which will allow an estimated model to be obtained
asymptotically giving thereby a correct input-output
description of the plant for the given sequence of inputs.
•This technique is successfully used for the plant model
identification in open loop.
•Note that on-line estimation of plant model parameters is
itself an adaptive system which can be interpreted as a
Model Reference Adaptive System (MRAS).
•The plant to be identified represents the reference model.
•The parameters of the adjustable predictor (the adjustable
system) will be driven by the PAA (parameter adaptation
algorithm) in order to minimize a criterion in terms of the
adaptation error (prediction error).
Using the indirect adaptive control schemes shown in Fig, one
can further elaborate on the ad-hoc use of the “certainty
equivalence” or “separation theorem” which hold for the linear
case with known parameters.
•In terms of separation it is assumed that the adaptive predictor
gives a good prediction (or estimation) of the plant output (or
states) when the plant parameters are unknown, and that the
prediction error is independent of the input to the plant (this is
false however during adaptation transients).
•The adjustable predictor is a system for which full information is
available (parameters and states).
•An appropriate control for the predictor is computed and this
control is also applied to the plant.
•In terms of certainty equivalence, one considers the unknown
parameters of the plant model as additional states.
•The control applied to the plant is the same as the one applied
when all the “states” (i.e., parameters and states) are known
exactly, except that the “states” are replaced by their estimates
Direct vs. Indirect Adaptive Control

• Indirect
– estimate plant parameters
– compute controller parameters

•Direct
– no plant parameter estimation
– estimate controller parameters (gains) only
Direct Vs Indirect Adaptive Control

•Comparing the direct adaptive control scheme Fig. with

the indirect adaptive control scheme Fig. one observes an
important difference.
•In the scheme of DACS Fig. the parameters of the
controller are directly estimated
(adapted) by the adaptation mechanism.
•In the scheme of IACS Fig. the adaptation mechanism 1
tunes the parameters of an adjustable predictor and these
parameters are then used to compute the controller
parameters.
To illustrate the idea of “re parameterization” of the
plant model, consider the following example. Let the
discrete-time plant model be:

where y is the plant output, u is the plant input and a

is an unknown parameter. Assume that the desired
objective is to find u(t) such that:
2

•(The desired closed-loop pole is defined by c1 ).

The appropriate control law when a 1 is known has
the form:
3
 However, (1) can be rewritten as:
4

and the estimation of the unknown parameter r0 will

directly give the parameter of the controller.

Using an adjustable predictor of the form:

5

and a control law derived from (3) in which r0 is replaced

by its estimates:

6
one gets:
7

which is effectively the desired output at (t + 1) (i.e.,

the output of the implicit reference model made from
the combination of the predictor and the controller).
Applications:
Direct Adaptive Control of a Phosphate Drying Furnace
•This application has been done at the O.C.P., Beni-Idir Factory,
Morocco (Dahhou et al. 1983).
•The phosphate, independently of the extraction method, has
about 15% humidity. Before being sold its humidity should be
reduced to about 1.5% using a rotary drying furnace.
•The drying process requires a great consumption of energy.
•The objective is to keep the humidity of the dried phosphate
close to the desired value (1.5%) independently of the raw
material humidity variations (between 7 and 20%), feed flow
variations (100 to 240 t/h) and other perturbations that may affect
the drying process.
•The dynamic characteristics of the process vary as a
consequence of the variable moisture and the nature of the damp
product. A direct adaptive control approach has been used to
achieve the desired performances over the range of possible
changes in the process characteristics.
A block diagram of the system is shown in Fig. The drying
furnace consists of a:
• feeding system,
• combustion chamber,
• rotary drying tube,
• dust chamber,
• ventilator and chimney.
•The combustion chamber produces the hot gas needed for the
drying process.
•The hot gas and the phosphate mix in the rotary drying tube. In
the dust chamber, one recaptures the phosphate fine particles
which represent approximately 30% of the dried phosphate.
• The final product is obtained on the bottom of the dust
chamber and recovered on a conveyor.
•The temperature of the final product is used as an indirect
measure of its humidity.
•The control action is the fuel flow while the other variable for
the burner (steam, primary air) are related through conventional
loops to the fuel flow.
•Significant improvement in performance was obtained with
respect to a standard PID controller (the system has a delay of
about 90 s).
•The improved regulation has as a side effect an average
reduction of the fuel consumption and a reduction of the thermal
stress on the combustion chamber walls allowing to increase the
average time between two maintenance operations.
Indirect and Multimodel Adaptive Control of a Flexible
Transmission
•The flexible transmission built at GIPSA-LAB, Control Dept.
(CNRS-INPG-UJF), Grenoble, France, consists of three
horizontal pulleys connected by two elastic belts (Fig).
•The first pulley is driven by a D.C. motor whose position is
controlled by local feedback. The third pulley may be loaded
with disks of different weight.
•The objective is to control the position of the third pulley
measured by a position sensor.
•The system input is the reference for the axis position of the
first pulley. A PC is used to control the system. The sampling
frequency is 20 Hz.
•The system is characterized by two low-damped vibration
modes subject to a large variation in the presence of load. Fig.
gives the frequency characteristics of the identified discrete-time
models for the case without load, half load (1.8 kg) and full load
(3.6 kg).
•A variation of 100% of the first vibration mode occurs when
passing from the full loaded case to the case without load. In
addition, the system features a delay and unstable zeros.
• The system was used as a benchmark for robust digital control
(Landau et al. 1995a), as well as a test bed for indirect adaptive
control, multiple model adaptive control, identification in open-
loop and closed-loop operation, iterative identification in closed
loop and control redesign.
Conclusion:
1.Adaptive control provides a set of techniques for automatic
adjustment of the controllers in real time in order to achieve or
to maintain a desired level of control system performance, when
the parameters of the plant model are unknown and/or change in
time.
2. While a conventional feedback control is primarily oriented
toward the elimination of the effect of disturbances acting upon
the controlled variables, an adaptive control system is mainly
oriented toward the elimination of the effect of parameter
disturbances upon the performances of the control system.
3. A control system is truly adaptive if, in addition to a
conventional feedback, it contains a closed-loop control of a
certain performance index.
4. Robust control design is an efficient way to handle known
parameter uncertainty in a certain region around a nominal model
and it constitutes a good underlying design method for adaptive
control, but it is not an adaptive control system.
 Adaptive control can improve the performance of a robust
control design by providing better information about the
nominal model and expanding the uncertainty region for which
the desired performances can be guaranteed.
Thank You