EEE4434
Intelligent Control System Design
Dr Samsul Bahari Mohd Noor
samsul@[Link]
Control System Design
– Basic Steps (review)
If we consider the local level, the typical steps in designing the control are:
1. Define the components of the process to be controlled, manipulated, or
which are fixed. Additional define which are the variables of interest.
2. Define the user control goals.
3. Get a draft model of the process and the attached signals.
4. Select which will be the manipulated and measured variables.
5. Choose a suitable control structure.
6. Translate into a control language (also suitable for the selected control
structure) the user requirements.
Control System Design
– Basic Steps (review)
7. Apply the controlled design methodology based on the decision taken in
the previous steps (variables, models, goals).
8. Validate (by simulation or experimentally) the design and tune the
controller parameters.
9. Define the controller implementation. In the case of digital controllers,
select the hardware and software to fulfil the control requirements.
10. Install the control in the process.
11. Evaluate the controlled system performances.
Control System Design
– Basic Steps
Most of these activities should be performed iteratively. For example, if the
designed controller does not match the requirements, either the control
structure or the variables should be revisited.
If the implementation of the controller introduces additional constraints like
time delays, computation time, etc. the design should be reconsidered, taking
into account the new requirements.
If the controlled system is driven out of the region of the validation of the
model, a new model should be acquired or the uncertainties should be
reduced.
Once the user goals have been stated, some decisions should be made to
design the controller:
• Which variables are selected as controlled, measured and manipulated
and which ones are going to be treated as disturbances.
• Based on this choice, which are the achievable performances?
• To meet the control requirements, which is the most suitable control
design approach?
Control Goals
From the set of usually unclear “user” control goals, mainly based on
qualitative and economical requirements as well as operational constraints,
the desirable controlled plant performances for control design should be
derived. They may concern different properties such as:
• Reference tracking, to follow the changes in the set-points or references.
• Control decoupling, to better understand and tune the different sub
processes or control variables.
• Disturbance rejection, to cope with non-manipulated external variables.
• Measurement noise rejection, to be able to use “imperfect” sensor and
transmission systems.
• Robustness against changes in the plant (model) or expected
disturbances.
Control Structures
The choice of the variables to be used as control variables as well as the
information used to generate the control actions will determine the control
structure.
The control structures to be considered are:
• Open loop vs Close loop
• Single/Multi loops
• Two degree of freedom - The control action may be computed into two phases. Foremost,
the control error is evaluated, and the control is based on the error. This is a feedback control
action. Subsequently, an additional control action is computed based on external inputs. There are
two degrees of freedom to design the controller and the design can be split to achieve tracking
(references) and regulation (output feedback) performances.
• Multi level Control - In this control structure, groups of input (output) variables can be treated
jointly to control a process variable. They will act locally, receiving commands (set-points) from the
higher decision levels, and sending information back to these coordination levels.
Performance Analysis
There are several criteria for defining “good nominal performance” in terms
of the close-loop transfer matrices:
• Settling time – it is related to position of the system poles and its internal structure i.e.
controllability and observability.
• Overshoot – although this can be checked for particular input/output combinations, it
is not used during the design phase.
• Steady-state gain – position errors for reference tracking are established, in most
loops by the DC gain of the sensitivity function. For disturbance rejection, the steady-
state gain may be determined.
• Bandwidth – many control requirements may be cast in the frequency domain, in
particular, ability to track references varying up to maximum rate and ability to reject
disturbances whose frequency components are mainly concentrated on a particular
band.
Distinguishing between Classical
and Intelligent Control
The mathematical approach is used in developing a classical control
system, the designer must mathematically model the plant.
In developing an intelligent control, the designer doesn’t need to know the
model of the plant to be controlled, only the plant behavior is required. In
many cases, the plant may be too complex to be modeled.
For classical control, the designer must model the plant and therefore the
intelligence (or knowledge) lies with him. The intelligence is shifted
towards the designer.
For intelligent control, the software abstractly models the plant and
therefore the intelligence (or knowledge) lies with it. The intelligence is
shifted towards the software.
Classical Control Methodology
Solve the control problem by doing the following steps:
Develop a model of the plant to be controlled.
Design a controller using the mathematical model.
Study plant performance using the mathematical model of the closed loop
system.
Implement the controller and evaluate the performance of the closed loop
system (again, possibly leading to redesign).
Methods for Intelligent Control
No System Modeling.
The designer only has to input the appropriate stimuli to the intelligent
control and evaluate it on its output. The intelligent control itself develops a
model of the system to be controlled.
Humans can perform complex tasks without knowing exactly how they do
them.
The types of intelligent control includes: fuzzy logic, artificial neural
networks, genetic programming, and others.
Definition of intelligent systems
Intelligence is a mental quality that consists of the abilities to;
learn from experience,
adapt to new situations,
understand and handle abstract concepts, and
use knowledge to manipulate one’s environment.
AI is defined as “the ability of a digital computer or computer-controlled
robot to perform tasks commonly associated with intelligent beings”.
Definition of intelligent systems
During the last four decades, researchers have proposed many model-
based control strategies that involve various phases such as modeling,
analysis, simulation, implementation and verification.
Zadeh has considered this decade as the era of intelligent systems “I
believe the system analysis and controls should embrace soft
computing and assign a higher priority to the development of methods
that can cope with imprecision, uncertainties and partial truth.”
Intelligent control is designed to seek control methods that provide a level
of intelligence and autonomy in the control decision that allows for
improving the system performance.
Hard vs Soft Computing
• Hard Computing: Hard computing uses traditional mathematical
methods to solve problems, such as algorithms and mathematical
models. It is based on deterministic and precise calculations and is ideal
for solving problems that have well-defined mathematical solutions.
• Soft Computing: Soft computing, on the other hand, uses techniques
such as fuzzy logic, neural networks, genetic algorithms, and other
heuristic methods to solve problems. It is based on the idea of
approximation and is ideal for solving problems that are difficult or
impossible to solve exactly
Hard vs Soft Computing
In hard computing, the prime objectives of the computations are precision
and certainty.
In soft computing, the precision and certainty carry a cost.
Therefore, it is realistic to consider the integration of computation,
reasoning, and decision making in order to provide a framework for the
trade off between precision and uncertainty.
The principal partners in such a consortium are fuzzy logic, neural
network computing, generic algorithms and probabilistic reasoning as
well as their integration.
Intelligent systems underlie what is called “soft computing.”
Hard vs Soft Computing
Soft computing tools are commonly used to enhance AI and incorporate
human expert knowledge in computing processes.
Their applications include the design of intelligent autonomous systems
(controllers) and dealing with unknown parameters.
Soft computing can:
– learn from experience,
– universalize into domains where direct experience is absent,
– can perform mapping from inputs to the outputs faster than serial analytical
representations by using parallel computer architectures that simulate biological
processes.
Intelligent control techniques that emulate characteristics of biological
systems offer opportunities for creating control products with new
capabilities.
Hard vs Soft Computing
Intelligent control has different tools for emulating the biological behavior
that could solve problems as human beings do. The main tools for IC are;
Fuzzy logic systems are based on the experience of a human operator, expressed in
a linguistic form (IF–THEN rules).
Artificial neural networks emulate the learning process of biologic neural networks, so
that the network can learn different patterns using a training method, supervised or
unsupervised.
Evolutionary methods are based on evolutionary processes such as natural evolution.
These are essentially optimization procedures.
Predictive methods are mathematical methods that provide information
about the future system behavior.
Design of ANFIS Controller
Outline the steps
1.
2.
3.
4.
5.
6.
Case Study:
ANFIS Controller for Grain Dryer
• Grain drying is an important stage of agriculture. The percentage wet
basis (%WB) needs to maintained at 14% or less during storage to avoid
damage. one way is to use a continuous drying technology. However,
the wet basis of the incoming grains are not known and can vary based
on the wheather during harvest.
• The system has been used for a few PhD studies. An ANFIS model is
already developed and available that relate the %WB at the end of the
conveyor and the time that the grains stays in the dryer (residence time).
• Propose an ANFIS control system for the dryer. Provide the structure of
the controller. Label all variables and indicate all assumptions that you
have made.
2
Further readings
1. Intelligent Control Systems Using Soft Computing Methodologies, By:
Ali Zilouchian & Mo Jamshidi, CRC Press, 2001, ISBN:0-8493- 1875-0.
Available online: [Link]
[Link]/book/593496/3c6d6e?regionChanged=&redirect=7850341
2. 2. Intelligent Control Systems with LabVIEW, By: Pedro Ponce-Cruz,
Fernando D. Ramírez-Figueroa, Springer, 2010, ISBN 978-1-84882-
683-0. Available online: [Link]
[Link]/book/812190/493ebf?regionChanged=&redirect=7855146
3. Taught Courses by Prof Kasim Mousa, Philadelphia University, Jordan