Report 1

Expert Systems
By
Ahmed Hussein Hosny
Delivered to
DR/ Walaa Gabr
Intelligent Control
Control Major
Electrical Engineering Department,
Benha Faculty of Engineering,
Benha University
Introduction:
Expert systems have been successfully used in engineering, as well as in many other applications, as a tool
to incorporate in a decision system the knowledge from an expert in the field. It has also been the case in
control engineering, where a number of applications using expert systems to implement control tasks
have been reported (see Expert Control Systems: An Introduction with Case Studies). This knowledge,
usually expressed as a set of behavioral or reaction rules, requires a sort of reasoning to produce the
output. Nevertheless, in most control applications, time is a critical issue and the controller should react in
real time; that is, producing the control action within a limited time interval. This requires the avoidance of
reasoning loops, leading to unpredictable computation time. Thus, expert systems in control should have a
well–defined, structured knowledge, with an efficient reasoning engine and always must provide a correct,
or at least safe, output. Control applications can be differently time demanding. For instance, in fault
detection or maintenance applications, time is not so critical. In general, in supervisory activities, the
control block uses long term data and information and a longer reasoning time is allowed. The local
control is assumed to be fast enough, and it copes with the normal disturbances and demands over the
process. In this case, the use of expert systems has been very fruitful.
Expert Control System:

As it is well known, fuzzy logic allows one to deal with approximated structured knowledge, but it will also
provide the framework to combine numeric and heuristic data. The intrinsic reasoning feature of fuzzy
systems is very appropriate to implement expert systems, as the expert knowledge is expressed by a set of
rules and facts easily translated into premises and fuzzy rules. A fuzzy system can be composed of many
fuzzy subsystems, each one of them implementing a partial activity. In this way, the structured fuzzy
system can be easy to understand and modify, and the execution time can be tracked, under different
operating scenarios. Moreover, a fuzzy expert control system may incorporate some traditional modules,
such as classical controllers or processors, integrating the whole control system operation. Based on the
conditions above, one of the key properties of an expert control system is to have a complete and
validated knowledge base. In order to define the required knowledge and describe the available tools to
validate it, let us review some basic concepts involved in any fuzzy expert control system. As already
pointed out in a previous chapter, dealing with fuzzy systems has some advantages but also some
drawbacks that should be taken into account. On one hand, the possibility to handle approximated
knowledge, the emulation of the human reasoning, or the capability to integrate different kinds of
information can be considered among the advantages. On the other hand, in defining a fuzzy system there
are so many options and alternatives that it is difficult to justify the optimality, or even the adequacy, of a
concrete selection of options including: the number and parameters of the linguistic variables, the shape
and parameters of the different membership functions, the kind of connectives and operators, as well as
the fuzzification and defuzzification approaches. In some other cases, the way these options are selected
leads to fuzzy systems that simply emulate other conventional systems, like many PID fuzzy controllers.
Another frequent situation is that, as a result of applying classical optimization tools to determine the
optimal parameters of the system, the final system has lost the basic interpretability feature which should
be in the basis of any fuzzy system.
The basic feature of an expert system is the ability to obtain some conclusions, from a set of facts, by
reasoning on previous knowledge. Data from the environment is converted into conditional statements
representing evidences, conditions or premises. Previous knowledge is expressed by means of reasoning
rules, that can be inference or production rules. Each rule expresses a partial knowledge connecting two
sets of facts, the premises and the consequents. This allows for the connection of rules, in such a way that
conclusions from a rule are used as premises for others. The reasoning can be forward, from data to
results or vice versa. Confidence factors are attached to each fact and rule, to express the uncertainty in
the knowledge. In general, expert systems only deal with a narrow knowledge area, to avoid the
uncontrolled nesting in the reasoning process. However various expert systems can cooperate in the
solution of a problem. There are also many options in defining an expert system, the key feature being the
knowledge represented in the rule base. It is composed of many atoms (each one of the rules), and the full
set of rules should have some properties, such as coherence, completion, and so on. Classical expert
system can provide many options as a result of the reasoning, leaving the final conclusion to the user. This
provides a richer result with alternatives to be considered in the actual application. But an expert system
to be used in control applications must fulfill some requirements. The control experts’ work is based on
their knowledge about the process, the control system, the goals and the constraints. Altogether this
conforms the expert knowledge and based on data from the actual plant, the pre-established reasoning
leads to some conclusions. These conclusions, based on data taken from the process and some other
external information, determine some actions that should be applied to the process. Thus, some basic
control-specific requirements should be added to the classical expert systems:
• Forwarding. The reasoning should proceed forward, from data to actions.
• Real time. The reasoning conclusion should be available at the predefined deadline, to be applied to the
system. In some cases, (soft real time conditions), this requirement can be relaxed or back-up actions can
be applied.
• Numeric data. The expert system must receive and send information from and to the process. This
information is numeric or logic. Nevertheless, this data should be fused and combined with other data
coming from the operator or other systems, probably of a heuristic nature.
• Clear interpretability. For security and confidence reasons, the expert system should be accepted by the
operator who will interact with it during the execution.
• Completion. A control system can not be without an action under certain circumstances. Thus, the
expert system knowledge should be complete and the reasoning system should provide a conclusion for
any input pattern.
Fuzzy expert control systems (FECS) combine the features of fuzzy controllers and expert control systems.
Facts are derived from data taken from variables that is fuzzified by application of membership functions.
This representation is flexible, intuitive and interpretable. It also allows for combining physical data with
heuristic and qualitative information, since the reasoning is performed over a unique type of variables,
namely fuzzy variables. A fuzzy expert control system easily fulfills the requirements of an expert control
system. The reasoning is forward and the looping should be avoided. In these conditions, and if the
number of operations is bounded, the reasoning will lead to a conclusion within a time interval, the
deadline, to forward the action to the process. The option of merging different kinds of data is typical in
fuzzy systems. Also, the interpretability of the reasoning is one of the key properties of a fuzzy system that
should be highly preserved when implementing a fuzzy expert control system. A modular construction
allows the design of different subsystems for different operating conditions, goals or constraints. Again,
the issue is the quality of the knowledge as well as its distribution and coordination. Another feature a
fuzzy expert control system may have is the learning capability, quite difficult to implement in a traditional
expert system. Learning is a step forward in the concept of adaptation. It involves not only changes in the
parameters or in the structure of the control system but in creating or deleting new/old activities in order
to accommodate to situations which were not completely defined and foreseen at the initial stage. It can
be considered as the facility to either incorporate new rules in the current KB, based on the experience
garnered under the operation of the system, or to delete those rules less used by the system, due to time
or memory limitations, according to the rules less used by the system. Tools should be provided to
guarantee the validity of the time-varying knowledge.

Report 1

Uploaded by

Copyright:

Available Formats

Report 1

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Report 1

Uploaded by

Copyright:

Available Formats

Expert Systems

Ahmed Hussein Hosny

Expert Control System:

• Forwarding. The reasoning should proceed forward, from data to actions.

You might also like