Intelligent Manufacturing € February € 1995 € Vol. 1 € No. 2
Intelligent Manufacturing € February € 1995 € Vol. 1
€ No. 2
Common to all manufacturing is the need to provide ongoing and
real-time support, monitoring processes to verify that all is in
order and diagnosing problems when they arise. These tasks apply to
discrete part manufacturing as well as batch and continuous
production.
The need for rapid response and enlightened observation provides the
motivation for better use of knowledge, and thus knowledge-based
systems. For this reason, manufacturing support has been one of the
dominant application areas for knowledge-based systems.
Knowledge-based systems enable computers to be broader and more
effective problem-solving tools; help users develop better designs
and better processes by making the best design and manufacturing
expertise readily available; and help users identify relationships
among design geometry, materials and production processes, and
leverage these relationships to provide the best combination of
product features, quality and cost.
The advanced computer-based information management technologies
called knowledge-based systems share a common goal: better
acquisition and application of knowledge. Knowledge is required to
design products and processes that are feasible, cost-effective and
functional. Once the product is designed and the manufacturing
process is under way, knowledge is used to analyze data, identify
trends and problems, and define appropriate corrective actions.
There are several distinct technologies within the category of
knowledge-based systems, and they meet the needs of manufacturers in
different way. Expert systems and fuzzy logic incorporate knowledge
in the form of rules. Neural networks actually discover knowledge in
the form of relationships hidden within data. These technologies have
provided manufacturers new yet proven tools, each with certain
strengths and limitations.
The strengths relate to the types of knowledge that may be captured
and the types of decision support provided. The limitations relate to
the resources and types of knowledge required to build a successful
system. To identify applications and select tools, manufacturers
should consider the nature of their applications and the information
available to support those applications.
Expert systems
Expert systems are the most established and recognized
knowledge-based technology. Expert systems capture human
problem-solving expertise in the form of rules. This approach is
useful for problems that are difficult enough to require special
training or experience and for which an expert is available to
identify concise and complete problem-solving rules. Ideally, the
expert has sufficient depth of experience to be able to look past the
obvious and provide insight in the form of hunches that provide
problem-identification shortcuts.
A user poses a problem or a situation to the expert system, which
then calls upon its internal understanding of the logic and issues
pertaining to that particular problem. A question-and-answer dialog
between the expert system and the user enables the expert system to
clarify the situation and ultimately provide the user a diagnosis or
recommended next step. The user may seek to better understand the
problem-solving approach or perform a reality test on the answer by
asking the system how it reached the given conclusion. The system
will respond with the step-by-step reasoning process based on the
specific rules used and the conclusions derived from these rules.
Expert system users are typically less experienced staff members.
They can use the system not only for decision support, but also for
training. The system can be an effective training tool because users
observe reasoning processes and applications of knowledge in specific
contexts. The decision support role of expert systems is useful for
both novices and experts. Novices draw upon the system's expert
knowledge to handle more challenging problems than they would
otherwise. Experts also benefit because they are available for the
more challenging or unusual tasks.
An expert system contains three basic elements. The expert's
knowledge is contained in the knowledge base. The second part, the
inference engine, contains the rules of logic and analysis that are
applied to information in the knowledge base. The third part, the
controller, manages the system, applying the inference engine
functionality to the appropriate knowledge.
The chances for a successful implementation are contingent on picking
the right problem (for instance, see "Expert Systems Solve Scheduling
Problems"). Some guidelines follow:
An answer exists and can be found by an expert in minutes to hours
based on rules that can be explicitly stated; the problem is neither
trivial nor one of common sense; the problem is deep in a particular
area rather than broad across a wide area; a clearly definable
benefit can be identified, with interested and supportive users; and
the system does not require ongoing maintenance to remain
current.
Fuzzy logic
Fuzzy logic is an increasingly visible knowledge-based technology.
After more than 20 years of benign neglect and overt criticism, it is
now increasingly viewed as a useful technology. It is incorporated in
commercial process controllers as well as a number of
application-specific controllers for consumer products. In fact,
fuzzy logic is becoming commonplace in Japanese consumer products.
Examples include camcorders, air conditioners, vacuum cleaners,
washing machines and automobiles.
Omron Electronics, a Japanese supplier of programmable logic
controllers (PLCs) and single-loop controllers, is a major proponent
of fuzzy logic for process controllers. Some of their PLC and
single-loop controller products incorporate fuzzy logic, which is
claimed to be particularly useful for the following types of
applications:
Fuzzy logic has found application primarily as a control
technology. In the applications for which it is best suited, the
benefits include easier programming because of a more natural
expression of control objectives. With temperature control, for
instance, the objective could be stated as "Keep the temperature
close to 70°." The fuzzy logic controller would implement
control actions of progressively smaller magnitude as the system
approaches the setpoint. This contrasts with the multiple rules
required in an expert system approach or the equations required in a
traditional approach to define the smaller control actions to be
taken as the setpoint is approached.
The stability of fuzzy logic-controlled systems cannot be proved
mathematically. In most cases this is of no particular practical
consequence. However, in safety-related and other critical
applications, this is a problem. In these cases, a necessary
implementation step is to perform exhaustive testing to verify system
performance under all possible circumstances.
An example is provided by a fully automated subway train with fuzzy
logic control that has been in operation in Japan since 1988. The
fuzzy logic controller is reported to make almost 70 percent fewer
errors in braking and acceleration than humans and to provide an
energy savings of 10 percent. But implementation was a challenge. The
safety implications are considerable in this control application. The
inability to mathematically prove control stability, i.e., to prove
that there is no possible scenario under which the control system
would respond inappropriately, required drastic measures. Observers
were on board for tests consisting of 300,000 runs! At this point, it
was agreed that safety had been sufficiently demonstrated, and the
system was commissioned.
Neural networks
Neural networks, like expert systems and fuzzy logic, share the
common goal of enabling computers to capture and apply knowledge, not
just manipulate numbers and letters. Neural networks have a number of
interesting properties: They can improve their own performance; they
can adapt; and they can discover relationships in data.
As an example, a common design problem for products made according to
a recipe is to find ways to vary the recipe, or proportions of
various ingredients, to provide new product properties. If a neural
network is provided recipe data along with the corresponding
properties data, it can discover the relationship between the two. In
this case, that relationship is a product formulation model that can
be used to design new or better products. The user need not know how
changes in recipe affect changes in properties. The user need only
provide data that captures that relationship.
This discovery capability provides an interesting alternative to
dependence on rules-based knowledge. Now, corporate databases can be
mined to discover knowledge, not just numbers. This knowledge
includes data such as marketing information, product designs,
manufacturing records and financial results. An opportunity exists to
discover interesting relationships among, say, how a product was
designed, how it was manufactured, how it performed functionally, and
how it performed as a source of revenue.
Neural networks are inherently adaptive, nonlinear and pattern
recognizing in nature. They learn to solve a problem by being shown
examples of situations and associated solutions. The network learns a
relationship that is valid between each of the sample situations and
the associated solutions. Upon finding this relationship, the network
is able to generalize to provide an appropriate solution to a new
situation.
In conclusion, technology trends will be an element in overall
business trends. Competition will remain fierce, and the successful
organizations will be those that respond rapidly and effectively to
new requirements and new opportunities. Information management will
become an increasingly critical competitive factor in a worldwide
marketplace. And knowledge-based systems will thus play an
increasingly visible and important role.
An expert system has improved overall forge shop operations by
improving scheduling. Scheduling, experience and logic are useful in
balancing constraints such as process capabilities and inventory
limitations, and in meeting a variety of scheduling objectives. These
objectives might include minimizing work in process, late orders, or
machine setups.
Ellwood City Forge (Ellwood City, Penn.) specializes in open-die
forgings of one quarter to 25 tons, generally with no two jobs alike.
To address the scheduling challenges of such a business, the company
adopted the Forge Shop Expert Scheduling System, incorporating
company policies, metallurgy knowledge and scheduling rules.
The system development goals were to achieve high quality and low
cost with reduced order turnaround time and improved efficiency of
material utilization. A secondary goal was to increase energy
efficiency. The system had to manage up to 2,000 orders in the
system, each of which could have unique materials and processing, and
many of which involve lot sizes of two or three pieces.
The following benefits were reported: Improved material yield and
energy efficiency; more accurate estimates of price and delivery; and
reduction of turnaround time from eight to less than three weeks.
These benefits were achieved through improved scheduling, as the
system schedules the melt shop, the forge and the furnace.
New business opportunities
Better scheduling capabilities have also enabled pursuit of new
business opportunities. Union Carbide Chemical and Plastics
(Charleston, W.Va.) has implemented a personal computer-based
scheduling expert system to handle the task of finite scheduling,
i.e., determining a production schedule that best accommodates finite
(limited) production capabilities. The development objective was to
enable rapid, yet realistic, responses to opportunities that would
periodically arise while rapidly producing products for export.
The decision on whether to accept an order had to be made within a
day. If accepted, these new orders would be inserted into the
existing production schedule, with some impact on the delivery dates
of the existing orders. The operational challenge was to accommodate
the impact of the added order without causing ship dates of existing
orders to exceed prior commitments. The scheduling challenge was to
rapidly determine whether this would be feasible before committing to
the new order.
When a constraint is violated by insertion of a new order, the expert
system helps schedulers determine which constraints to override to
meet demand. Schedulers may, for example, consult a screen that
combines a Gantt chart and expert system results. The Gantt chart
shows the time intervals and timing relationships for the various
manufacturing steps. The expert system window displays scheduling
rules and specific problem resolution advice to help users juggle
conflicting demands and, hopefully, identify a feasible approach.