Intelligent Manufacturing € March € 1995 € Vol. 1 € No. 3

What Makes Manufacturing Intelligent?

By David Blanchard
Editor

The First World Congress on Intelligent Manufacturing, held last month in Mayaguez, Puerto Rico, opened with a particularly interesting question: What exactly is intelligent manufacturing?

Most of the conference participants, who were primarily researchers and developers at various R&D; centers, wrestled with defining what seems on the surface to be a simple concept.

The first definition came from Nam Suh, a researcher at MIT. "The goal of intelligent manufacturing," according to Suh, "is satisfying customer needs at the most efficient level for the lowest possible cost." Almost immediately after the words were out of his mouth, reaction came from the other conference-goers, which could be summed us thusly: "All well and good, but that's the goal of just-plain-old manufacturing as well. Where does the intelligence come into play?"

The answer, it was agreed, was in the involvement of computers. However, given that the concept of computer-integrated manufacturing (CIM) has been with us for more than 20 years, the incorporation of computer technology alone does not necessarily result in intelligent manufacturing. It is the introduction of human-like decision-making capabilities into the manufacturing system that makes it, indeed, intelligent. In particular, knowledge-based systems will dominate the manufacturing landscape of the late 1990s much as CIM was all the rage in the 1980s, and flexible manufacturing systems in the 1970s.

In fact, according to Vladimir Milacic of the University of Puerto Rico and the conference's organizer, manufacturing technology will play an important role in future human developments if it is based on new knowledge-based platforms. That's a big "if," though, and much of the conference was spent debating exactly what impact knowledge-based systems will have on manufacturers (see last month's article on knowledge-based systems).

"An organization must take full advantage of the knowledge at its disposal," noted Steve Kim of Lightwell (Charlottesville, Va.). "This goal translates into the effective use of knowledge ranging from design to production and maintenance. To this end, the knowledge captured in disparate modules in the organization must be liberated and directed synergistically to support integrated systems for engineering and fabrication."

Unfortunately, there is an overall lack of awareness about new tools and techniques among manufacturers, noted Tor Guimares of Tennessee Technological University. Too often, manufacturers focus on using existing tools, rather than investigate the benefits of knowledge-based system (sometimes called expert system) technology. Expert system vendors can help the situation, Guimares said, by improving their shells to fit defined requirements for successful manufacturing applications. While there have been some off-the-shelf applications aimed at manufacturers, the marketplace is far from saturated with quality products.

Elaborating on his earlier definition, MIT's Suh explained that "intelligent" manufacturing can be achieved in three basic ways:

• Existing manufacturing processes can become intelligent by monitoring and controlling the state of the manufacturing machine;
• Existing processes can be made intelligent by adding sensors to monitor and control the state of the product being processed; or
• New processes can be intelligently designed to produce parts of the desired quality without the need for sensing and control of the process.

An intelligent manufacturing process, Suh concluded, has the ability to self-regulate and/or self-control to manufacture the product within the design specifications. Taking the concept a step further, Milacic boldly predicted factories of the future, where products are manufactured in an artificial life environment. In short, the World Congress was as lively for its discussion of what exactly "intelligent manufacturing" implied as it was for the presentation of systems, processes and applications.


Implementing an Expert System
At the University of Texas at Austin, Rockson Huang and Davor Juricic have been working on adding intelligence to a computer-aided drafting system. According to Huang and Juricic, recent developments in feature-based solid modeling techniques for design representation and AI applications have shown a potential to substitute expert systems for some decisions currently handled by a designer's knowledge of design rules and practices.

As part of their approach to implementing an AI solution, Huang and Juricic arrived at a formula which, though specific to their drafting and design application, reflect the basic structure of an expert system:

1. Select an existing software that will first be augmented by an expert system, and then integrated with the task solution software.
2. Collect the trade standards, heuristic rules, past experiences, and recommendations from books, experts and standard codes.
3. Organize a complete set of rules governing this task, and describe them in a formal way acceptable to a list-processing software.
4. Generate a list of features and other part attributes called for in the rule set.
5. Generate relevant external data by extracting the features and other necessary part attributes from the model data of an object.
6. Organize the rules governing this task into a knowledge base.
7. Implement an inference engine as a reasoning mechanism to find a solution based on the knowledge base and external data.
8. Incorporate the results into the model and its representation to obtain a final solution.

A Look at Some Applications
Long after the debates over terminology are forgotten, the applications will remain, and the World Congress shone an international spotlight on some of the intelligent systems that have been developed for manufacturers. For instance, at the University of Trondheim (Norway), an interactive tool for short-term production scheduling has been developed. The system features a graphical interface that allows the operator to interactively control the schedule generation and see the influence on key parameters.

"A constraint management subsystem checks that the current schedule agrees with the restriction imposed from the production environments, and if not, the reason for the conflict is identified and presented to the planner," explained Johan Haavardtun and Dag Kjenstad, the system's developers. This knowledge-based system is able to effectively react to unexpected events or delays. Neural network techniques come into play, repairing an inconsistent schedule gradually toward a consistent or an optimal schedule.

Case-based reasoning is an integral part of the approach being taken at Ohio University to interpret the process parameters involved in metal forming. Human experts need to be consulted to properly interpret analytical results of the forming processes, providing an opportunity to improve on the procedure with an expert system. Case-based reasoning represents knowledge as "cases," i.e., examples of past problems and their solutions. Applications are said to improve with use because the expert system will "learn" from past cases.

The Ohio U researchers developed a Windows-based intelligent design and analysis software package, which they applied to a pack rolling application. The expert system produced optimal design parameters without violating any material or machine constraints, while providing a better methodology for the design and analysis of metal forming processes.

A joint effort between the Italian universities of Perugia and Trieste is using genetic algorithms for optimal assembly planning. Genetic algorithms, sometimes referred to as adaptive computation, are based on the evolutionary concepts of natural selection and survival of the fittest. In simple terms, genetic algorithms generate new rules to replace the least useful rules already in place. These software tools allow users to solve complex problems, such as scheduling a large number of conflicting tasks, finding the shortest route that connects a number of locations, or streamlining communications networks.

The Italian researchers are using a genetic algorithm to optimize the search routine used in assembly planning, with the goal of improving the assembly process of mechanical products, thereby minimizing time and cost. Applying the genetic algorithm to an automotive water pump assembly plan, a relational graph that took more than an hour to produce using conventional means required only a few seconds with the genetic algorithm.

While most of the systems and applications discussed at the World Congress on Intelligent Manufacturing are probably destined to remain inside the halls of academia, it is important to realize that the students working on these projects today will be the manufacturing managers of tomorrow. Furthermore, the technological achievements presented at the conference will inevitably find their way to industrial shopfloors, either at your own facility... or at your competitor's.