Prototype and Model Development

There is a strong need to adapt the prototype CVDE described in this book to function in commercial applications settings that require access to com-mercial databases. Linked to this is a need for development or acquisition of more realistic and complete domain-specific models. Such advanced im-plementations would require a significant emphasis on developing graphical user interfaces and displays that help users participate in the decision-making process, and multi-criteria decision-decision-making methods that include human decision makers [Josephson et a/., 1998] will especially be critical to the success of such systems. Ongoing work at General Electric Research fo-cuses on the development of such interactive multi-criteria decision-making tools for complex problems [Cheetham, 2003].

9.2.6 Applications

The methodology presented in this book can have a fundamental impact on the principles and practice of engineering in industrial product develop-ment in the network-based distributed environdevelop-ment that is emerging within and among corporate enterprise systems. In addition, the conceptual frame-work of the approach to distributed decision systems presented in this book may have much wider implications for network-based systems ranging from intelligent agent-based browser systems, to enhanced consumer and busi-ness services, and intelligent search techniques in scientific and commercial databases.

The following discussion presents two application examples; one from integrated circuit design and manufacturing planning, and the second from decentralized air traffic management.

9.2.6.1 IC Design and Manufacturing

Integrated circuit (IC) design and manufacturing, is an area deemed critical to the sustained growth of an economy driven by advances in information technology and one consuming considerable resources.

The largest delay in bringing new ICs to market has been, and will con-tinue to be for the foreseeable future, the design and verification process.

This is particularly true of mixed-signal designs, Systems on a Chip (SOC) implementations and densely packed Application Specific Integrated

Cir-cuits (ASIC). Design reuse of complex macro-cells, also called Intellectual Property (IP) cores, is a growing response to the design and verification pro-cess in which previously developed and proven sub-chip modules are reused in new ICs. This concept extends the well-developed use of in-house design libraries to include complex macro-cells as basic building blocks, similar to the use of integrated circuits as basic building blocks for electronics prod-ucts. This is a significant advantage both for the IC designer, and with widespread implementation, the industry as well [Gutmann et at, 1999;

Gutmann, 1999].

In support of this trend, several commercial IC foundries have helped reduce the technological gap enjoyed by vertically integrated industrial lead-ers, and as a consequence, most IC designers have access to near state-of-the-art fabrication facilities. Licensing of fully verified complex macro-cells can thus reduce the time-to-market appreciably, and simultaneously re-duce new product risk. The advantage of such a design approach will lead to more standardized design and manufacturing processes among design groups and leading-edge IC foundries, and offset the IC designer mind-set of designing from scratch, which is a principal factor causing design de-lays. The time-to-market advantage of reusing IP cores will therefore be overwhelming.

In the anticipated design and manufacturing environment, IC virtual design becomes analogous to virtual design in the printed circuit assembly domain, where selection of IP cores is analogous to selection of packaged ICs, and selection of foundries is analogous to selection of assembly facili-ties. In this environment, most of the IC design effort would focus on the development of interconnects and "glue" circuits, while benefiting signifi-cantly from the reuse of complicated but fully verified basic building blocks.

A common unifying theme in these applications domains is that multi-ple logically interrelated decision resources (IP cores, their suppliers, and foundries) are physically distributed, and information for global decision-making using these resources is available from several network-distributed databases.

9.2.6.2 Decentralized Air Traffic Management

The current air traffic control system in place in the United States, Europe and other countries is based on a network of geographically contiguous sectors, one or more of which are intersected by en-route flights between any two points. Each sector may be further decomposed into a set of

Conclusions 149

smaller control air-spaces. Each such airspace is typically managed by an expert air traffic controller who monitors the numbers and trajectories of the various flights in their given airspace and coordinates the flights in their airspace in collaboration with neighboring controllers. Since controllers wish to retain as much control over flight paths as possible, aircraft are currently disallowed to pick arbitrary routes and are assigned to one of a few standard routes between an origin and destination. While this system has been in place for decades and is considered reliable and robust, the increasing demand for air travel is frequently causing the system to reach an overloaded congested state at several sectors. Congestion occurs when the number of active flights in a controller's airspace exceeds the number of flights that they can safely manage.

Current efforts at General Electric Research to alleviate this congestion problem include the development of methods to better predict congestion using temporal reasoning tools, and global optimization to realize optimal flight routes. Since a flight between an origin and destination is typically restricted to a few standard routes, the assignment of flights to routes such that the expected congestion is minimized is a discrete planning problem not unlike the class of problems discussed in this book. In fact, a suitable representation for solving this problem is similar to the representation for the printed circuit assembly planning problem discussed in this book. From a practical perspective it would be impossible to simultaneously optimize the trajectories of all flights in a given time window over the entire national airspace. In addition, since a given sector is tightly coupled to only a few other sectors in a time window due to shared intersecting flights, flight planning and optimization for the entire national airspace for given time window horizons is more efficiently performed in a decentralized manner.

The algorithms and architectures presented in this book may be readily applicable in this environment.

A p p e n d i x A