Application Domain

Design and manufacturing of printed circuit (PC) assemblies is typically realized in three distinct but highly coupled stages. First, a design (string of parts) and suppliers for the parts constituting the design are specified along with a specification of the layout of the parts on a printed circuit board. Next, a fabricator of the printed circuit board is selected. Finally, a manufacturing facility that can assemble the product given a part string and circuit board, is selected. These stages are highly coupled for decisions made at the design stage have maximum impact on product cost, product realization time, and manufacturing, and manufacturing choices have max-imum impact on design, product cost, and product realization time. Design and manufacturing decisions affect choices of parts that are assigned to a design, selection of suppliers who will supply the parts, and selection of manufacturing resources that can produce the design.

Traditionally, however, design and manufacturing tasks are executed se-quentially with little or no interaction between the phases, and choices are typically made using the experience base in an organization. This results in numerous time consuming iterations and design-supplier-manufacturing solutions that are not anywhere close to optimal in terms of time, cost, or a trade-off function of both time and cost. Moreover, as suppliers and manufacturers increase in number, the combinatorial complexity of opti-mal selection grows rapidly. An integrated planning method thus assumes importance in this environment as a powerful means to improve competi-tiveness of products. Such planning methods respect the inherent coupling among the decision stages, and seek to select those plans that result in lower overall cost and lead-time to realization.

Figure 7.1 shows the nature of the integrated design, supplier, manu-facturing planning problem for printed circuit assemblies, a formal model

for which is presented in Section 3.3. In this domain, parts, suppliers, and PC Board Fabricators PC Assemblers

Parts Library Parts Suppliers

Fig. 7.1 Nature of the integrated design, supplier, manufacturing planning problem for printed circuit assemblies. Lines with arrowheads indicate assignments. Identical parts in various designs have solid lines between them.

manufacturing decision resources are logically interrelated, and information for global decision-making using these resources is available from multiple network-distributed databases.

7.2.1 Configuration of the Networked Environment

Figure 7.2 shows a high-level configuration of the networked environment that consists of several logical clusters of network nodes and a product design node. Nodes in a logical cluster correspond to a class of functionally equivalent entities, and in general are physically distributed over a network.

In this environment, three logical clusters of network nodes are considered:

• Parts Distributor Nodes: Each node in this cluster corresponds to a parts distributor (parts warehouse) that stocks parts from several

man-Coevolutionary Virtual Design Environment 99

Logical Cluster of Parts Distributor Network Nodes

Product Design Node

Logical Cluster of PC Board Fabricator Network Nodes

Logical Cluster of PC Assembly Network Nodes

Fig. 7.2 High-level configuration of the networked environment for integrated printed circuit assembly planning.

ufacturers.

• Printed Circuit Board Fabricator Nodes: Each node in this cluster cor-responds to a printed circuit board manufacturer that may have several alternative board manufacturing lines, each of which is capable of man-ufacturing printed circuit boards.

• Printed Circuit Assembly Nodes: Each node in this cluster corresponds to a manufacturing facility with alternative manufacturing lines, each of which is capable of manufacturing printed circuit assemblies.

Each of the nodes in the above three logical clusters has a locally res-ident database that stores information specific to the node. A parts dis-tributor node's database stores information on parts and their functionally equivalent alternatives, part characteristics, their costs and lead times for availability. A printed circuit board fabricator node's database stores infor-mation that reflects the capabilities of the corresponding fabrication lines, and a printed circuit assembly node's database stores information on the available manufacturing resources within lines, their capabilities, process costs and delays.

In addition, there is a product design node that generates functional specifications that serve as partial templates for virtual designs. These templates specify the required parts equivalence classes (module types) and their respective instances for realizing a specific printed circuit board prod-uct.

While the search at a parts distributor node is over the space of func-tionally equivalent designs that correspond to a functional specification, and is achieved by selecting alternative parts and suppliers for those parts, the search at a printed circuit fabricator node is over the space of available board fabrication resources, and the search at a printed circuit assembly node is over the space of available assembly resources.

7.2.2 Application-Specific Assumptions

This section lists several assumptions specific to printed circuit assem-bly planning that influence the organization and implementation of the decision-making environment.

It is assumed that for each module type listed in a functional specifica-tion, there are multiple part alternatives that can satisfy the requirement.

Part alternatives may differ not just by their supply sources, costs and lead times, but also by their physical characteristics. A commonly occurring physical variation is packaging, whereby the same part may be available in multiple package variations. This type of variation has a significant im-pact on manufacturing processes. Another commonly occurring variation is through function subsumption, whereby a given function occurring in a cer-tain part occurs similarly in its part variant along with additional available functions. This type of variation has the potential to affect the physical size of a part, which in turn has consequences in manufacturing. The important issue is that variations in part characteristics and their supply sources have coupled and nonlinear affects on manufacturing, product cost, and product realization time.

An important assumption is made regarding the availability of a printed circuit layout generator. Ideally, this generator would take as input a parts string and a specification of the electrical interconnectivity between the parts, and output a layout on a printed circuit board. However, the discus-sion in this book assumes the existence of a model-based system that can identify the principal characteristics of the resulting printed circuit board given a design.

Another important assumption is made regarding the manufacturability

Coevolutionary Virtual Design Environment 101

of printed circuit board products. It is assumed that any generated printed circuit board is manufacturable at any of the available fabrication lines, and printed circuit assemblies can be assembled at any of the available assembly lines. The rationale is that rather than generating infeasible production options, production options are allowed to incur cost and time penalties consistent with the degree of infeasibility, through the objective function.

A single functional specification is used as a starting point to search over the coupled space of equivalent designs, printed circuit board fabricators, and printed circuit assemblers. However, in principle, one could implement an augmented search that not only searches over equivalent designs cor-responding to a single functional specification, but also searches over the space of functional specifications itself. To realize such a system one would require a mechanism (application) that resides at the product design node and generates a space of equivalent functional specifications.

A functional specification is an abstraction at the level of part module types. If it is possible to generate equivalences at a higher level of ab-straction (sub-system level), one could implement a much broader search.

An example is used to illustrate this point. Assume that a sub-system in a printed circuit board requires one instance of a specific integrated circuit to perform a function. Let there exist a sub-system level alternative wherein the same function can be realized using purely discrete components (resis-tors, capaci(resis-tors, transis(resis-tors, etc.). Then these two sub-system alternatives are considered to have equivalent (not identical) functional specifications from the perspective of the module types they encompass.