Distributed Optimal Design

Optimal Design Distributed Optimization Beyond Systems Engineering

Every engineer on a design team makes strategic decisions for your company with each material selection and each dimension they enter into a CAD drawing. In the course of developing an aircraft, a ship, or a vehicle of any sort, tens of thousands, if not millions, of design decisions are made trading manufacturing cost, efficiency, reliability, and a dozen other measures of importance to your customers. These decisions will cumulatively determine your revenue and profit through the life of the product. Yet, none of the designers have any quantitative sense of the impact of their choices on price and profit -- thus, they have no guidance to help them make the best choice every time.

Design Optimization

Optimization is the art or science of finding the best. To optimize, one must have a set of alternatives (or a process for generating alternatives) and a method to measure the goodness of alternatives. Value Models evaluate the goodness of whole products. But what about engineers who want to optimize a component, such as a wheel on an aircraft, or a turbine in an engine?

Distributed Optimization

Optimizing components is of little value unless the combined effect of optimizing individual components is the optimization of the whole product. Distributed Optimization is a process by which the optimal solution to a large problem is found by decomposing it into smaller problems. When the smaller problems are solved, their solutions combine to form the optimal solution to the larger problem. Design for the Marketplace takes a measure of goodness for a complex product, based on the objective function, and uses it to construct measures for independent components. When the components are designed to be optimal according to their separate measures of goodness, the overall product is optimized according to its measure.

Design for the Marketplace uses scoreboards to communicate component objectives to component designers, and contract incentives to pass objectives to suppliers and partners. The incentives make sure that the most profitable design for the supplier is the best design for the overall program. Each participant is motivated to locally optimize their design, and DFM technology ensures that this optimizes to whole product.

Beyond Systems Engineering

In the 1950's the aerospace industry learned how to make truly complex products, such as intercontinental ballistic missiles, nuclear submarines, and reliable jet-powered aircraft. Each of these products is so complex that no one individual can comprehend it in sufficient detail to design the whole product. Thus, they are not designed by individuals; they are designed by organizations. The process of systems engineering has enabled organizations to design such complex products. The essence of systems engineering is that a product design is transformed into a hierarchy of design problems. Each problem is specified by a set of requirements. When a problem is too large to solve directly, it is decomposed into a group of smaller problems, each specified by its own requirements. The interrelationships between the smaller problems is designed such that if they are solved for each to meet its requirements, and the resulting components are interfaced together according to the design, then the larger problem is solved.

At each level of the design, the result of the design activity may include drawings (instructions to manufacturing or assembly) and specifications for lower levels. For example, the design of a lubrication system may include specifications for an oil pump, filter and tank; oil tubing; spray nozzles; and an assembly drawing to provide directions for putting the system together. The design is done correctly if, when each part satisfies its specification and when all the parts are assembled according to the assembly drawing, then the lubrication system necessarily meets its specification.

The key property of systems engineering is that, if each design group does its job correctly, a whole system can be created without any one group comprehending more than its own component. This property makes it possible for organizations using systems engineering to create products that are more complex than any individual can create. The systems engineering process has enabled our civilization to construct products that are barely short of miraculous: the Saturn V moon rocket, cellular telephone systems, jet airliners, Pentium microprocessors, and any current day automobile. However, it only makes these products possible—it does not provide any process by which an organization can design the best product. In other words it does not provide facility for optimization.

As an example, consider a hypothetical story about automobile designed to requirements, using target costing. Wind tunnel tests show that the body drag is higher that specifications, so a ridge is added to the trunk lid to act as a mini-spoiler, increasing the cost $100 and improving highway fuel economy by 1 mpg. Meanwhile, the exhaust pipe is over cost, so the diameter is reduced from 3" to 2" saving $25, although fuel economy is decreased 2 mpg. The net result of the two changes is $75 more cost and the loss of 1 mpg—clearly a change for the worse, even though the changes to the individual components seemed to be for the good, in the context of meeting requirements.

The traditional systems engineering approach, which places requirements on the design of every component, frequently leads engineers to work at cross purposes as each attempts to meet his or her component requirements. Design for the Marketplace is distinguished by the framework it provides for cooperation throughout a distributed engineering team.