Mixed Continuous Variable and Catalog Search Using Genetic Algorithms

In this paper a general design and catalog representation is proposed and implemented. The representation is applied to mixed continuous variable and catalog search using genetic algorithms. The representation addresses a number of common catalog search design scenarios. In addition to catalogs of static data (e.g., different sizes of pipes), it is possible to model/search subdesigns that can provide performance data based upon operating conditions. The use of catalog hierarchies allows the search to simultaneously consider catalogs from different vendors and at different levels of detail, therefore achieving the capability of modeling solutions containing general information. The capabilities of the representation are demonstrated through an object-oriented computer implementation that uses a genetic optimization and search algorithm. INTRODUCTION In practice, solving many product design problems involves choosing elements from catalogs and combining them with unique custom components. Therefore, methods to search for optimal solutions to such problems, which involve both continuous and discrete variables, are of practical importance. From an automated search perspective, however, these combinatorial problems tend to have undesirable characteristics. In this paper we will present a general catalog architecture and search methodology for this class of design problems. 1 Please address correspondence to this author: phone (617)253-2655, fax (617)258-9637, [email protected] We have frequently encountered a number of catalog selection scenarios when designing products. Perhaps the most obvious case involves choosing elements defined as static collections of data. For example, an element in a catalog of standard pipes would include an inside and outside diameter, a cross sectional area, a second moment of area, etc. Each element is described simply by a set of static data. However, one can imagine more complex cases where catalogs should not be limited to static data. Such catalogs might include different types of manufacturing processes or electric motors. Consider a catalog of motors. An element in the catalog ( ie , a specific motor) will certainly have static data, such as its geometry. Other characteristics, such as the operating torque or speed, must be calculated using motor specific models and operating conditions governed by other aspects of the design. Even the motor's cost will be a function of purchase volume. Ideally, each motor in the catalog should be able to receive operating conditions from the design and calculate its corresponding performance. Thus, some catalogs elements should contain embedded models in addition to static data. Furthermore, when choosing a catalog element, the component may add new requirements to the design. For example, imagine that we are choosing a thermocouple from a catalog. The design requires a certain accuracy and cost. On the other hand, each thermocouple in the catalog will require the design to operate within various temperature ranges. Therefore, catalog elements should be able to add new requirements to a design. Often, we want to choose a catalog element, but we are not sure which catalogs to use. We may need to choose a motor but are not sure whether the catalog from vendor X or vendor Y is