A hybrid agent approach for set-based conceptual ship design through the use of a fuzzy logic agent to facilitate communications and negotiation.

Ship design, like many engineering problems, involves both objective knowledge such as mathematical models and subjective knowledge such as expert opinion or design requirements. Fuzzy set theory is a method of systems analysis that bridges the gap between traditional methods of analysis and the real world relationships between variables that are ambiguous. This dissertation focuses on a set-based design approach to the cross-functional design team (Integrated Product Team) activity in early ship design. A hybrid agent analogy with human agents aided by intermediate computer agents is utilized. A new implementation of a fuzzy system has been developed to enable these designers to communicate their preferences over a range of parameter values. The system involves a new input sequence compared to classical fuzzy systems and improves scalability through the use of meta rules. The new hybrid agent fuzzy system is unique in that the inputs are the membership functions mui(x) which are changed by the human agents to express their individual preferences for a particular design parameter as a design progresses. The fuzzification, linguistic fuzzy rule bank, fuzzy inference, and defuzzification components are typical. The input membership functions and the output variable span the set-based range currently under consideration from the parameter's lower limit to its upper limit. The output is, however, a crisp value function of the design variable indicating the joint preference of the team of human design agents over the current set of that variable. A new fuzzy logic software agent to enable hybrid agent design team communication and negotiation has been developed in JAVA(TM). A series of pair comparison design experiments in which non-collocated student design teams designed a small feeder containership using the new agent software or a chat window environment have been completed. The goal of the pair comparison experiments was to eliminate experimental biases and give an accurate evaluation of the fuzzy logic approach. The results and analysis of the experiments have demonstrated that the fuzzy logic software agent does enable set-based design thus increasing the probability of achieving a global optimum when compared to conventional team-based design systems.Ph.D.Applied SciencesOcean engineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/123704/2/3096201.pd