An exercise in design process simulation using agent models based on Bayesian global optimization

The frequent schedule delays and cost overruns in the design of modern complex engineered systems is motivating current research to provide improved understanding of design process dynamics. A particular interest is to understand the aggregated decision making behavior of multiple individual designers engaged in design tasks associated with multiple disciplines, subsystems, or components. In this context, this paper focuses on defining and implementing an agent model and multi-agent coordination schemes to enable simulation of the design process in an engineering firm. This research represents an initial step by the authors toward the overall goal of simulating the fundamental features of the activity of designing complex engineered systems. First, a Bayesian agent model able to model designer preferences and uncertainties in designer knowledge is described. Next, three coordination schemes inspired by MDO architectures are formulated. The agent model and the coordination schemes are then applied to demonstrate the framework with a simple example problem. Results show that the approach represents a promising avenue for future extensions and follow-up studies of more complex and realistic design process structures.