Design of Experiments: Applications in the Simulation Profession
(Room S320B)
26 Nov 18
12:45 PM
-
2:15 PM
Design of Experiments (DOE) provides new types of information to modeling and simulation developers and users. DOE is used to calculate relatively accurate models of a system quickly, identify the most significant inputs (factors) that affect the outputs (responses), find optimal solutions, and calculate settings to meet target values. The models of the systems are sets of equations that determine the relationships between the responses and the factors. These models can be used to characterize (calculate) the response values anywhere in the region modeled. The benefits of using DOE include a thorough upfront analysis process; a wide variety of possible designs that can be used; a straightforward way to estimate needed sample sizes; development of equations that can be tuned to optimize or otherwise hit targeted values; and a wide array of use cases. The use cases for DOE include simulation, training, education, test and evaluation, systems engineering, consumer product design, quality improvement, cybersecurity, model validation, and human factors design information. All of these use areas will be mentioned in the briefing and several will be included in the example case studies in this tutorial.
This tutorial will discuss the upfront analysis steps for the DOE process, key benefits of using DOE, and current use cases. These use cases include the development of functional models of systems or processes in order to characterize how the systems or processes perform within the region modeled. The tutorial will then provide examples to illustrate how DOE models are developed and used to define a relationship between factors and responses for the purpose of analysis, tradespace studies, evaluation, and optimization. Response surface graphs will be used to illustrate how human-systems integration issues can be detected. Use cases such as Model Based Systems Engineering, education, training, cybersecurity, and validation of models will be mentioned. There are no prerequi