Interaction Effects in the Design of Computer Simulation Experiments for Architecting Systems‐of‐Systems

Abstract

ABSTRACTThis paper demonstrates the applicability and benefits of optimal design of experiments (ODOE) for architecting systems‐of‐systems (SoSs). All experiments depend on appropriate models for their design and analysis. In keeping with the sparsity‐of‐effects principle, a general linear model with main‐effects‐plus‐two‐factor‐interactions (MEPTFI) is assumed to realistically characterize the relationship between the SoS elements and the response of interest. The designed experiment requires additional simulations to account for the two‐factor interaction effects, and the main effects cannot be interpreted as defining the optimal solution. The optimal performance architecture is determined by optimizing the fitted MEPTFI model. The proposed method is applied to the problem of architecting a SoS to protect against small boat attacks. A statistical analysis confirms that the MEPTFI model has excellent predictive capability. It also proves the fallacy of the applicability of the standard Taguchi method for architecting SoSs by demonstrating that it gives misleading results because of the omission of active interactions. The ODOE method can be implemented using commercially available general‐purpose statistics software. Given its mathematical formulation, it can be generalized to solve systems engineering and architecting problems that require complex models beyond the MEPTFI model.