Increasing the modularity of system architectures is generally accepted as a good design principle in engineering. In this paper, we explore whether modularity comes at the expense of robustness. To that end, we model three engineering systems as networks and measure the relation between modularity and robustness to random failures. We produced four types of network models of systems — component, component-function, component-parameter, and function-parameter — to further test the relation of robustness to the type of system representation, architectural or behavioral. The results show that higher modularity is correlated with lower robustness (p < 0.001) and that the estimated modularity of the system can depend on the type of system representation. The implication is that there is a trade-off between modularity and robustness, meaning that increasing modularity might not be appropriate for systems for which robustness is critical and for those whose modularity estimate differs largely between each type of system representation.

This content is only available via PDF.
You do not currently have access to this content.