Quantitative validation of complex systems integration principles

Abstract

The integration of complex systems is an important aspect of systems engineering.Previous research derived six integration principles and qualitatively validated four of them using a data set of 14 systems. Of the two non-validated principles, one was determined to be confounded with two of the four validated principles and is hence not considered in this research. This paper describes the quantitative validation of the resulting five integration principles based on an expanded data set of 52 systems. This expanded data set is analyzed statistically, and the interactions between integration principles are also evaluated. This research quantitatively validates four of the five integration principles and identifies three principle interactions that are significantly related to integration success, solidifying validity of the principles, and identifying three cases where the principles interact that must be further explored.

Machine learning approaches for improving integration of advanced sensors on legacy aircraft

“…Grumbach and Thomas defined the Principle of Reuse and Evolvability that allowed for existing components to adapt to new applications with a small increase in system complexity and a large decrease in uncertainty 17 . Grumbach and Thomas further validated their results and found that success in component reuse is affected by adherence to hierarchic verification approaches and treating subsystem integration as part of full system integration 18 …”

“…17 Grumbach and Thomas further validated their results and found that success in component reuse is affected by adherence to hierarchic verification approaches and treating subsystem integration as part of full system integration. 18 Other researchers have explored reinforcement learning algorithms more generally to improve their performance against complex problem sets. Gupta, Roy, and Dutt compared Instance-Based Learning (IBL) and Q-Learning algorithms in a dynamic test environment and found that IBL is more effective for dynamic environments.…”