MODEL DIAGNOSTICS AND DYNAMIC EMULATION: ENHANCING THE UNDERSTANDING AND IMPACT OF COMPLEX MODELS IN SYSTEMS ENGINEERING

This paper proposes and demonstrates sensitivity-informed model diagnostics as applied to constellation design (CD). Model diagnostics provide guidance on how high fidelity, computationally intensive, simulations can be simplified to yield substantial computational savings while minimally impacting accuracy. Moreover, current CD methods average performance at multiple locations across the globe over a year; preventing nuanced evaluation of systems, and the tailoring of design for specific applications. Model diagnostics discovered the most important inputs, times, and locations for analysis; highlighting key dynamics typically occluded by averaging. Model diagnostics benefits are demonstrated in this study with a specific example of guiding the creation of dynamic emulators, with significant potential for improving the computational tractability of design optimization.