Transient Surrogate Modeling for Thermal Management Systems Andrew Van Zwieten, * Gokcin Cinar, † Elena Garcia, † Jonathan C. Gladin, † Dimitri N. Mavris, ‡ Aerospace Systems Design Laboratory, Atlanta, GA, 30332 In typical multidisciplinary design optimization problems, with varying missions, aerody- namic data, engine data, Thermal Management Systems (TMS), and other parameters it can take upwards of millions of runs to cover the full design space which result in extremely large computational burden, reducing the effectiveness of the design process. The objective of this work is to create a technical approach, leveraging existing concepts in surrogate modeling and other relevant fields, for the creation of a Transient TMS surrogate model. This work utilizes Design of Experiments to intelligently sample a model’s design space, surrogate modeling to enable instantaneous predictions, and state space modeling to allows surrogates to carry pre- dictions forward. By leveraging these three components, a six step methodology was developed. This paper explains the developed methodology with an application on a notional Transient TMS. We first pre-process the time dependent data and possible correlations between input variables, then break down a set of input variables into a series of step functions which repre- sent input schedules in a fraction of the time. We create Artificial Neural Network models to predict the future response of a metric of interest using the current response of both the input step functions and the corresponding output. We finally test whether the surrogates which were created with step functions could be used to predict the future response of the metric of interest for a full time trace of a sample aircraft mission. We show that this methodology yields acceptable predictions for both the partial and full time trace, with a maximum error of 5% and 10%, respectively. I. Nomenclature ACS = Air Cycle System ATTMO = Air Force Research Laboratory Transient Thermal Modeling and Optimization DOE = Design of Experiment EFTMS = Engine Fuel Thermal Management System RPM = Revolutions Per Minute MDO = Multidisciplinary Design Optimization TMS = Thermal Management System u = Input Variables VCS = Vapor Cycle System x = State Variables y = Response pps = Pounds per Square Inch II. Introduction T  constraints need to be considered in the design process when solving Multidisciplinary Design Optimization (MDO) problems for aircraft systems. This will become increasingly important as future aircraft systems are projected to require much higher thermal demands[1]. MDO problems can take upwards of millions of runs to cover the full design space[2]. These high fidelity models reduce the effectiveness of the design process due to their extremely * Graduate Research Assistant, School of Aerospace Engineering, AIAA Student Member † Research Engineer II, School of Aerospace Engineering, AIAA Member ‡ S.P. Langley Distinguished Regents Professor and Director of ASDL, School of Aerospace Engineering, Georgia Tech, AIAA Fellow 1 brought to you by CORE View metadata, citation and similar papers at core.ac.uk provided by Scholarly Materials And Research @ Georgia Tech