Vapor compression system data-driven surrogate models for aircraft Environmental Control Systems

Abstract

Commercial aircraft manufacturers are actively working to develop more efficient Environmental Control Systems (ECS) in line with the More Electric Aircraft (MEA) concept. Novel ECS architectures may integrate supplementary cooling units based on Vapor Compression Systems (VCS), which provide more efficient cooling compared to traditional Air Cycle Machines (ACM). The design, optimization, and evaluation of these new ECS configurations rely on numerical simulations. Given the inherent complexity and computational demands of VCS, this study explores the use of data-driven surrogate models for efficient simulation of these systems. The main objective was to develop surrogate models for the VCS cooling unit, designed as steady-state equivalents of physics-based models, which have provided the training and testing data. To achieve this, we developed data-driven surrogate models using Gaussian Processes (GP) and Multi-Layer Perceptrons (MLP), both trained on outputs from high-fidelity simulations. The results show that GP models perform best when the output data exhibits smooth, continuous behavior, while MLPs are better suited to capturing complex, nonlinear behavior. This complementary performance highlights the strengths of each approach depending on the nature of the output data. Once model accuracy is verified, the VCS data-driven surrogate models were seamlessly adapted to be used under the Modelica/Dymola environment to comprehensively assess their accuracy and computational performance. The results demonstrated a significantly lower CPU time consumption when comparing physics-based models to data-driven twins across various scenarios, including VCS steady-state conditions, ECS steady-state conditions, and ECS dynamic conditions.