Performance Prediction of Cyber-Physical Systems Product Lines in Dynamic Environments

Evaluating performance properties across highly configurable Cyber-Physical Systems (CPS) remains a significant challenge due to the size of configuration spaces and the cost of domain-specific simulations. Computing the fitness of a configuration is itself complex, as CPS behaviors depend on non-linear, system-level interactions that are difficult to model and predict. Although sampling and prediction techniques have shown strong potential in software domains, their extension to CPS contexts involving physical simulations remains largely unexplored. In this work, we highlight this challenge through a CubeSat product line case study and provide preliminary results combining simulation and machine learning to predict system-level metrics. Our findings, based on baseline prediction accuracy, illustrate the difficulty of the problem and motivate further research on performance prediction methods for CPS product lines.