One of the most critical issues for monitoring and managing the health status of various systems, including space systems, is the assessment of performance degradation and the prediction of remaining useful life. However, estimating remaining useful life with highly accelerated life testing methods is very challenging and costly, and is practically infeasible for unique and expensive systems. Given the common degradation trends in model-based performance–requirements systems, this article proposes a framework based on degradation trend that can, using monitoring data and with informed of the physics of degradation, provide confidence in the actual system performance under real operating conditions. The framework comprises a complementary model consisting of two configurable neural networks. The first network uses dynamic weighting to model the effects of influential parameters on degradation, while the second network, informed by discovered relations, provides predictions for the degradation trajectory. The proposed framework, by employing dynamic weighting, addresses the data-scarcity problem and the lack of knowledge about dynamic system interactions. Through a system-level study on environmental condition data and load profiles of compressor-turbine performance, the article demonstrates the effectiveness of the proposed approach, showing accuracy improvements in predictions of up to 99% and outperforming neural networks that lack dynamic weighting and do not uncover system physics. By applying the proposed method, the remaining useful life and the degradation trajectory of space systems and their health status can be effectively assessed, enabling the optimization of their design and fabrication.