In simulation-driven product development, evaluating the signal response behavior of signal transmission devices often depends on repeated electromagnetic simulations, which become increasingly time-consuming as design complexity grows. This creates a need for data-driven methods that can support faster and more efficient early-stage design evaluation. To address this challenge, this paper proposes a design knowledge–powered Graph Attention Network (GAT) framework for predicting frequency-domain signal response intensity from product design representations. The proposed approach transforms CAD models into hierarchical knowledge graphs that capture assembly structures and geometric dependencies and uses these graph-based representations together with simulation results to train a predictive model for multi-curve response estimation. The framework enables efficient and structure-aware prediction of signal response behavior, providing a practical surrogate modeling solution for accelerating design exploration.

Keywords

Knowledge Graph, Design Model, Graph Attention Network, Sign Transmission Device, Performance Prediction