Electromagnetic relays are critical in aerospace and military systems, affecting the safety and stability of applications like aircraft control, satellite communication, and missile launchers. However, the scarcity of degradation data and complex variations in contact resistance pose challenges. Traditional methods often struggle with small samples. To address these issues, we propose a novel framework integrating TimeGAN with a CNN-LSTM-Attention model. TimeGAN generates synthetic degradation data that aligns with the statistical distribution of the original dataset, mitigating data scarcity. Data quality is evaluated using PCA and KDE. The CNN-LSTM model captures multi-scale temporal features, while the attention mechanism highlights critical features to improve contact resistance prediction accuracy. Experimental results show that the proposed framework outperforms traditional methods, demonstrating robust performance even without data augmentation. These findings offer a valuable foundation for health monitoring and fault prediction in high-reliability systems.