Abstract: This paper proposes a physics-informed and data-driven framework based on Transformer architecture for early anomaly prediction in aero-engines.The framework embeds physical constraints into the bearing fault simulation data generation process,provide physically consistent simulation training samples for the Leffler-Kernel-enhanced Transformer model.This integration significantly enhances the model’s physical fidelity and generalization capability.Experimental results on a real-world dataset—where faulty bearings were manually replaced in an aero-engine testbed and vibration signals from both the rotor and casing were collected—demonstrate that the proposed method achieves strong predictive performance in early fault detection.The approach shows high potential for practical engineering applications and contributes to predictive maintenance and fault prevention in aero-engine systems.