Surface roughness prediction in end milling is challenging because the final surface quality is jointly affected by machining geometry, cutting-force-induced tool deformation, and dynamic process disturbances. This study proposes a physics–vibration collaborative framework for surface roughness prediction. A three-dimensional physical surface model is established by incorporating the effective end-edge geometry and cutting-force-induced tool deflection and tilting, from which the physics-derived roughness is obtained. In parallel, triaxial vibration signals are processed into experiment-level feature sequences, and an attention-enhanced bidirectional long short-term memory–bidirectional gated recurrent unit (AE-BiLSTM-BiGRU) network is employed to extract dynamic process representations. The physics-derived and vibration-derived features are then fused for final roughness prediction. A dataset covering 40 milling conditions is used for validation, with 25 samples used for model development and the remaining 15 samples reserved for independent testing. The independent-test results show that the proposed fusion model achieves better prediction accuracy and closer agreement with the measured roughness than the physics-only and vibration-only models, demonstrating the complementary contributions of physics-based surface information and vibration-based dynamic features.

end milling,physics–vibration fusion,AE-BiLSTM-BiGRU,Surface roughness prediction