To address the difficulty of traditional formula-based surface modeling methods in accurately describing real manufacturing errors, this study proposes a new approach based on variational mode decomposition (VMD) for separating surface profile components and building non-ideal surface models. Most existing methods generate non-ideal surfaces by assuming predefined theoretical models. However, due to the multi-scale nature of surface geometry and the strong coupling of different types of manufacturing errors, such methods often fail to capture the true shape of actual surfaces. In this work, we introduce the VMD algorithm into the field of surface modeling to adaptively separate multi-frequency components from surface data, enabling frequency-domain decoupling of manufacturing errors. Specifically, VMD is used to extract low-frequency components related to form errors and mid-frequency components related to waviness, which are then used to construct a realistic surface model. Simulation results show that, compared to traditional Gaussian filtering, the proposed method achieves more accurate surface reconstruction and reveals clearer spatial patterns of manufacturing errors. This method provides high-fidelity geometric input for downstream assembly error analysis and improves the reliability of precision assembly evaluation.