Active magnetic bearings (AMBs) are a new type of high-performance bearings with advantages such as frictionlessness, high rotational speed, and high controllability. However, the rotor mass of AMBs cannot be absolutely uniformly distributed, leading to medium-term unbalanced forces at the same frequency as the rotational speed during high-speed operation, which in turn causes unbalanced vibration. Additionally, the rotor of AMBs transitions from being rigid to flexible at high speeds. Especially at the critical rotational speed where the resonance point is reached, the physical form of the AMB rotor differs significantly from that of a rigid one, and its characteristic data is difficult to collect due to the resonance. Based on the aforementioned problems, this paper implements sample generation for bearing data based on the second-order transfer function mechanism model of AMBs and Generative Adversarial Networks (GANs). Subsequently, an in-depth analysis of the problems with the generated results is conducted, leading to a generation method that combines the mechanism model of the second-order critical rotational speed with the Wasserstein Generative Adversarial Networks (WGAN). Finally, a fusion modeling method for AMBs that integrates mechanisms and data is achieved.