In practical engineering scenarios for rolling bearing fault diagnosis, physical sensing limitations often result in incomplete or insufficient real-world data acquisition, posing significant challenges to accurate fault identification. To address this issue, this study proposes an intelligent fault diagnosis method for rolling bearings based on kinetics simulation data. First, a geometric model of the bearing is constructed using SolidWorks and imported into Adams to establish a dynamic simulation model, generating comprehensive simulation datasets. Subsequently, a Deep Adaptive Representation Transfer (DART) learning framework is developed, where simulation data serves as the source domain and limited measured data as the target domain. An Improved Joint Distribution Adaptation (IJDA) mechanism is innovatively introduced to align marginal and conditional distributions between domains, significantly enhancing domain adaptability. Additionally, the I-Softmax loss function is incorporated to improve feature separability. Experimental validation on a rolling bearing test rig demonstrates that the proposed DART method achieves high diagnostic accuracy even with limited target domain samples, confirming its feasibility and practical advantages under physical sensing constraints.