Recently, the intelligent fault diagnosis models gain increasing attention due to the development of artificial intelligent and state monitoring technology. However, obtaining massive defect data in advance in the actual diagnostic environment is difficult. Constructing diagnostic models on small sample datasets will easily lead to serious over fitting problems and loss of generalization ability, which is referred to as the small sample problem in this study. The simulation model method has made some progress in addressing the small sample problem. However, establishing an effective simulation model is difficult and time-consuming. The simulation signals also have a certain deviation between the actual signals. To address the above problem, a simulation data-driven adversarial domain adaptation fault diagnosis framework was proposed, which is based on dynamic modeling and adversarial domain adaptation approach. First, a reliable and complete dynamic model is established by considering the actual operating state of the faulty part. Second, the failure geometric defects are added to the model as displacement excitation, and the vibration response of the classical fault is simulated. Finally, adversarial domain adaptation approach is utilized to extract the common features of the simulated and measured samples to identify the faults. The effectiveness of the proposed method is validated and discussed on axial piston pump dataset and other dataset. It indicates that the proposed method can effectively solve the small samples problem in different mechanical equipment.