A fault diagnosis method based on digital twins is proposed to address the issue of lag in traditional methods for diagnosing switch machine faults. A virtual model of the switch machine is established to achieve visual interaction of digital twins. Then, the convolutional neural network model used in the fault diagnosis module of the digital twin is validated. Taking the ZD6 switch machine as an example, the simulation results show that the accuracy of fault diagnosis reaches 99.4%. The fault diagnosis method proposed in this article can be used to detect the real-time status of the switch machine.