Abstract: As a core component of industrial systems, the accuracy of the dynamic model of rotating machinery directly affects the reliability of fault diagnosis and health management. To address the issues of parameter mismatch and large model errors in traditional physical modeling methods under practical operating conditions, a bidirectional model correction method combining "physical constraints and data-driven approaches" is proposed, which is based on the Multi-Layer Perceptron (MLP). First, the Sobel sensitivity analysis method is used to screen key correction parameters, and Poisson disk sampling is employed to achieve uniform sampling in the high-dimensional parameter space, generating datasets of structural and bearing parameters. Then, an MLP neural network is constructed to perform joint optimization of structural and bearing parameters, achieving high-precision mapping between the dynamic model and the measured response. A typical dual-rotor system is used as the object for simulation and experimental verification. The results show that this method can significantly reduce the error between the finite element model and the actual system in key dynamic indicators such as the critical speed, with the post-correction critical speed identification error being less than 2%. This method offers advantages such as automation, high precision, and strong engineering adaptability, providing effective technical support for digital twin modeling and intelligent diagnosis of rotating machinery, and demonstrates a promising outlook for engineering applications.