n solving finite element problems in mechanical models, the speed of reaching the response and accuracy are important. Issues such as "limited computational memory and the need for real-time response" affect solution speed and accuracy. What is proposed to address these problems is the development of methods that reduce the problem solution time while maintaining accuracy to an acceptable level. These problems will also be more severe for models with fine meshes and dynamic solution methods.One of the proposed approaches for this class of problems is reduced-order modeling of the finite element model. In this paper, reduced-order modeling for a nonlinear dynamic model is presented, and the accuracy and speed of reaching the response are investigated.This method is based on proper orthogonal decomposition of "singular vectors and values," and optimal use of eigenvectors and eigenvalues. This method has been validated for heat transfer and displacement resulting from thermal expansion and pressure applied to a compressor blade model.The goal is to develop and train a model and test it with new boundary conditions.For this model, and based on the computational system used, the overall computational speed for the two heat transfer and static solutions has increased by a factor of 7.634. Additionally, the maximum error percentage of the reduced-order model compared to the full model is 3.6%. This trained model can provide a faster response regarding the blade's behavior in crisis situations. Furthermore, due to its lower computational footprint, it is capable of running on low-end processors.