Modeling and system identification on flexible multi-bearing rotors with two-concentrated disks are presented in this study. Both rotor unbalance vibration responses through critical speed were experimentally obtained through accurate control of journal bearing static load. Vibration simulations of this laboratory rotor-bearing system were performed as a linear system. The simulation predicted that critical speeds were higher than the experimental ones. These differences strongly occurred because of errors in estimating journal bearing damping and stiffness coefficients. Estimated journal bearing coefficients were carried out by using optimization methods to enhance the model prediction capabilities to the actual test results. The pattern search method was used to solve an inverse problem for the global parameters. A good agreement, in terms of instability threshold speeds and system responses, was found between the experimental results and the optimized model. Both the Response Surface method and the Neural Network method were used to build a Metamodel for predicting the system coefficients. For all tested static loads, the predicted error of the original model was in the range of 9% - 18%.