<p>Current studies on the nonlinear dynamic modeling of bolted joints often neglect the influence of the effective contact area between interfaces. This assumption introduces significant deviations in the equivalent dynamic models due to the discrepancy in contact stiffness at the interface. To address this issue, this paper proposes a twin finite element modeling approach incorporating gradient stratification based on localized contact characteristics. First, a finite element model is established to analyze the contact pressure distribution in the bolted joint interface. A fourth-order polynomial function is used to fit the contact pressure distribution curve. Using 10 % of the peak contact pressure as the threshold, the validity of the selected pressure range is verified through an area integral formula, ensuring it accounts for over 99 % of the total pressure. This allows the effective and ineffective contact areas to be accurately identified. Based on the gradient distribution characteristics of contact pressure, the effective contact area is further divided into virtual material layers with gradient stratification. Subsequently, a deep neural network (DNN) model is employed to achieve a nonlinear mapping between the contact states of the bolted joint, the gradient-stratified regions, and the equivalent dynamic parameters. A twin finite element model is then constructed. Finally, a particle swarm optimization (PSO) algorithm is used to identify the equivalent dynamic parameters of the virtual material layers. The identification process minimizes the root-mean-square error between the first four predicted modal frequencies of the twin finite element model and experimentally measured values. The experimental results demonstrate that the proposed modeling method achieves a prediction error within 1 % for the first four modal frequencies compared to the measured values.</p>