This paper proposes an uncertainty propagation analysis method based on transfer learning-enhanced Physics-Informed Neural Networks. The proposed method establishes an efficient framework for response distribution reconstruction by sequentially evaluating reliability information at multiple response levels and reusing previously learned physical information between adjacent response points. By leveraging the ability of PINNs to capture underlying physical knowledge, the method avoids repeated calls to the true performance function. A transfer learning-enhanced iterative mechanism is then employed, in which network parameters are sequentially inherited between adjacent response points so that the learned physical information can be continually reused, thereby reducing redundant retraining and accelerating convergence. The iteration continues until a prescribed reliability threshold is reached, after which the resulting discrete reliability information is used to reconstruct the cumulative distribution function of the response. Four examples are used to illustrate the effectiveness of the proposed method.
