Abstract: To address the high cost of long-term direct stress monitoring, limited sensor coverage, and the difficulty in continuously updating the fatigue-life status of critical welds in metro bogie frames, a remaining life rolling prediction method based on an acceleration-stress surrogate model is proposed. In this method, the continuous monitoring signals are partitioned into fixed-length windows, and the root mean square (RMS) of acceleration within each window is extracted. An acceleration-stress surrogate model is adopted to predict the corresponding stress RMS for each window. Furthermore, taking the predicted stress RMS as an equivalent indicator to quantify the stress response intensity of current window, the damage rate per unit mileage, cumulative damage and remaining operable mileage are updated with the accumulated operating mileage. Validation results using samples constructed from synchronously measured acceleration and strain data demonstrate good agreement between the predicted and measured stress RMS values. Mileage simulation results further show that the proposed method can effectively capture the accumulation of weld fatigue damage and the degradation trend of remaining life under increased vibration conditions, so that it provides a theoretical basis for online condition assessment, risk prioritization, and maintenance warning of critical welds in metro bogie frames.