<p>Accurately predicting the residual strength of blended hydrogen natural gas pipelines containing a crack-in-dent defect is critical for ensuring their safe and stable operation. Machine learning methods offer an effective approach for predicting residual strength. However, the application of machine learning models to the prediction of residual strength in blended hydrogen natural gas pipelines is often challenged by limited sample sizes and the difficulty of interpreting feature interactions. To address these challenges, a model integrating a tabular foundation model with interpretability analysis techniques was proposed. This model not only achieves highly accurate residual strength prediction under limited data conditions but also facilitates interpretability analysis of feature interactions. Comparative experimental results indicate that the model outperforms baseline models in predictive accuracy, achieving an <i>R</i>² of 0.9961. The model also demonstrates outstanding predictive stability, with the majority of absolute errors falling within 0.15 MPa and the maximum absolute error reaching only 0.2858 MPa. In addition, the interpretability analysis method facilitates the interpretability analysis of both individual features and their interactions, thereby improving model transparency. This methodological framework holds significant potential for supporting safety assessments and intelligent decision-making in the operation of blended hydrogen natural gas pipelines.</p>