The study of the fatigue performance of corroded steel wires in bridge cables holds significant scientific value for advancing structural theory and informing engineering practice. To address key challenges in fatigue life prediction such as the scarcity of the complexity of nonlinear relationships and the lack of model interpretability, this study proposes a progressive solution framework consisting of integrated optimization, transfer validation, model interpretation, platform development. A dual-source heterogeneous database (A/B) was first constructed by integrating 422 sets of specimens data from the literature with 30 sets of experimental data obtained through independently conducted corrosion tests. An integration strategy based on stacked-transfer models is used to couple the strengths of six different machine learning (ML) models. The improved sparrow optimisation (ISSA) algorithm was employed for hyperparameter optimization. The results demonstrate that the proposed Stacking model surpasses both individual base learners and existing mathematical models from literature and specifications in prediction accuracy. When transferred to new independent datasets, the model maintains excellent predictive performance, validating its strong generalization capability. Furthermore, by incorporating the SHAP framework, the study systematically deciphers the model’s decision-making mechanism and quantifies the contribution distribution of individual parameters to fatigue life. Finally, to enhance model applicability, a web-based human–computer interaction platform for intelligent fatigue life prediction was developed based on the stacking-SHAP model. This study provides a data-algorithm-platform trinity solution for the whole life cycle management of bridge cables.
扫码关注我们
求助内容:
应助结果提醒方式:
