Answers are grounded in this paper and cite their sources.
Accurate prediction of fretting-fatigue life remains challenging due to strong nonlinear interactions and limited experimental data. In this study, a physics-informed multiplicative Kolmogorov-Arnold network (PI-M-KAN) is developed for fretting-fatigue life prediction under limited-data conditions. The proposed model integrates a multiplicative interaction structure with a Smith-Watson-Topper (SWT)-based constraint through a two-stage training strategy. A dataset covering multiple aluminum alloys is used to evaluate predictive performance under repeated cross-validation. The results demonstrate that incorporating multiplicative interactions substantially reduces prediction error and improves consistency compared with conventional multilayer perceptrons and tree-based models. Statistical analysis indicates that the dominant performance gain arises from the multiplicative representation of input interactions, while the SWT-based constraint provides an additional positive contribution when applied through the two-stage strategy. Interpretability analysis shows that the trained model captures interaction-dominated relationships among contact and loading parameters. The proposed framework offers a practical approach for fretting fatigue life prediction and structural integrity assessment under limited-data conditions, effectively combining flexible nonlinear representation with physically guided stabilization.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: