A data-driven decision support system for high-pressure hydrogen-induced rubber volume expansion testing: Part I. A classification model for predicting test validity

Hydrogen exposure testing of sealing rubbers is essential for evaluating their volume expansion behavior, yet these experiments are costly and can occasionally yield data unsuitable for quantitative analysis. This study develops a machine-learning framework that predicts, prior to testing, whether a rubber material will provide a reliable measurement of its maximum volume expansion ratio after high-pressure hydrogen exposure. Dynamic mechanical analysis (DMA) features and tensile stress–strain (SS) features were extracted from various rubber materials, and each sample was labeled as “ pass ” or “ fail ” based on the measurability of its pressure–expansion relation. Among the tested algorithms, the linear support vector machine (SVM) achieved the highest discrimination performance. SHapley Additive exPlanations (SHAP)-based feature interpretation identified maximum of tanδ (tand_max) and Young's modulus (Ess) as the most influential predictors, revealing two formulation-dependent tendencies associated with measurement failure: Both tendencies are consistent with a mechanism in which internal blisters form and propagate to the surface, followed by rapid hydrogen desorption and abrupt shrinkage, yielding unreliable expansion measurements. The proposed model therefore serves as an effective pre-screening tool and suggests possible physicochemical factors associated with measurement failure, helping reduce experimental cost while guiding future investigations of hydrogen–rubber interactions.