Response signatures for seismic diagnosis and inspection prioritization of corroded buried gas pipelines

Localized corrosion can amplify seismic demand in buried gas pipelines, but translating response changes into local diagnosis remains challenging. This study develops a physics-guided framework that uses finite-element response histories to identify corrosion states and rank inspection priorities. Fourteen monitoring-location samples from eight scenarios were used for development. Twenty-six samples from 14 additional scenarios, covering two new ground motions and corrosion depths of 3 and 6 mm, were used for external numerical evaluation. Four feature representations and five low-capacity classifiers were compared using scenario-grouped cross-validation. The selected model combines 16 strain-stress indicators, z-score standardization, and a balanced linear support vector machine. Its fitted scaler and classifier were frozen before external evaluation. Intact responses under each excitation remained required reference inputs. The global acceleration-displacement representation achieved 0.214 accuracy. The selected model classified all 14 development samples correctly in grouped validation and 23 of 26 external samples correctly, giving 0.885 accuracy and 0.917 balanced accuracy externally. All 14 new-ground-motion samples and 9 of 12 depth-variation samples were correctly classified. A baseline-independent local representation classified 25 of 26 external samples correctly. Feature ablation indicated that stress-amplitude information supports depth transfer, while bootstrap intervals and diffuse class probabilities exposed small-sample uncertainty. A relative response-demand index ranked dual external corrosion highest under axial-dominant shear-wave excitation. These findings support conditional transfer within the tested numerical domain. Experimental validation and reliable field decision thresholds remain to be established.