Author(s): Alessandro Lombardini; Cristian Cappello; Lorenzo Carmelo Zingali; Cristiana Bragalli; Angelo Leopardi; Carla Tricarico
Linked Author(s):
Keywords: Data-driven modelling; Network heterogeneity; Pipe failure; Deterioration ranking; Water distribution networks
Abstract: This study examines the robustness and transferability of data-driven pipe deterioration models across heterogeneous Water Distribution Networks (WDNs) located in Northern and Southern Italy. A unified modelling framework, based on a common set of physical and hydraulic predictors and implemented through four classification algorithms, is applied to explore how network specific characteristics affect predictive performance. The findings reveal significant accuracy discrepancies among all models when applied to the networks, pointing to the influence of differing deterioration dynamics, operational contexts, and data quality. These outcomes highlight the importance of explicitly considering network heterogeneity when designing predictive models for WDN asset management.
Year: 2026