Sustainable and Explainable Machine Learning for Risk-Based Maintenance Prioritization of Airport Stormwater Pipe Networks

Abstract

Airport stormwater pipe networks operate under demanding conditions where infiltration, structural defects, debris accumulation, hydraulic inefficiencies, and water-quality concerns can gradually reduce system performance. Conventional inspection-based maintenance can identify these deficiencies, but reacting to large inspection datasets and determining which pipe segments require the earliest intervention remains challenging. This study proposes an explainable machine-learning framework for risk-based maintenance prioritization of airport stormwater infrastructure. An anonymized dataset containing pipe characteristics, inspection observations, hydraulic conditions, infiltration indicators, and water-quality measurements is used to develop and evaluate a proof-of-concept machine learning framework for identifying higher-priority maintenance locations. Random Forest and gradient-boosting algorithms are explored within the framework, while explainable-AI techniques are used to identify the factors influencing maintenance priority. The framework is intended to support faster engineering response, more targeted inspections and rehabilitation, reduced unnecessary intervention, and improved asset life-cycle management. The proposed approach provides a data-driven pathway toward more resilient and sustainable maintenance of underground drainage infrastructure in complex airport environments.