Resolving Left Truncation Issues and Enabling Pipe-Specific Likelihood of Failure Models

Utility companies manage extensive water distribution pipe networks in urban areas, where aging infrastructure presents challenges in efficiently prioritizing rehabilitation and replacement projects. To address this issue, accurately assessing the condition of each pipe segment is essential. While various approaches exist, questions remain regarding their ability to handle the left truncation problem, create pipe-specific condition models, and maintain model interpretability. This study introduces a method that employs the inverse of time between failures (TBF) to represent pipe deterioration conditions. The method utilizes parametric statistical models to construct idealized pipe condition curves and leverages machine learning to capture complex relationships among factors influencing pipe deterioration, ultimately generating unique, pipe-specific models. The results demonstrate that the TBF approach, in conjunction with parametric models, aligns well with domain knowledge-based physical mechanisms. Moreover, machine learning offers the flexibility to include covariates. These findings are expected to offer utility companies a structured, data-driven approach for prioritizing pipe rehabilitation and replacement initiatives, enhancing the efficiency and reliability of infrastructure management.