An assessment model of water pipe condition using Bayesian inference

Abstract

An accurate understanding of the condition of a pipe is important for maintaining acceptable levels of service and providing appropriate strategies for maintenance and rehabilitation in water supply systems. Many factors contribute to pipe deterioration. To consolidate information on these factors to assess the condition of water pipes, this study employed a new approach based on Bayesian configuration against pipe condition to generate factor weights. Ten pipe factors from three pipe materials (cast iron, ductile cast iron and steel) were used in this study. The factors included size, age, inner coating, outer coating, soil condition, bedding condition, trench depth, electrical recharge, the number of road lanes, material, and operational pressure. To address identification problems that arise when switching from pipe factor information to actual pipe condition, informative prior factor weight distribution based on the literature and previous knowledge of water pipe assessment was used. The influence of each factor on the results of pipe assessment was estimated. Results suggested that factors that with smaller weight values or with weights having relative stable posterior means and narrow uncertainty bounds, would have less influence on pipe conditions. The model was the most sensitive to variations of pipe age. Using numerical experiments of different factor combinations, a simplified model, excluding factors such as trench depth, electrical recharge, and the number of road lanes, is provided. The proposed Bayesian inference approach provides a more reliable assessment of pipe deterioration.