Predicting failure time and location in buried cast iron water pipes due to external Corrosion

Water distribution networks comprising of pipes transporting water throughout a region perform an essential societal function, providing potable water for use. Failure of water pipes, however, can result in significant economic costs, loss of water resources, environmental damage and disruptions to people’s lives. There are a considerable number of in-service aged cast iron pipes throughout water distribution networks within Australia. The failure rate of these pipes has been increasing largely due to external corrosion which is an ongoing threat to pipe integrity. The water industry desires to reduce pipe leading to research interest to develop methods for predicting pipe failure time and location which enable proactive maintenance. External corrosion, a leading cause of pipe failure, is more likely to be initiated by pitting corrosion rather than uniform corrosion. In pipes, pitting corrosion is driven by the material and soil parameters, and generally appears as wall thinning, which can reduce structural strength leading to potential leakage, mechanical collapse, and bursting of the pipe. Modelling corrosion of buried cast iron pipes is a useful remaining-life predictive tool as undertaking pipe condition assessments is difficult and costly. Considering the complexity of cast iron corrosion behaviour in soil and the effects of pipe installation and environmental parameters on the probability of failure makes this investigation a multi-disciplinary topic that requires study in macro and micro scales.  In the micro-scale, the pitting corrosion process is a repetitive cycle of multiple sub-processes including pit initiation, growth, and repassivation. The research plan for modelling the corrosion process through two steps is achieved in this study. First, an integrated model for pit initiation and its transition to growth for a single pit is developed based on the Point Defect Model, with a deep understanding of the electrochemical aspects of the pitting process, including mass balance, charge neutrality, and temporal variation of pH and the concentration of aggressive ions. The model determines the pit initiation time as the transition time from oxygenated condition dominant, to anoxic condition dominant. The developed model presents the probability of pit initiation over time, which aligns greatly with the previous regression models. Then, the single pit model is extended to a surface by considering pit interactions and the spatial variations of material and soil characteristics such as pH, concentration of aggressive ions, soil pores and moisture, material graphite and inclusion sites, and the number of rainy days during a year. Random Field theory is employed to model the pipe material surface, soil texture, and particle distribution. The developed 3D corrosion model employs the Cellular Automata (CA) technique to simulate the corrosion growth and pit depths distribution on the pipe surface by defining the CA transition rules based on the ongoing chemical reactions and boundary conditions at the pitting locations. The estimated pit depth by the developed CA corrosion model has been verified with the results from two experimental data sets, the National Bureau of Standard dataset and a random selection of Australian water pipes. The outcome shows the efficiency of the developed CA model in estimating the pit depth based on the soil characteristics with a 6% Mean Square Error for short and long-term exposure times. Sensitivity analysis shows that pH, the concentration of aggressive ions, and soil moisture content have a threshold to effectively accelerate the pit initiation time and increase the corrosion growth. At the macro scale, spatial modelling of pipe networks, pipe structural analysis, and failure assessment due to external corrosion are studied. The developed CA corrosion model for estimating the corrosion growth in pipes is employed in this reliability study. Different structural models are used for analysing the internal and external stresses on buried corroding pipes considering the pipe installation, traffic and soil stress variables. The designed reliability system comprises of leakage and bursting failure modes to predict the probability of failure due to the estimated external corrosion. The hazard rate resulting from the developed model is verified with three different datasets of burst failure records collected from Australian water authorities; Sydney, Hunter and Yarra Valley Water. Results show that the first peak of failures happens between 45 to 60 years from their installation time, followed by a slight decrease, and then increase again after 90 years from their installation time. The contribution of the variables on increasing the probability of failure and hazard rate is investigated and highlights the significant variables which include pH, pipe wall thickness, traffic load, and material characteristics. The ultimate goal of this study is to predict the failure time and locations along a pipe network comprised of kilometres of pipe branches and sections installed at different times and soil conditions with differing installation parameters. Achieving this goal requires modelling the spatial distribution of pipe network variables and implementing the developed models along the pipe network of interest. For this purpose, graph theory is employed to model the pipe network. Random field theory is used to reconstruct the spatial distribution material, soil, and environmental variables along the pipe network. The spatial variation of variables along the pipe length is used to efficiently segment the pipe network with variable length, as each segment has a distinctive set of variables compared to its adjacent segments. Implementing the developed corrosion and reliability model for each segment results in predicting the probability of failure due to considered failure modes. Eventually, the potential failure time and locations can be predicted by following the first passage of the probability of failure theory and comparing the predicted probability of failure to the asset owner's acceptable threshold. The application of the developed methodology in this study is graphically presented by colormap videos of a worked example pipe network as a tool for identifying the pipe failure locations over time. Monte Carlo realisations are implemented in all simulations to produce unbiased results. The outcomes of this research can potentially reduce the consequences of failure by assisting asset owners in rationalizing their inspection and maintenance plans.