Markov and Neural Network Models for Prediction of Structural Deterioration of Storm-Water Pipe Assets

Abstract

Storm-water pipe networks in Australia are designed to convey water from rainfall and surface runoff. They do not transport sewerage. Their structural deterioration is progressive with aging and will eventually cause pipe collapse with consequences of service interruption. Predicting structural condition of pipes provides vital information for asset management to prevent unexpected failures and to extend service life. This study focused on predicting the structural condition of storm-water pipes with two objectives. The first objective is the prediction of structural condition changes of the whole network of storm-water pipes by a Markov model at different times during their service life. This information can be used for planning annual budget and estimating the useful life of pipe assets. The second objective is the prediction of structural condition of any particular pipe by a neural network model. This knowledge is valuable in identifying pipes that are in poor condition for repair actions. A case study with closed circuit television inspection snapshot data was used to demonstrate the applicability of these two models.