ESTIMATING THE CONDITION OF UNINSPECTED SEWER PIPELINES

The quantity and quality of information that owners of collection systems in the U.S. are currently collecting come in the wake of a long history of less-than-complete record keeping for pipeline maintenance, condition assessment and rehabilitation efforts. Business models of collection system planning and operation such as asset management techniques are proving the value of high-quality information when used in a decision-making environment. However, complete knowledge of the condition of a collection system is difficult and expensive. The data gap created by the uninspected portion of the collection system represents a large source of uncertainty for decision makers, who apply decision-support models for long-term capital and O&M spending. The purpose of this paper is to describe a discrete classification technique used to estimate uninspected pipeline conditions and present results from an application of this method to a sewer collection system in Vallejo, California.Linear regression techniques using Ordinary Least Squares (OLS) have been applied in many instances for predicting the state of the uninspected portion of a pipe network. These models tend to have a limited ability to predict the condition of uninspected pipes. One problem is that these models are usually based on pipe age. Intuitively pipe age should be highly correlated with condition. However, unrecorded historical rehabilitation efforts cloud the relationship between time and condition, along with other environmental and spatially based factors that affect condition in a non-homogeneous fashion. For example, a pipe may be recorded as being 50 years old, but there may be no record that the same pipe was rehabilitated several times over this period, and my be in excellent condition. Likewise, a different 75 year-old pipe may be in good condition, while a relatively new pipe (e.g. 15 years old) may be in very poor condition because it was installed in corrosive soils and underneath a highly traveled roadway.Discrete statistical models may be more appropriate for this problem. Alternative regression techniques such as logistic regression and discriminant analysis techniques that include spatial relationships as well as the consequences of error (i.e. classifying a pipe's condition as good when it should be poor), can be an improvement over linear regression models based on age. The model developed for Vallejo uses an inspected pipeline condition dataset and descriptive characteristics of the inspected pipes (e.g. type, dia meter, groundwater stage, etc.), to correlate the condition of the uninspected pipes based on their descriptive characteristics using the alternative regression techniques. Once the conditions of the pipes in the uninspected dataset are estimated, this dataset can be coupled with the inspected pipe condition database to provide comprehensive information on the condition of the entire pipeline network. The condition of each pipe (whether inspected or uninspected) can then be coupled with rehabilitation and replacement costs to develop a rehabilitation prioritization plan. The model developed for Vallejo has the advantage of increasing the reliability of the uninspected condition estimates and has successfully been applied to a collection system with over 8,500 pipes.