Hydrostatic Performance Of Reinforced Concrete Pipe For Infiltration
Abstract
Groundwater infiltration into underground sewer systems has long been a costly issue for municipalities. With reinforced concrete pipe (RCP) being a primary option for sewer systems, existing hydrostatic testing methods conducted by manufacturers to measure internal pipe pressure, as required by specifications, do not reflect in-situ external hydrostatic conditions. This thesis records the development of a novel testing method to evaluate the RCP joint performance for infiltration. The test is safe and easy to conduct by RCP producers at the factory. The test method mimics field conditions of possible RCP joint gap and joint offset. Over 100 tests were conducted, including 600 mm, 900 mm and 1200 mm RCP with conventional single offset self-lubricated gaskets. This study also evaluates the gasket performance for infiltration. Pipe joint performance curves were developed based on the test results. Comparison to laboratory load deformation tests on gaskets were conducted, indicating that predictions of the sealing potential derived using gasket geometry agreed well with results of infiltration tests. The study shows that the joint gap plays an important role in the sealing potential. The developed apparatus allows the observation of gasket movement under infiltration pressure against the gasket leading to failure. The performance curves also allow the prediction of an infiltration potential leading to a practical applicational procedure to guide RCP installation. A case study of deep RCP pipe subjected to groundwater pressure illustrated the usefulness of the performance curves to derive maximum allowable joint gaps, which contractors could rely on during RCP installation. The findings should allow deducing technical guidance on how water tightness of RCP can be achieved at installation below the prevailing groundwater level. Two oversampling methods: Synthetic Minority Over-sampling Technique (SMOTE) and Density-Based SMOTE were employed to address the unbalanced dataset. Accordingly, applying advanced machine learning techniques, the scale of variation in the test data can be analyzed and accurately predicted using tree-based supervised classification methods: random forest, extra trees and gradient boosting.