Boxall, J. B., O'Hagan, A., Pooladsaz, S., Saul, A. J. and Unwin, D. M.
Department of Probability and Statistics and Department of Civil and Structural Engineering, University of Sheffield, Sheffield, England
Publication details: Water Management 160, 73-82, 2007.
The ability to forecast the burst behaviour of pipes in water distribution systems is a fundamental requirement of proactive planning for investment, replacement and rehabilitation strategies. However, burst behaviour is a complex function of many uncertain contributing factors. Attempts to develop accurate predictive models have been limited by: data quality and quantity; the lack of associations between individual events and pipes; and the statistical techniques available and applied.
Predictive expressions for annual burst rate in cast-iron and asbestos-cement pipes are derived for two sample datasets from the UK. The available data was interrogated to make rigorous associations between individual events and pipes and filtered for missing, default and erroneous data. Using statistical analysis, the theory of generalised linear models, it was then possible to derive predictive expressions for burst probabilities. The resulting expressions show strong associations between annual burst rate, diameter and length, and a complex association with age. Analysis was also undertaken to investigate associations with available soil data, a relative lack of consistent dependence of burst rate on this data is demonstrated.
Keywords: Water supply, bursts, statistical analysis, asset management.