Predictingthe Failure Performance of Individual Water Mains

ReportNo WSAA 114

July 1996




Water-supplymains have been examined to determine the optimum timing for the replacement ofa deteriorating asset. It has been found that, with suitable filtering of raw failure data for a pipe asset, it ispossible to make reasonable predictions of the future failure behaviour of theasset. Drawing on work in data filtering by Minetti (1995) and by Constantineand Darroch (1995) for partial data sets, models have been developed whichpredict the time to the next failure and then suggest whether the main shouldbe repaired or replaced in order to minimise life-cycle cost.


Thereport incorporates a set of guidelines for operators of water-supply mains.These can be applied by any Authority to a main with a history of four or morefailures. Authorities are expected to refine these models according to theirown operating environment and specific corporate goals. However, there is astrong recommendation that a standard format be adopted for the collection ofperformance data so that future research projects in this area can be assuredof data consistency across the country (see Appendix B).


Thefindings of this project will enable more effective management of water-distributionnetworks by ensuring that a main is replaced at the end of its economic life.


Keywordsin this study are: asset management; water-supply pipe; failure; pipereplacement rules; data filtering; economic model.


Earlierthe Urban Water Research Association of Australia (UWRAA) research hasindicated the need to develop a reliable means to predict the failureperformance of individual water mains. The project team has analysed therecords of small-diameter asbestos-cement and cast-iron pipes and from theanalysis developed failure and decision support models. The end product is aset of guidelines which operators can use to decide how to minimise thelife-cycle cost of a main under their control.


Afundamental requirement of any water-main failure model is that it shouldrepresent the aging, or gradual deterioration, of the pipe in its operatingenvironment. To develop a model on this basis it is necessary to remove datarelating to externally imposed factors such as bad repair of a previous failure,faulty operation of the supply system, accidental damage during adjacentexcavation work and intentional damage. To achieve this filtering, a processhas been developed which screens the raw performance history data and leavesthe failures due to inherent pipe characteristics. The latter are then employedin the failure prediction model to forecast the time to the next failure.


Afterextensive trials with data supplied by Melbourne Water and the Sydney WaterBoard, a model with an exponential form that predicts interfailure time wasselected as the most appropriate. This failure model requires the full burst history of the main. In thisregard, information on four or more failures are needed following the filteringprocess. If this is not available, the reader is referred to the companionUWRAA study presented in Research Report AM22 (Constantine and Darroch, 1995),which provides a failure prediction model for incomplete datasets.


Thebasic building block of predictive modelling is an accurate database of failuredata on all mains.


Thereport presents results from a survey of water authorities that examined thedata currently being collected in Australia. A major point coming from thissurvey was the obvious need for a set of standard definitions and fields forfailure databases. It is strongly recommended that a standard be developed toprovide consistency across the Australian water industry. A suggested format isprovided in Appendix B.


Thesurvey of water authorities also covered the currently used procedures foreconomic evaluation of water mains, and the replace/repair decision inparticular. Greater emphasis is now being placed upon indirect costs associatedwith burst water mains and the total cost to the community. Using thisinformation and applying the failure forecasting model developed in the earlierpart of the project, a decision model was developed which employs the conceptof Total Failure Cost to decidewhether to repair or replace a failing main. Computer software has been writtento assist authorities apply the model and this is available from theInfrastructure Research Unit, Department of Civil and Geological Engineering,RMIT.


Thefindings of this project are incorporated in a set of guidelines providingoperators with a package of tools aimed at helping them minimise life-cyclecosts. Advice on the collection and interpretation of performance data isfollowed by instructions on the application of filtering, the failureprediction process and the determination of an optimal replacement date.


It issuggested that authorities apply the guidelines included in this report andwith local data, over a period of time, refine them to better manage theirresources. The guidelines can be used immediately but they must be converted toyour own environment using your own failure data.


Copies of the Report are available from WSAA, price$A100. Orders may be placed through the Bookshop at or by email to