Predictions of the structural condition of urban drainage networks enables us to estimate future investment costs under various rehabilitation strategies. This facilitates proactive, far-sighted sewer asset management, which, in turn, contributes to a better balance between expenses and system performance.
Many sewer deterioration models have been formulated to predict the physical aging of pipes in terms of discrete sewer condition states. However, in many cases there is not sufficient information available to calibrate them. A severe issue is that historical asset data is not available. Specifically, the lack of the following information often hinders us to calibrate sewer deterioration models:
- (i) Condition records of sewer pipes which have been replaced by new ones. The corresponding records are discarded from the database and replaced by the new information.
- (ii) Condition ratings of renovated or repaired sewer pipes from the time before such action. The repaired or renovated pipes are reassessed. On the basis of this reassessment, the condition rating prior to repair or renovation is overwritten by a new (typically better) condition rating. Thus, such records should be excluded from the analysis. The condition of such pipes is the result of not only aging but also rehabilitation.
When a sewer deterioration model is calibrated without explicitley accounting for effect (i) and (ii) the predictions suffer a ‘survival selection bias’ (Scheidegger et al., 2011).
The SWIP-SDM represents a simple probabilistic sewer deterioration model and provides the possibility to infer the deterioration parameters from data lacking historical information as defined by case (i) and (ii). Details of the model are given in Egger et al. (2013).
The R code provided below allows for model calibration and forecasting of condition states of the network. Model calibration is done by Bayesian inference (Egger et al., 2013). The likelihood functions described in Egger et al. (2013) (Eq. 1 and 6) are extended so that also condition data from two subsequent inspections per pipe can be used for parameter inference.
SDM.zip contains the required R code. Look at SDM_inference.R to do parameter inference and at SDM_forecast.R to predict future condition states for a given model parameters.
For demonstration additionally two datasets are provided. Both contain the condition ratings of a sewer network which underwent substantial rehabilitation in form of pipe replacements in the last 30 years. Each row of the datasets represents one pipe with one or two observed condition states. Dataset dataset_example_complete.RData contains the condition records of all pipes ever constructed, i.e. it includes also pipes that have been replaced in the past. The data set dataset_example_incomplete.RData is more realistic. It contains only those pipes that are still in service, i.e. the data set lacks historical data of pipes replaced in the past (according to case (i) defined above).
Matching parameters for a log-normal prior distribution are contained in prior_complete.RData and prior_incomplete.RData respectively.
The development of SWIP-SDM was part of the National Research Project 61 (NRP 61) http://www.nfp61.ch/E/Pages/home.aspx funded by the Swiss National Science Foundation (SNF) (project number 406140_125901/1).
Egger, C., Scheidegger, A., Reichert, P. and Maurer, M. (2013) Sewer deterioration modeling with condition data lacking historical records. Water Research 47(17), 6762-6779.
Scheidegger, A., Hug, T., Rieckermann, J. and Maurer, M. (2011) Network condition simulator for benchmarking sewer deterioration models. Water Research 45(16), 4983-4994