Evaluating the role of deterioration models for condition assessment of sewers

Abstract

Ensuring reliable structural condition of sewers is an important criterion for sewer rehabilitation decisions. Deterioration models applied to sewer pipes support the rehabilitation planning by means of prioritising pipes according to their current and predicted structural status. There is a benefit in applying such models if sufficient inspection data for calibration, an appropriate deterioration model, and adequate covariates to explain the variability in the conditions are available. In this paper it is discussed up to what level the application of sewer deterioration models can be beneficial under limited data availability. The findings show that the indirect nature of the explanatory covariates which are commonly used in sewer deterioration models makes it difficult to harness any benefit from modelling sewer conditions at a network level, but that the deterioration model application still may be beneficial for prioritising inspection candidates. The prediction power of the current sewer deterioration models is limited by the adequacy of the explanatory variables available, and by the fact that different failure modes are mixed in the aggregated condition class, and not modelled explicitly.

Pipe failure modelling for water distribution networks using boosted decision trees

“…There exist several approaches for sewer networks, but these are only partly comparable since the factors affecting pipe failures in water networks are different from the factors in the sewers. Rokstad and Ugarelli (2015) compared random forest algorithms with statistical deterioration models for sewers and found that random forests are not suitable to estimate condition states. Syachrani et al (2013) employed a decision tree-based deterioration model for sewer pipes to predict the 'real' age of their pipes, using prior clustering to get slimmer decision trees.…”

Sewer asset management – state of the art and research needs

“…Existing sewer deterioration models can be classified into three basic groups: deterministic, statistical and artificial intelligence (AI) models. For a detailed review of modelling approaches, the authors refer to Ana and Bauwens (2010), Kley et al (2013), Marlow et al (2009), Rokstad and Ugarelli (2015) and Tscheikner-Gratl (2016).…”

“…Survival analysis and Markov-chain are the most common types of statistical deterioration models on a network level (e.g. Caradot et al 2017;Duchesne et al 2013;Egger et al 2013;Le Gat 2008;Micevski, Kuczera, and Coombes 2002;Rokstad and Ugarelli 2015). Prior to model calibration, pipes are generally grouped in cohorts, i.e.…”

Sewer Life Span Prediction: Comparison of Methods and Assessment of the Sample Impact on the Results

“…They found that there was uncertainty relating to both predicting the overall condition distribution of the network and the state of individual pipes. Rokstad and Ugarelli [20] compared the predictions given by the GompitZ model and a random forest (RF) model and found that RF generally outperformed GompitZ in terms of uncertainty. However, when the amount of data was reduced, both predictions exhibited clear biases.…”

“…In cases where actual data were applied, their extent and characteristics varied. In Rokstad and Ugarelli [20], the data set comprised some 27% of the Oslo VAV sewer network, mainly inspected within a five years period. Caradot et al [3] had an extensive data set where all the network pipes had been inspected at least once and more than 50% at least twice.…”