Support tools to predict the critical structural condition of uninspected pipes for case studies of Germany and Colombia

Abstract

Several deterioration models have been used to predict the structural condition of sewer pipes, and some have been applied in different cities in the world. However, each one of these models has not been proved simultaneously for case studies with different characteristics (topographic conditions, soil uses, demographic growth, utilities' service operation and city's dynamic) and the use of their predictions have not been analyzed to support different management objectives. Therefore, the objective of this work was to assess the prediction results of two models (based on Logistic Regression and Random Forest (RF) methods), which previously have been identified as successful in other experiences, for two different case studies (a city in Colombia and a city in Germany). The prediction assessment was carried out by three analysis techniques (Positive Likelihood Rate (PLR) index, performance curve and deviation analysis). According to the results, we found that: (i) the model based on RF was the one that could be useful as a support tool in the sewer asset management of both case studies; (ii) for the German city, the prediction results could be useful for designing strategic investment plans in order to know the number of pipes that the utility should rehabilitate each year; and (iii) for the Colombian city, the predictions are appropriate to make decisions concerning inspection or rehabilitation plans, since the probability of identifying the sewer's assets in critical condition (C4) correctly (according to the analysis of the sample of the 10% of sewers with the highest probability to be in this condition) is around 63% and could be 83% if the stakeholders also consider in these plans the misclassification of those pipes in a bad structural condition (C3).

Is it possible developing reliable prediction models considering only the pipe’s age for decision-making in sewer asset management?

“…(2) Scenario 2: the age together with other sewer available characteristics such as the material, the type of effluent (sewerage), the pipe's depth, length, slope and diameter as input independent variables that could influence the structural condition of the sewer assets (Ariaratnam et al, 2001;Baik et al, 2006;Younis and Knight, 2010;Tscheikner-Gratl et al, 2016;Hern andez et al, 2018).…”

A Two‐Stage Data‐Driven Spatiotemporal Analysis to Predict Failure Risk of Urban Sewer Systems Leveraging Machine Learning Algorithms

“…In the search for efficient models and tools to predict the physical condition of underground sewer infrastructure, studies such as Sousa, Matos, and Matias (2014), Jiang, Keller, Bond, and Yuan (2016), Santos et al (2017), Caradot et al (2018), andHernández, Caradot, Sonnenberg, Rouault, and were aimed to compare a collection of different models, and identifying the ones that produced the best results under several conditions. Additionally, Laakso, Kokkonen, Mellin, and Vahala (2018) and Elmasry, Hawari, and Zayed (2017) coupled different models as a part of a single framework with the idea of combining the predictive capabilities of such models in a single tool.…”

“…These models were used to predict three levels of sewer condition: good, medium, and bad. Hernández et al (2018) evaluated two different models' predictive outcomes, namely, LR and RF, for two different case studies, a city in Europe and a city in South America. The models were used to predict the critical structural condition of sewer pipes in both cities on a four-level scale.…”

Sewer asset management – state of the art and research needs

“…Few studies evaluated the performance of deterioration models to simulate the condition distribution of the network (Caradot et al 2017(Caradot et al , 2018Duchesne et al 2013;Hernández et al 2018;Ugarelli et al 2013). They indicated that survival analysis and Markov models outperform a simple random model for predicting the condition distribution of the network, especially in the case of low data availability.…”

“…Main machine learning methods used as deterioration models are Random Forest (e.g. Harvey and McBean 2014; Laakso et al 2018; Rokstad and Ugarelli 2015), Support Vector Machines (e.g Hernández et al 2018;Mashford et al 2011;Sousa, Matos, and Matias 2014),. and Neural Networks (e.g Jiang et al 2016;Sousa, Matos, and Matias 2014;Tran, Ng, and Perera 2007)…”