Data-driven network asset management: Focus on sewer systems

Research output: ThesisDoctoral ThesisCollection of Articles

Abstract

Water and wastewater networks are critical infrastructure and vital for modern society. In Finland, the majority of the networks has been built in the 1970s and after and for the first time continuous network renovation is needed everywhere in the country. In this context, data analysis can efficiently support the decisions needed in planning the renovations. The aim of this dissertation was to discover, what kind of data there are available for supporting risk-based network inspection and renovation and how these data should be used for decision making. Both network and environmental data sets were used in the assessments. The dissertation addresses failure consequence estimation, sewer condition prediction and sewer life span estimation. Additionally, current data utilization at Finnish water and wastewater utilities is studied and compared to the state-of-the-art. Failure consequences were assessed for a water distribution network and a sewer system. Although some relevant data, such as hydroynamic modelling results, were not available at the time of the study, the data at hand enabled diverse assessment of failure consequences. The machine learning methods applied for pipe condition and life span modelling - random forests and random survival forests – performed equally or slightly better than the statistical methods logistic regression and Weibull survival. Environmental data sets were particularly useful for consequence assessment, whereas pipe attributes were essential for condition modelling. Partial dependence plots were found to be useful for visually studying the interconnections between pipe condition and explanatory factors. The sewer condition data set contained more pipes in a poor condition than the network as a whole. The data set was suitable for condition prediction, but deficient from the perspective of life span modelling. In order to overcome the obstacles caused by the sample not being representative of the whole network, a method was developed for estimating roughly the range in which the actual life span of the network lies. In order to gain the full benefit from life span models, sewer inspections need to be allocated not only to pipes suspected to be in a poor condition but also to a random sample of pipes whose condition is monitored consecutively in time. Based on the results of this dissertation, currently available data provide relevant support for both estimating failure consequences and predicting pipe condition. The survey on the data use in practice supported the view that existing data sets are not utilized to their full potential. However, there is interest in the adoption of proven new technologies. In the future, the amount of data and the possibilities to their use will increase. This will require water and wastewater utilities to gain new technological, managerial and security-related competencies.
Translated title of the contributionDatalähtöinen verkosto-omaisuuden hallinta - painopisteenä viemäriverkostot
Original languageEnglish
QualificationDoctor's degree
Awarding Institution
  • Aalto University
Supervisors/Advisors
  • Vahala, Riku, Supervising Professor
  • Kokkonen, Teemu, Thesis Advisor
Publisher
Print ISBNs978-952-60-3852-0
Electronic ISBNs978-952-60-3853-7
Publication statusPublished - 2020
MoE publication typeG5 Doctoral dissertation (article)

Keywords

  • asset management
  • condition modeling
  • life span modeling
  • risk assessment
  • water and wastewater networks

Fingerprint

Dive into the research topics of 'Data-driven network asset management: Focus on sewer systems'. Together they form a unique fingerprint.

Cite this