Classification of wastewater manholes, procedures and opportunities by applying machine learning
Master thesis
Permanent lenke
http://hdl.handle.net/11250/2614636Utgivelsesdato
2018Metadata
Vis full innførselSamlinger
Sammendrag
Wastewater manholes are not subject to proper condition assessments in the same manner as pipes. Although manholes have a significant impact on the performance of the wastewater system, no official procedures for how to best operate them exists. To achieve such procedures, inspections and subsequent condition assessments need to be carried out. In order to achieve valuable inspections, a classification system describing possible defects and how they should be weighted is of tremendous advantage.
Here a classification system for wastewater manholes is proposed and tested on a set of manhole images acquired from Trondheim kommune. During the process of classifying manholes the classification system has been continuously altered such that the resulting distribution of condition classes seemed plausible. As former research on manholes is very scarce, the final classification system presented here is still only tentative and more research is needed to develop an official system for wastewater manholes, analogous to the system which exists for pipes (Norsk Vann, report 145)
Since the classification procedure for manholes is based on visual inspections, the possible use of computer vision as an aid has been assessed. Machine learning algorithms have been trained for detecting defects in the manholes, achieving promising results. How such programs work and the possible benefits to be gained are widely discussed, interesting opportunities for further research is also presented.