Data analytics for detecting blockages in waste water pipe networks

Abstract

Waste water blockages cause significant harm and incur a financial cost to remove. This thesis includes a literature review touching upon waste water blockage formation, blockage detection and blockage management. Gaps in the literature around waste water blockages have been identified and presented. The methodology of this thesis, creating a computational dataset of blockage events and implementing novel real time blockage detection methodologies upon this dataset, has been explained. Two novel real time blockage detection methodologies have been introduced, one approaching the issue in the frequency domain using the coherence metric and the other using phase portraits. These real time blockage detection methodologies use real time data from level sensors distributed around networks to generate evidence of blockage with each time step of new data. Control rules are used to transition to an alarm once the sufficient evidence is gathered. The performance of these two novel real time detection methodologies has been explored for a range of independent variables including blockage severity, blockage detection time, measurement resolution, measurement noise, sensor distribution, rainfall, control rules and training data. Due to the lack of reliable real time field data capturing blockage events, a large computational dataset of 29464 hydraulic cases has been generated before training and testing the two novel blockage detection methodologies on this synthetic dataset. The results for these two methodologies have been compared and discussed. While both methodologies have been shown to successfully detect modelled blockages, the phase portrait method generally outperforms the coherence method.