Biofilms in drinking water distribution systems (DWDS) challenge water quality, infrastructure and public health. Current monitoring methods often disrupt biofilms or lack spatial coverage. This study explores two novel, non-intrusive techniques to measure biofilm thickness: one based on heat resistance, the other on changes in hydraulic residence time. Experiments were conducted in a lab-scale DWDS simulator replicating realistic pipe conditions. Both methods were evaluated against traditional destructive sampling to assess accuracy. Results show that the residence time method yields consistent, reliable estimates closely matching physical samples, while the heat resistance approach shows greater variability and requires refinement. Sensitivity analyses further demonstrate that the residence time method is more robust under varying operational conditions. These findings highlight its potential for field deployment, offering a scalable and minimally invasive solution for real-time biofilm monitoring. This advancement could support improved water quality management through targeted interventions in actual DWDS environments.

Biofilms in drinking water distribution systems (DWDS) pose a critical challenge to water quality. If left unchecked, they can compromise the biological stability of delivered water and ultimately public health. Existing biofilm sensing techniques primarily focus on metabolic or genetic indicators of activity, often using local and destructive methods. While rich in information, such data are difficult to apply in developing practical biofilm growth models. Biofilm thickness, however, is a more representative and scalable metric for this purpose. Yet, limited research exists on non-invasive thickness sensing in DWDS. This study introduces two non-destructive methods for measuring biofilm thickness by leveraging changes in heat resistance and residence time. Heat resistance was evaluated using ambient and water temperature measurements, while residence time was assessed with a conservative tracer. Both techniques were tested in the Slimer experimental setup (50 m long, 13.2 mm diameter PVCp pipe) under realistic hydraulic conditions. Results showed a strong correlation between biofilm thickness and residence time drift, indicating flow disturbance as a reliable indicator of biofouling. In contrast, heat resistance sensing exhibited considerable natural variability, limiting its analytic value. The findings highlight residence time analysis as a promising, non-invasive approach for estimating biofilm thickness. This method offers continuous, non-destructive monitoring, enabling early detection of biofilm-related anomalies and providing valuable input for both laboratory and field applications aimed at enhancing DWDS resilience.

This study presents a collaborative framework developed by the Water Futures team of researchers for the “Battle of the Water Demand Forecasting” challenge at the 3rd International WDSA-CCWI Joint Conference. The framework integrates an ensemble of machine learning forecasting models into a deterministic outcome consistent with the competition formulation. The water demand trajectory over a week exhibits complex overlapping patterns and non-linear dependencies to multiple features and time-dependent events that a single model cannot accurately predict. As such, the reconciled forecast from an ensemble of models exceeds the performance of the individual ones and exhibits higher stability across the weeks of the year and district metered areas considered.

Researchers and engineers employ machine learning (ML) tools to detect pipe bursts and prevent significant non-revenue water losses in water distribution systems (WDS). Nonetheless, many approaches developed so far consider a fixed number of sensors, which requires the ML model redevelopment and collection of sufficient data with the new sensor configuration for training. To overcome these issues, this study presents a novel approach based on Long Short-Term Memory neural networks (NNs) that leverages transfer learning to manage a varying number of sensors and retain good detection performance with limited training data. The proposed detection model first learns to reproduce the normal behavior of the system on a dataset obtained in burst-free conditions. The training process involves predicting flow and pressure one-time step ahead using historical data and time-related features as inputs. During testing, a post-prediction step flags potential bursts based on the comparison between the observations and model predictions using a time-varied error threshold. When adding new sensors, we implement transfer learning by replicating the weights of existing channels and then fine-tune the augmented NN. We evaluate the robustness of the methodology on simulated fire hydrant bursts and real-bursts in 10 district metered areas (DMAs) of the UK. For real bursts, we perform a sensitivity analysis to understand the impact of data resolution and error threshold on burst detection performance. The results obtained demonstrate that this ML-based methodology can achieve Precision of up to 98.1% in real-life settings and can identify bursts, even in data scarce conditions.