Evaluation of ANN model for pipe status assessment in drinking water management

Abstract

Non-revenue water due to pipe leakages presents a significant global challenge, impacting both the economy and environmental sustainability. The current approach to pipe management for water utilities in Sweden is mainly reactive; leaks are repaired when detected, sometimes with large costs if the leakage is extensive and critical. With this study, we want to focus on proactive pipe network management by using an Artificial Neural Network (ANN) model to estimate the probability of leakage in water pipes. The ANN model was trained on leaks that occurred over 10 years. A comparison with leaks reported after the training shows that the model succeeds in identifying groups of pipes with a higher leakage frequency. Evaluation of both new and historical leaks in four different water pipe networks in Sweden showed that a higher prediction value from the ANN model was linked to a higher occurrence of leakage. This indicates that the ANN model succeeds in identifying some of the combinations of attributes that lead to leakage. An evaluation of the input attributes in the ANN model found that the most important attributes for leakage prediction were pipe material, pipe age, adjacent problems on the pipe stretch, pipe length and pipe dimension. HIGHLIGHTS An ANN model for pipe status in four Swedish water utilities has been assessed.; The ANN model demonstrates effectiveness in predicting leakages.; There are clear relationships between prediction values and reported leakages.; The most important attributes for leakage prediction are material, age, adjacent problems, length, and dimension.; The ANN model can identify potential leakages and enable proactive maintenance strategies.;

Keywords

• ai
• artificial neural networks
• asset management
• pipe renewal
• water mains
• water supply