Machine Learning Approaches for Reliability Prediction in Wastewater Pipeline Networks

The improper use of water resources throughout the world is one of the main topics in advancing for better sustainability. The 6^th Sustainable Development Goal (SDG) of the United Nations aims to ensure availability and sustainable management of water and sanitation. However, according to the SDG report published in 2025, only 56% of domestic wastewater is safely treated. With the current rate, the world will not achieve sustainable water management until at least 2049 [1]. For this reason, the optimal synthesis of wastewater treatment (WWT) networks is an important challenge that needs to be improved upon. Current approaches for synthesizing WWT networks rely mostly on cost optimization without looking into other important metrics such as reliability [2] [3]. In process systems, reliability refers to the probability that the system performs its intended function correctly despite component failures [4]. The current methods for estimating reliability are based on historical observations and expert opinion. However, even experienced experts can have a biased approach in looking at reliability. Additionally, this increases the investment of both time and money to fully evaluate an entire network.

In the past, our team members developed a failure probability estimation via ML classification models [5]. However, this approach was limited to that specific plant and lacked predictive and extrapolation capabilities for other similar WWT networks. In this context, machine learning (ML) models for regression can be used to minimize bias, improve the efficiency of obtaining reliability values, and lower investment costs. We implemented ML methods such as eXtreme gradient boosting (XGBoost), Neural Networks (NNs), and Random Forest (RF) to calculate pipeline network’s reliability based on normal available features (pipeline length, material, installation year, etc.) in the WWT network. This method allows us to calculate reliability of pipeline networks in an efficient way which enables companies to anticipate and prevent process disruptions thus maintaining safe and reliable operations. These reliability estimates in combination with Graph Theory allows for the prediction of WWT networks with additional pipes or units that increase the overall utility and functionality.

References

[1] United Nations, “The Sustainable Development Goals Report 2025,” United Nations, New York, Aug. 2025. Accessed: Sept. 22, 2025. [Online]. Available: https://unstats.un.org/sdgs/report/2025/The-Sustainable-Development-Goa…

[2] A. B. Smith and R. W. Katz, “US billion-dollar weather and climate disasters: data sources, trends, accuracy and biases,” Nat. Hazards, vol. 67, no. 2, pp. 387–410, June 2013, doi: 10.1007/s11069-013-0566-5.

[3] United States Environmental Protection Agency, “Drinking Water Infrastructure Needs Survey and Assessment,” Environmental Protection Agency, United States, EPA 810R23001, Sept. 2023. Accessed: Sept. 21, 2025. [Online]. Available: https://www.epa.gov/system/files/documents/2023-09/Seventh%20DWINSA_Sep…

[4] F. Friedler, Á. Orosz, and J. Pimentel Losada, P-graphs for Process Systems Engineering: Mathematical Models and Algorithms. Cham: Springer International Publishing, 2022. doi: 10.1007/978-3-030-92216-0.

[5] J. Stengel et al., “Systematic Development of a Machine Learning-Based Asset Management Tool for Wastewater Pipeline Networks,” ACS EST Water, vol. 4, no. 12, pp. 5555–5565, Dec. 2024, doi: 10.1021/acsestwater.4c00608.