(86b) Development of Predictive Modeling Tool for Wastewater Asset Modeling and Management

One of the most critical elements of urban infrastructure is the Water and Wastewater Distribution Network (WWDN) (Rahbaralam et al. 2020). With the steadily rising population, many utility companies are forced to expand their WWDN to meet growing demands while meeting sustainability and reliability requirements (Bakri et al. 2015). For the last two decades, the American Society of Civil Engineers (ASCE) has given the wastewater infrastructure fairly poor ratings. Ever since the start of the infrastructure report card system in 1998, the infrastructure sector has never been given a rating higher than a “D+”, demonstrating the continued ineffectiveness and deterioration of the infrastructure (ASCE 2017; 2021). While many aspects contribute to the low scores, the most influential factors are the WWDN’s age and asset deterioration (Mazumder et al. 2018). Depending on the infrastructure’s location, the system often experiences a wide variety of natural phenomena which can have a negative impact on the system (Mazumder et al. 2018). In addition, the last major modernization of the water distribution network was done when the Clean Water Act (CWA) was passed in the 1970s (US EPA 1972; Malm et al. 2013). Most pipelines have an average lifespan of roughly 50 to 75 years; therefore, many assets have used their expected lifespan if they were not rehabilitated (Raghuvanshi et al. 2017; Turner 2016). Since many assets are at their end of life, many utility companies have seen a rise in asset failures which result in large capital losses and negative environmental impacts. The commonplace solution in the industry is to use a reactionary asset management approach, where the utility company needs to wait until a failure occurs before fixing the asset (Baah et al. 2015; Fenner 2000). Figure 1 shows the basic steps to developing an asset management plan which is standardized by the United States Environmental Protection Agency (US EPA) (Allbee and Rose 2017). While this standardizes the development process, the guidelines are very vague and leave much room for interpretation in the industry. However, the main failure of this process is the human bias that is introduced in steps 2 and 3. The generated bias often leads to ambiguity within the reactive asset management plan and can result in even greater capital losses when unexpected events occur. To that end, by using computational models we can modify the reactive approach to become a proactive objective-based method to minimize the negative economic and environmental impacts seen in a reactive plan.

Methodology

The backbone of the asset management plan is solved using machine learning classification algorithms to develop a predictive model for the whole WWDN and which is applied to each individual asset. Predictive modeling plays a vital role in the future of the wastewater industry since the ability to forecast pipe failures is crucial for planning the strategic modernization of WWDN infrastructure (Rahbaralam et al. 2020). By using machine learning, we can minimize the human bias that is introduced into the reactive asset management plan, and develop a predictive model based on objective mathematical correlations. Using Python to develop a Random Forest Classification (RFC) model for the WWDN, we create a framework that can take in a dataset from industry and assess the top deterioration factors. The objective of this model is to identify high-risk assets, potential failures, and environmental consequences for each asset in the WWDN. By identifying these key components, we can develop a list of corrective actions to minimize asset failures, environmental impacts, and large capital losses. To allow for ease of use, we have also developed an application that has the predictive asset management model built-in its framework. This application allows anyone in the wastewater industry to use the predictive framework, without needing any programming knowledge.

Results & Summary

To test the validity of the framework, we have partnered with a utility company in the wastewater industry. The provided dataset is comprised of many variables such as the pipe diameter, material type, segment length, expected remaining life of pipe, and flow rates. After preprocessing the data, RFC is used to generate the predictive machine learning model. The model generates a failure score for each asset, which can be used to predict how likely the asset is to fail. In addition, we are able to see which variable contributes the most to the failure score. Once identified, we develop a list of corrective actions which target the variable with the highest impact. In this model, the variable which has the highest impact is the number of years since the asset was last inspected. Therefore, by developing an inspection schedule, and prioritizing high-risk assets, we can minimize financial losses on assets that do not need to be inspected. This schedule and model information is then put into a graphical user interface (GUI), along with pertinent asset information for the user in the industry to see how the WWDN is functioning as a whole. By using this GUI to its full extent, a user in the industry can use our objective-based predictive approach to minimize major economic expenditures while minimizing negative environmental consequences resulting from asset failures.

Acknowledgments

We thank the Atlantic County Utilities Authority for supporting and funding the research for this project. We also thank the Department of Chemical Engineering at Rowan University for their continued assistance with acquiring the tools required for this research.

References:

Allbee, Steve, and Duncan Rose. 2017. “Fundamentals of Asset Management Session 0 - Exectutive Overview.” US EPA. https://www.epa.gov/sustainable-water-infrastructure/asset-management-wo....

ASCE. 2017. “2017 Infrastructure Report Card.” ASCE. https://www.infrastructurereportcard.org/cat-item/wastewater/.

ASCE 2021. “Infrastructure | ASCE.” American Society of Civil Engineers. March 2021. https://www.asce.org/infrastructure/.

Baah, Kelly, Brajesh Dubey, Richard Harvey, and Edward McBean. 2015. “A Risk-Based Approach to Sanitary Sewer Pipe Asset Management.” Science of The Total Environment 505 (February): 1011–17. https://doi.org/10.1016/j.scitotenv.2014.10.040.

Bakri, B., Y. Arai, T. Inakazu, A. Koizumi, H. Yoda, and S. Pallu. 2015. “Selection and Concentration of Pipeline Mains for Rehabilitation and Expansion of Water Distribution Network.” Procedia Environmental Sciences, The 5th Sustainable Future for Human Security (SustaiN 2014), 28 (January): 732–42. https://doi.org/10.1016/j.proenv.2015.07.086.

Fenner, R. A. 2000. “Approaches to Sewer Maintenance: A Review.” Urban Water, Sewer Systems and Processes, 2 (4): 343–56. https://doi.org/10.1016/S1462-0758(00)00065-0.

Malm, A., G. Svensson, H. Bäckman, and Gregory M. Morrison. 2013. “Prediction of Water and Wastewater Networks Rehabilitation Based Current Age and Material Distribution.” Water Supply 13 (2): 227–37. https://doi.org/10.2166/ws.2013.011.

Mazumder, Ram K., Abdullahi M. Salman, Yue Li, and Xiong Yu. 2018. “Performance Evaluation of Water Distribution Systems and Asset Management.” ASCE Library 24 (3): 24.

Raghuvanshi, Smita, Vikrant Bhakar, Chelikani Sowmya, and K.S. Sangwan. 2017. “Waste Water Treatment Plant Life Cycle Assessment: Treatment Process to Reuse of Water.” Procedia CIRP 61: 761–66. https://doi.org/10.1016/j.procir.2016.11.170.

Rahbaralam, Maryam, David Modesto, Jaume Cardús, Amir Abdollahi, and Fernando M. Cucchietti. 2020. “Predictive Analytics for Water Asset Management: Machine Learning and Survival Analysis.” ArXiv:2007.03744 [Cs, Eess, Stat], July. http://arxiv.org/abs/2007.03744.

Turner, Jon. 2016. “San Simeon Community Services Districe - Estimated WWTP Life Expectancy Analysis.” Phoenix Civil Engineering, Inc.

US EPA. 1972. “FEDERAL WATER POLLUTION CONTROL ACT.” Office of the Law Revision Counsel of the U.S. House of Representatives, 234.