March 2022 NPN Webinar

Led by one of CNAM’s new professionals, this webinar will showcase emerging research and analysis being undertaken to advance the asset management body of knowledge. This event is an opportunity for new and seasoned professionals alike to gain new knowledge in a particular area of research.

The webinar will run on March 24th from 12:00 pm – 1:30 pm (ET).

You will receive your Zoom link with instructions the morning of the event.

Details

Comparison of Machine Learning Models for Prediction of Water Main Deterioration Across Canada

The failure of water mains jeopardizes the essential delivery of clean water and the safety of water users. To address these risks, previous studies have proposed a variety of predictive models. These models have been developed for only one or a few case study systems and the factors considered depended on local data availability. The purpose of the present study is to compare the accuracy of machine learning models in predicting water main deterioration across Canadian water networks. The study was conducted by the UrbanLinks research group of Concordia University, in collaboration with the National Water and Wastewater Benchmarking Initiative. Water main inventory and break data was collected from thirteen Canadian water utilities. A variety of factors, including intrinsic, environmental, and operational, were compiled.  Random forest (RF), artificial neural networks (ANN), extreme gradient boosting (XGBOOST), and logistic regression (LR) were compared. While results were generally similar, XGBOOST was found to perform the best for most systems (9 out of 13) because it is able to identify intricate relationships between factors. Models were less accurate for systems with less break data or for materials such as PVC which have experienced fewer breaks. Overall, physical and protection attributes, including material, length, lining status and age were found to be key predictors of main breaks. And for systems that had data on pressure and pipe roughness those were also found to be important predictive factors. Because the models provide probability of failure results for each pipe in the system, they can easily be mapped and facilitate the visualization of deterioration. 

 Learning Objectives

1. What information is being collected on water main breaks by different systems across Canada?

2. What are the state-of-the-art models applied in predicting water main deterioration?

3. What are the best models to be applied in predicting water main breaks in Canada?

4. What are the main factors driving the deterioration process? 

5. What additional data should be collected to better understand and predict water main deterioration?