Conference Agenda

Much has been publicized in the industry about the potential for digital or “smart water” technologies to address modern challenges. Some impressive successes have been demonstrated by digital technology providers that are leveraging sensors, data systems and machine learning to optimize systems and backstop operator transitions. A digital roadmap to complement renewal, replacement and upgrade plans for physical infrastructure can help a utility access the value of digital tools, increasing resilience and allowing more to be done with fewer resources. To implement digital technologies, the water utility manager must navigate a myriad of different technology providers and products to evaluate, prioritize and select the right solutions, and then figure out how to implement, integrate, and manage through the associated changes necessary to realize the benefits of the new digital technology products.

This presentation will offer an overview of digital water technologies and their potential benefits, explain the scope and importance of a digital roadmap within the master planning context for utility infrastructure, and review methods for selecting and implementing technologies. Case studies will be reviewed including the following:

1. Deploying sensors in collection systems establishes historical data for use in planning activities while enabling conditions-based maintenance. Olathe, Kansas suffered from high I&I. They needed dependable sensors for data collection backed by software with analytical capabilities so they could assess their basins and determine where to achieve the highest impact in reducing I&I.
2. Leveraging sensors, data and machine learning can help manage collection systems and avoid overflows. Houston, Texas experienced frequent dry weather SSOs due the accumulation of FOG in random locations in their collection system. Sensors, a digital twin and machine learning were utilized to identify forming obstructions prior to overflows occurring so that crews could be dispatched to clear the obstructions.
3. Implementing a digital twin for a WRRF can improve communications between stakeholders and optimize sub-systems for less consumption of energy and chemicals. The utility needed to unify their team around data-driven decision making and capture institutional expertise ahead of expected retirements. A digital dashboard of the 70 MGD CAS WRRF was implemented for these purposes.

The Environmental Protection Agency’s (EPA) Lead and Copper Revision Rule (LCRR) requires utilities to submit a service line inventory for all services including the public and private side of the line in October of 2024. Many water agencies have invested time and resources over the past two years to establish an initial inventory, but many are faced with a daunting number of service lines with unknown materials. Physical inspection is the most reliable way to determine if a service line is lead, but often requires excavation and these inspections can be costly, ranging between $500 - $1,500 or more per property. Many communities are turning to digital methods such as Machine Learning to reduce costs and streamline the removal of lead service lines.

Collaborative Intelligence is a process in which Machine Learning is paired with human intelligence to enhance each other’s capabilities. Models excel at analyzing vast amounts of disparate data and identifying patterns; however, they need to be told which data should be included and how the output can be used.

This presentation will focus on the strategy and approach for training, using, and explaining Machine Learning to predict service line material with the goal of reducing the number of unknowns in the service line inventory in a cost-effective way. An initial model can be built using available field verification data, but if no verifications have taken place (or there are not enough to effectively train a model), data collected from historical records can be used to train the initial model. The initial model will be beneficial for targeting locations to perform further field verifications, and it can then be iteratively improved with additional field data from targeted and opportunistic inspections and replacements. We will look behind the scenes of this collaborative human and machine modeling process, highlighting techniques and approaches that help to boost model performance, ensure model reliability, and explain the outcome in terms homeowners and regulators can understand. We will present two case studies highlighting the differences in outputs due to quality and availability of model features and volume of training data (inspections).