Smarter Systems, Stronger Cities: AI’s Role in maintaining Modern Infrastructure

At 7:00 in the morning, the ground under a busy roadway collapses. A car falls into the sinkhole, and by the time crews arrive, the entire roadway has been shut down, backing up traffic. Citizens flood the city’s phone lines with concerns. It’s a bad day for residents, commuters, local officials, work crews, and city engineers.

“Even though cities do everything right, catastrophic failures still happen regularly for infrastructure systems in Texas,” says Matt Stahl, P.E., Practice Leader, Applied Innovation for Pape-Dawson. “Although cities have the data, unplanned utility failures still can pop up when cities haven’t yet implemented predictive modeling that helps them flag problems early and make better decisions.”

Fortunately, things have changed. Today’s Artificial Intelligence (AI) tools can help predict and minimize such events. Pape-Dawson’s recent success in AI comes through careful experimentation in three main areas: vision AI, analytical AI, and network AI.

Vision AI

Also called computer vision or object detection, vision AI makes sense of unstructured data, like images, videos, and point clouds. Prior to joining Pape-Dawson, Matt led a team that trained a vision AI model on 14,000 different structural and operational defects in storm drains. The team applied the model to review municipal closed-circuit television camera footage and flag areas of concern that warranted another look by the on-site operator. “We kept human expertise and discernment,” he says, “but AI helped find high-risk issues that may have been missed, so we could review them.”

The results impressed Matt and his clients. Adding AI to human capability cut overall time by 50%, boosted productivity, and improved prediction accuracy. Matt sees opportunities to apply vision AI in detection, tracking, counting, and analysis tasks, “not taking the expert out of the loop, but as a way to focus experts’ efforts.”

Analytical AI

Rather than inspecting unstructured data like images, analytical AI works with structured data (such as historical failures or inspection results) that is already formatted as a table. Analytical AI replaces traditional, rule-based logic with lessons it learns from the data set itself.

In Fort Worth, Texas, a traditional prioritization system correctly identified about 55% of the infrastructure problems that would concern engineers, says Matt. “My team implemented an analytical AI approach to generate a forecast for critical conditions and events.” It improved positive identification rates of severe infrastructure problems to over 80%.

The return on investment can be substantial, he adds. “Every dollar spent in proactive rehab and repair typically saves two to three times that in emergency repairs.” This sort of predictive tool could have identified the location of that sinkhole for further investigation, he concludes.

Matt expects analytical AI to take hold in other use cases, too. “It can predict water main breaks, erosion, overtopping at low water crossings, water quality degradation, and other problems, so we can address them before it’s too late.”

Network AI

“Network AI allows us to more fully understand the impacts of failures and outages based on topology, dependencies, redundancies, and the way assets are connected,” Matt explains. It shows promise in pinpointing infrastructure impacts and timely response options because roads, water pipes, sewer lines, traffic signals, electrical distribution, work crews, and more can be represented as connected networks.

Importantly, network AI can dynamically highlight Consequence of Failure (CoF), which quantifies system impacts due to damage or malfunction. While traditional CoF predictions primarily consider geographical proximity and nearby land use, network AI also considers topology, adjacent assets, and a problem’s upstream and downstream implications. For example, a traditional CoF evaluation of a water system might prioritize a trunk main serving a high-density central business district over one serving a lower-density rural area. In contrast, says Matt, “The network AI perspective might bump the urban main down in priority if it has multiple rerouting and redundancy options for that pipe, and it might deem the rural pipe high-risk if it is the single delivery source for local customers.”

Matt concludes: “The traditional approach to CoF could miss many of the critical relationships that are key for operational decisions. Network AI gives utilities a powerful tool to capture those relationships and adjust the plan of action accordingly.” Network AI shows promise across the architecture/engineering/construction field, including in transportation, stormwater, capital planning, and disaster response.

The Future with AI

Senior Vice President Zubin Sukheswalla, P.E., CFM has seen some Pape-Dawson clients reluctant to deploy AI. “We help them see that AI can really make their lives easier by identifying hidden short-term efficiencies while setting them up for long-term programmatic success,” he says.

AI’s use will continue to grow. Currently, engineers can employ it when they have well-structured data and clearly defined questions, but that’s sure to change. “In less than five years, AI will take unstructured data and find meaningful predictive results,” says Vanessa McMahan, Vice President, Innovation. That means it will help predict the next sinkhole in plenty of time to prevent it.