Pipe Dream: CEAT professors research development of AI virtual assistant

Pipe Dream: CEAT professors research development of AI virtual assistant

Wednesday, October 30, 2024

Media Contact: Kristi Wheeler | Manager, CEAT Marketing and Communications | 405-744-5831 | [email protected]

A municipality's ability to manage its sewer assets can be wrought with challenges, such as repairs not happening until things break or difficulties repairing infrastructure. Issues also include a lack of computerized data or a siloed asset data management approach.
Two College of Engineering, Architecture and Technology professors at Oklahoma State University are working to help alleviate these challenges as part of a two-year study.

Dr. Yongwei Shan, an associate professor of civil and environmental engineering, and Dr. Weihua Sheng, a professor of electrical and computer engineering, are working on an addition to a sewer system data management platform that was first created in 2023.
The platform utilizes a geographic information system (GIS) and a relational database as its backbone, allowing users to access the location, design information, current and historical conditions, and work orders of all sewer assets in a municipality.

The platform has been in the hands of the City of Stillwater for testing, and the professors are working to add a virtual, artificial intelligence assistant to it that will help improve the process used by cities to monitor sewer assets.

Shan said in 2018, he and other researchers worked on a National Science Foundation Innovation Corps project to explore what is needed for sewer asset management.

"We were tasked with interviewing those municipalities, contractors and engineering firms to see what is needed in this industry," Shan said. "Then, we later found out that for the majority of municipalities, they don't really have a good data management tool for the sewer infrastructure to help them make data-driven asset management decisions."

Shan said his inspiration for pursuing this research was that wastewater infrastructure in the U.S. was given a D+ rating in the 2021 American Society of Civil Engineers America's Infrastructure Report Card. He said reactive asset management, meaning things are fixed when they break, is common, which can be due to a lack of accessible data or data potentially not being accessible to all members of an organization.

The data management tool was created through Cowboy Technologies with a startup company called InfraTie Solutions, co-founded by Shan and Hossein Khaleghian, a former Ph.D. student of his and currently the chief research engineer for InfraTie. The company was created to commercialize OSU's intellectual property. Cowboy Technologies is a shareholder in InfraTie Solutions, as they gave pre-seed money to help the company develop its prototype tool. The company and OSU have signed an Option for Exclusive License Agreement.

This photo shows a map of sewer assets in Stillwater on a platform designed through InfraTie Solutions.

Once the virtual assistant is developed, it will also be tested by workers in the City of Stillwater's waste department.
The virtual assistant will help tie all the information from the GIS of a city's infrastructure and could allow quicker access to the information for workers such as city engineers and crews in the field on the same platform.

Through Large Language Models, the AI assistant is trained to better understand how humans naturally communicate. This allows users of the platform to ask the virtual assistant a question, via voice or text, and it will respond appropriately.

"We are developing a companion robot which can talk to people in their natural language," Sheng said. "Dr. Shan approached me and wanted to make this sewer asset management system more human-friendly. Maybe it's like a tablet or a smartphone, so the crew onsite can just talk to it and get the information they need."

The LLMs will be used for two purposes: the first is to correct human voice recognition errors, and the second is to be able to query information. An extensive database is required to query the data, and LLMs help map the natural language query the user asks through an SQL (standard structured query language) query.

This photo shows the hours of a sewer management staff after the information was asked for by a person to an artificial intelligence assistant designed to aid cities with the management of sewer assets.

The research project has two phases. The first phase is to use the LLMs to train the AI algorithms to accurately understand human interaction queries that will create SQL.

The second phase is to create the user interface and integrate those AI elements into the data management tool. Once the City of Stillwater tests the tool, it will then be fine-tuned based on those findings.

Shan said the goal of creating the virtual assistant is to help a city "get the information quicker so that it can save them time."
The platform can provide a better overall picture of sewer system conditions as well as different information on a specific asset to help municipalities make a more informed decision, Shan said. The platform that the City of Stillwater has used allows workers to monitor the status of a specific sewer system asset and make proper maintenance or repairs when needed instead of fixing something after it breaks.

Sheng and Shan will have a team of OSU engineering students working on this project, including a Ph.D. student, to help develop the virtual assistant and plan to get undergraduate students involved to train on LLMs and AI.

"I think this is a good opportunity for students to get involved in AI research," Sheng said.

Shan noted this type of research can motivate him to spend as much time in the lab as possible developing something that will aid cities in such an important area of infrastructure asset management.

"This is applied research, and I think what we are proud of is we are able to take the things that we're creating in our research lab and bring it to the practical field," Shan said.

Photos: Provided

Story by: Tanner Holubar | IMPACT Magazine

Public link

Please use the above public link if you want to share this noodl on another website.