The Amazing AI Race: Pit Stops, Detours, and Red Flags

The Amazing AI Race: Pit Stops, Detours, and Red FlagsAbstractAcross North America, the National Association of Sewer Service Companies (NASSCO) has pioneered the development of a standardized observation and defect coding system for underground infrastructure systems. Using NASSCO's the Pipeline Assessment and Certification Program (PACP), inspections can be evaluated in a consistent and defensible manner. Standardization serves as a cornerstone for establishing benchmarks and comprehensive guidelines for assessing pipe's current condition and tracking deterioration over time. Using PACP, allows owners, engineers, and operators to discuss pipe condition using the same language, which minimizes the risk of misinterpretation. In keeping with the broader trend seen in many industries, gravity sewer condition assessment has seen advancements with artificial intelligence (AI) coding. While NASSCO's coding system provides consistent guidelines for certified professionals to use, it doesn't account for experience level and bias. These factors can attribute to missed or improperly documented defects. However, AI technology is not intended to completely replace human interaction with closed circuit television (CCTV) and condition assessment. The goal would be to help employees work more efficiently, allow utilities to better allocate resources, and to produce more consistent condition assessment results to help track pipe deterioration over time. The extent to which AI can impact a sewer maintenance program depends on the unique goals and structure of each utility. Some utilities have an abundance of CCTV comprising information from various sources. This data can be from historical inspections that don't meet NASSCO standards, new construction acceptance assessments, cross-bore investigations, or routine inspections done in house or by a contractor. AI can be used to bring historical data up to date with modern coding standards quickly and effectively. For utilities facing staffing constraints, they can utilize AI to process and code videos to be flagged for emergency repairs, cleaning issues, or long-term construction planning. For more robust utilities with capital improvement programs, AI can help streamline processes and help drive decision making tools. Ultimately, AI can upgrade a utility's potentially underutilized data into a wealth of knowledge for stakeholders. For one case study, AI was introduced into a robust program. Historically, CCTV data was collected and coded in the field by operators. Office staff would review all the codes for accuracy and update accordingly. The codes were then processed through an algorithm to produce rehabilitation, replace, or repair (3R) recommendations and estimated costs for repairs. As a potential cost and time saving measure, AI was introduced between when the data is recorded and when the data is coded. The utility contractors will submit uncoded data to the AI and the AI will process the videos and provide PACP codes. Compared to historically coding review, there was an 80% reduction in review effort required by office staff. The AI generated codes are still run through the same 3R algorithm and costs are produced. When comparing human coding to AI coding, two key points stood out. AI detected more structural codes than a human coder, and according to the 3R algorithm, those codes in general were higher risk and indicated a more robust repair. A more detailed condition assessment must be accounted for in future budgeting as the utility considers adopting AI. While there are significant costs savings during data collection and processing, more capital funds are potentially required for system improvements. Modifying the 3R algorithm can help to accommodate for those increased costs, but there can be downsides to changing the algorithm. Another case study introduced AI in part to keep up with the amount of 360 data they are collected via an untethered large diameter floating camera. This data is collected with basic header information and was previously left for engineers to process. The goal for the AI is to code these inspections so that the engineers can prioritize which inspections they are reviewing based on structural or operation and maintenance (O SewerAI 2;SourceProceedings of the Water Environment FederationDocument typeConference PaperPublisherWater Environment FederationPrint publication date Feb 2024DOI10.2175/193864718825159289Volume / Issue Content sourceUtility Management ConferenceWord count11