Comparison of AI with Optimized QC vs Human Condition Assessment of Sanitary Sewers in Boston, MA

Abstract

The Boston Water and Sewer Commission (BWSC) collects 10-miles (16 kilometers) of closed-circuit television (CCTV) inspection videos per year. These inspections are performed by a combination of in-house staff and contracted resources. Manual coding of CCTV inspections by different resources can be subjective. To address this, in-house engineering staff review inspection videos to confirm the identified reportable observations and to develop action plans when needed. To gain a better insight into the subjective nature of manually coded reportable observations, a side-by-side comparison was completed to quantify differences between manually coded reportable observations and those identified using artificial intelligence (AI) software (DragonflySM) plus optimized quality control (QC) review. Optimized QC review was performed to ensure that AI coding was meeting high quality National Association of Sewer Service Companies (NASSCO) Pipeline Assessment Certification Program (PACP) output. The side-by-side comparison included a sample size of 20 pipe segments totaling approximately 4,000 linear feet (1,219 linear meters) of varying pipe materials, shapes, diameters, age, and condition. Results from the comparison showed that AI software with optimized QC review offered a consistent and accurate alternative to manually reviewed CCTV inspections and was able to correctly identify reportable observations by human inspectors, while also identifying major and nuanced reportable observations (e.g., cracks and fine roots) that may have been overlooked or disregarded by human inspectors. The comparison demonstrated how AI with optimized QC review of CCTV pipeline inspections can provide more consistent pipeline condition grading of both structural and maintenance defects, which can affect the prioritization of linear assets requiring prompt intervention. This presentation on leveraging AI for visual detection and recognition of reportable defects for pipeline inspections, and the importance of utilizing an optimized QC process to rapidly verify AI outputs, will assist utilities with streamlining operational procedures, allocation of resources, and capital planning.