Perfecting Defect Coding and Recommendations Using AI: The Power of Optimized QC and Historic Data

Utilities across the country are dealing with aging wastewater collection infrastructure as well as increased sewer flows due to new developments. These pressures require careful system management. To assess collection system health and to find deteriorated pipes in need of rehabilitation or replacement, utilities perform condition assessments, typically conducted using pole cameras, acoustic equipment, or closed-circuit television (CCTV) inspections. CCTV being the most robust method as the entire length of the pipeline is able to be visually observed and assessed. Performing CCTV inspections and coding pipe condition defects is a time-consuming and often tedious task which requires defect coding training and refreshers. Even with training, managers need to assure inspectors are coding uniformly. It is common for two different defect coders to describe a defect differently and to apply different codes. Additionally, watching CCTV footage for hours is monotonous and requires long-term attentiveness as the coder scans the pipe for observations. This can be exhausting and lead to mistakes and staff turnover. To resolve these problems, the industry is moving towards automated coding of CCTV defects using artificial intelligence (AI). The benefit is that AI processes the camera feed and scans for defects without getting tired. The downfall is that while CCTV AI coding has progressed, AI coding is not yet perfect. Since utilities rely on defect coding's accuracy to prioritize maintenance, replacement, and management of their collection systems, AI CCTV coding companies have had to incorporate a layer of QC to ensure the condition assessment results of the AI output can be relied upon by the utilities. In our presentation, we will focus on the history of AI tools for the effective management of buried assets. We will look at two applications of AI: visual recognition AI, used for automated coding of CCTV defects on inspection videos, and predictive AI, used for determining pipe degradation based on existing defects. We will not only look at the benefits of AI but also look at the limitations. We will drill into the most difficult defect types for AI to recognize and the importance of quality control (QC). Through two case studies, we will compare AI coded pipeline condition assessments to inspector coded pipeline condition assessments. We will explore how AI and optimized QC was able to detect nuanced defects that were either missed, overlooked, or misclassified by the human inspectors. We will also look at the benefits of keeping and maintaining historic data records. Utilities should note that condition assessment is not a once-and-done process. Repeated condition assessments help track deterioration over time and help establish a reliable and resilient conveyance system. Condition assessment tracking also helps utilities adopt proactive maintenance cycles instead of just reactive maintenance. We will look at an additional case study where historic CCTV inspections were re-coded to enhance their asset management analysis, leading to significant cost savings. This presentation on leveraging AI for visual defect recognition and coding, importance of augmented QC and data driven asset management will assist utilities with streamlining operational procedures, allocation of resources and capital planning. These tools also help utilities to catalogue inspection data effectively essentially creating a comprehensive database that is easily accessible.