Streamlining Sewer Inspection and Planning Processes with Artificial Intelligence

Managing buried sewer infrastructure requires ever-increasing time and financial resources to stay ahead of failures. While today's processes are well established, they tend to be reactive, manually intensive, and prone to errors, which results in inconsistent data. One such example is condition assessment of gravity sewers. The purpose of this presentation is to review an application of artificial intelligence (AI) enabled defect coding and how it has assisted First Utility District of Knox Count (FUD) in more efficiently inspecting and maintaining their gravity sewers. The audience will learn how AI analysis is applied for defect coding, leading to time and cost savings for the utility, as well as how the insights gained from the solution assist with more effective management of the system. Gravity sewers are generally inspected using a variety of inspection methods including pole cameras, acoustic equipment, and CCTV, with the latter typically being the most prevalent technique. Trained technicians view all the visual data as it is collected and assign specific names, codes, and severity levels to each defect observed. Engineers and managers then review the data provided by the technicians and decide which pipes need corrective or preventive maintenance and prioritize the next steps against available budgets. However, throughput is limited by the slow and tedious nature of the work, human output is highly error-prone, subjective, and inconsistent, and resources are limited and time-constrained resulting in a lack of information on the collection system and continually increasing sewer inspection backlogs. Hence, utilities are now turning to AI-based solutions for CCTV analysis, since these solutions can rapidly identify defects observed during inspection of the collection system, assign specific names, codes, and severity levels to each defect and free up time for staff to focus on other high priority/value tasks, enabling efficient management of the collection system. AI does not fatigue and consistently and accurately identifies sewer defects. As AI does not fatigue as humans, most AI based tools are consistent and offer over 95 percent accurate results. These tools save approximately 30-40 percent coding time as operators are expected to operate the camera and not spend time on coding. Due to the quick turnaround times, utilities can see at least 25 percent savings in their operations and maintenance costs. Some AI tools available in the market are cloud based for either automated or manual uploads or the AI assisted coding software is installed on CCTV trucks. Most of these AI defect coding tools are inspection software vendor agnostic, but only a few can provide engineering insights based on the defect scores using predictive analytics. Based on the utility size and preference, tools can be selected and used for their CCTV video analysis. This presentation will discuss how FUD used one of the AI defect coding tools called DragonflyTM in a pilot study to analyze 84,000 linear feet of gravity sewers. Figure 1 shows a comparison of the AI generated reports augmented by optimized QC process with human-coded reports for the pilot study. In most instances, the AI identified defects are comparable to experienced human inspectors but also identified a few that were missed by the human eye. Overall, for 162 pipe inspections, manual coding identified 75 defects, whereas the AI solution identified 131 defects. As with most AI defect coding tools, DragonflyTM tends to be more conservative and identifies more nuanced defects that help prioritize pipes based on their condition. Figure1: Structural grade by length This solution provided a more thorough condition assessment of the collection system and included detailed asset management insights that enabled cost-effective system management. The AI defect coding solution provided insights that enabled the staff to make quick decisions, which resulted in significant time and cost savings. An API was developed to allow seamless transfer of data to a secure location. Based on the successful completion of the pilot study, FUD decided to implement DragonflyTM for the District's sewer defect coding and condition assessment program for their entire collection system. FUD has a highly aggressive target to inspect their entire collection system of 478 miles of gravity sewer every five years. Adoption of the AI-enabled coding solution provides significant savings (75% reduction in cost for pipe evaluation), increases their productivity and efficiency in collecting and managing data, and frees up internal resources to focus on problem areas. In addition to high quality reports, for large utilities with sizeable sewer networks such as FUD, a cloud-based service that requires no software install and no license fees and automated CCTV video uploads on weekly basis enabled smooth transition from manual review to the AI based analysis. Since beginning the implementation, the AI-enabled defect coding solution has processed over 229,000 linear feet of gravity sewers for FUD. Figure 2 shows the insights provided by the tool (powered by Argon, an asset management tool that is integrated with DragonflyTM regarding rehabilitation/replacement for priority pipes, reinspection schedules, and maintenance schedules, effectively providing a roadmap to FUD staff for renewal of their system. The work is ongoing, with FUD uploading videos for coding roughly once per week or as the CCTV videos are obtained. As they inspect the remaining parts of their system, the knowledge gained will assist them with capital and maintenance planning, helping them to budget their repairs in the system at the assets in most need of renewal. Figure 2. Recommended Next Steps for FUD Based on AI-based Defect Coding