Automated Sewer Defect Identification for Rapid Sewer Inspection (ASDER)– An application of Privacy and Transparency-Focused Artificial Intelligence

英国有一个超过393,460公里的下水道网络(国际贸易部，2023年)，收集了超过9000个污水处理厂处理的超过110亿升的废水(DEFRA，2022年；《独立报》，2014年)。包括下水道大多超龄(超过100年)、快速城市化导致严重产能不足、频繁的极端天气事件和不当处置等因素，所有这些都会导致下水道间歇性坍塌、堵塞，尤其是持续不断的泄漏和相关的污染事件。因此，法院正在对2021年南方水务公司开出创纪录的9000万英镑的罚款(环境局，2021年)，目前正在计划对2.5亿英镑的罚款(DEFRA，2023年)。这些处罚是毁灭性的，收入大幅下降，因此，下水道业主现在急于避免泄漏。规避主要是通过频繁的检查，导致快速干预。然而，下水道检查过程效率低下，导致许多下水道检查公司难以处理积压。它包括将装有闭路电视摄像头的有线水平机器人通过检修孔或检查室送入下水道。该机器人是从一辆可以访问闭路电视视频的面包车上远程控制的。控制器是一名下水道检查员，它缓慢地移动机器人，绕着摄像头摇摆，检查每一米处的缺陷，并记录每一处缺陷。对于一公里长的下水道，这一过程可能需要10天。视频和文件在办公室进行评估，以便在另一天生成缺陷检查报告，从而导致11天(即15,840分钟)。下水道公司的多辆货车/检查员未能解决积压问题，因此，市场对高效的下水道检查程序的需求尚未得到满足，以便能够进行频繁的检查。该项目(ASDER)将开发人工智能和大数据模型，这些模型将在5分钟内自动从下水道检查视频中识别缺陷，并生成可用于验证结果的透明度的报告(值得信赖的人工智能)。这将意味着检查机器人可以以合理的速度(例如，6公里/小时)通过下水道进行视频捕获，在10分钟内完成1公里的下水道视频捕获。这15分钟的总时间比目前使用的平均11天的生产率提高了1000%以上，并将在短时间内实现更多的检查，从而帮助清理积压、降低成本、避免污染事件和相关罚款，以及增加可持续性等好处。

The UK has a network of over 393,460 kms of sewers (Department for International Trade, 2023) which collect over 11 billion litres of wastewater treated at over 9,000 sewage treatment works (DEFRA,2022; The Independent, 2014). Factors including the sewers being mostly overaged (over 100 years old), increased rapid urbanisation leading to acute under-capacity, frequent extreme weather events and inappropriate disposal, all cause the sewers to experience intermittent collapse, blockages, and particularly incessant leaks and associated pollution episodes.Sewer leaks/discharge cause pollution of 64% of UK water bodies (UK Parliament Committee,2022). Consequently, punitive fines are being issued by courts with record £90m fine for Southern Water in 2021 (Environment Agency,2021) and plans for £250m fines are currently in the works (DEFRA,2023). These penalties are devasting and revenue-decimating hence sewer owners are now desperate to avoid leaks. Avoidance comes mainly through frequent inspections that begets quick intervention.The sewer inspection process is however inefficient, causing many sewer inspection companies to struggle with backlog. It involves sending cabled horizontal robots with CCTV cameras into the sewer through a manhole or inspection chamber. The robot is controlled remotely from a van with access to the CCTV videos. The controller, a sewer inspector, moves the robot slowly, pans the camera circumferentially to check for defects at every metre, and documents each defect. This process can take 10days for one-kilometre sewer length. The video and document are assessed in the office to produce a defect inspection report on another day, leading to 11days (i.e., 15,840-minutes). Sewer companies' multiple vans/inspectors have not solved the backlog issue.There is therefore an unmet market need for a highly efficient sewer inspection process to enable frequent inspection. This project (ASDER) will develop Artificial Intelligence and Big Data models that will automatically identify defects from sewer inspection videos in 5-minutes and produce reports that can be used to verify results for transparency (trustworthy AI). This will mean that inspection robots can go through sewers at a reasonable pace (e.g., 6km/hr) simply for video capture, completing a one-kilometre sewer video capture in 10-mins. This total of 15mins represents over 1000% productivity increase on the average 11-days used currently and will enable a lot more inspections in a short space of time thus helping clear backlogs, reduce costs, avoid pollution episodes and associated fines and increase sustainability among other benefits.