Artificial Intelligence defect detection and condition assessment

Abstract

VAPAR has recently completed defect detection and condition assessment projects with Australian, UK and NZ water utilities on both sewer and stormwater pipe inspection footage. The collection of data from these projects yielded an assessment of the types and frequencies of defects present in pipes of different materials and diameters and the use of artificial intelligence to detect said defects. The outcomes of these projects have successfully demonstrated that artificial intelligence (AI) can meet the same level of service as humans at detecting defects, but also determined what additional factors should be taken into consideration for pipe grading maintenance activities. The current pipe inspection standards provide challenges for software and operators alike to adequately capture information from pipe inspections and make decisions on the most appropriate actions to take. This paper explores processes that can be used to achieve further insights from the information already collected by operators and utilities. A brief history The concept of AI has been around since ancient Greeks and can be traced back to philosophy describing how humans think, the term “Artificial Intelligence” was first officially used in 1956 at a Dartmouth College conference in Hanover, New Hampshire. In 1960 the US Department of Defense used the concept to design computers to solve problems using basic reasoning. It wasn’t until the beginning of this century where AI started to see broader adoption outside defence and research. Nowadays, AI implementations can be seen in industrial to consumer applications. This is predominantly due to the accessibility of data, converging of networks i.e. isolated network access via public connections, and costs of electronics. In short, AI is an algorithm/s that defines the output of specific data sets corresponding to some model or pattern. Pipe Application Although in the wastewater and stormwater sector there is recognition of the opportunities presented by AI, adoption is still behind compared to other sectors. One of the key opportunity areas for AI in our industry is for pipe defect identification and condition scoring of CCTV inspection footage. Page 1 of 5 No-Dig Down Under 2021 Artificial Intelligence defect detection and condition assessment VAPAR, Sydney, NSW Australia Inspections are generally completed using crawler robots with cameras to traverse the inside of the pipe feeding back the video for recording and evaluation. The video is often played back in an office environment and reviewed by a person for defects and hazards which may cause operational issues. Pipe condition is typically rated against two categories, structural and service defects. A structural defect is one which has an impact on the physical integrity of the pipe itself. Over time, structural defects may worsen significantly to the point that the pipe requires significant repair work, or even replacement to prevent complete failure. Commonly recognised structural defects include cracking, breaking, and surface damage. Service defects are those that have an impact on the operational capacity of a pipe, impairing the pipe's effectiveness to convey wastewater through the pipe network. Commonly recognised service defects include displaced joints, debris and root intrusions. Findings For the purpose of this paper, the results from approximately 70km of stormwater and sewer pipe were used. The CCTV footage was provided by local councils and municipalities from Australia, New Zealand and the UK. Certain standards are followed during sewer and stormwater pipe inspection including camera resolution, robotic crawler speed and title frame display information. The overlay information is relevant data about the inspection that is inserted at the beginning of the inspection video, e.g. date, address, diameter, material etc. Once overlaid, this data is embedded onto the footage and used by the condition assessor. In this study, the AI technology is used to recognise and interpret this information. Sewer Pipes From the sewer pipe footage that was analysed, concrete and vitrified clay were the most commonly identified construction materials, representing 45.25% and 42.02% of the dataset, respectively. Page 2 of 5 No-Dig Down Under 2021 Artificial Intelligence defect detection and condition assessment VAPAR, Sydney, NSW Australia

Techniques