This article was reviewed in accordance with Science X’s editorial process and guidelines. The editors highlighted the following attributes while ensuring the credibility of the content:
fact-checked
trustworthy source
Proofreading
OK! The PIG is lifted into a sewer for inspection. Photo credit: Leiden University
× close
The PIG is lifted into a sewer for inspection. Photo credit: Leiden University
They only come to mind when they become damaged and then become a huge problem: our sewers. Your maintenance could be much faster and more accurate, Ph.D. Candidate Dirk Meijer found out. Even deep underground, algorithms are proving to be a stroke of luck.
Sewer inspections are usually carried out by specialized companies. A small electric car (PIG) with a camera mounted on it is sent into the sewer. An inspector then assesses whether there are any cracks, leaks or other problems. And this is exactly where, says Meijer, things often go wrong. “If you show two inspectors the same sewer on different days, they will find different things. Companies are often paid by the hour or by the mile, and that can get in the way of accuracy.”
Individual assessment
A second problem is the way reporting is currently done. The severity of damage to sewer pipes is now classified from 1 to 5, with number 5 being given for the most severe damage. But in reality, these numbers are used to decide whether or not a pipe should be replaced, says Meijer. “If a slightly damaged pipe poses a threat to groundwater, an inspector will mark it as a 5 because the pipe needs to be replaced quickly. But that is an individual assessment.”
And that shouldn’t be the case. Through automation, inspections should be carried out more consistently and with a lower error rate. Using machine learning, inspection reports and images of around a thousand kilometers of sewer pipes, Meijer worked on algorithms that can help identify problems. “The algorithm filters the inspection images for vulnerabilities in the sewer system and then it is up to an inspector to review the selected images.”
Shadow or crack?
This check by an examiner remains necessary for the time being, because even trained computers make mistakes. “We know our model contains false observations because we have fed it the reports of inspectors who make human errors. For example, let’s take a shadow in the image of a sewer pipe that is mistaken for a crack. To get the algorithm to inspect better than humans, more research is needed. But the algorithm can already make inspections less time-consuming.”
Meijer hopes that companies will also apply his research, because a major challenge awaits the Netherlands. “Concrete sewer pipes have an average lifespan of around 60 to 80 years, which means that most sewers in the Netherlands need to be replaced. There will be a real boom in inspections and they could be carried out much more efficiently.”
Comments are closed.