Case: AI Pipe Inspection

Case: AI Pipe Inspection

In today’s fast-paced world, it’s easy to take plumbing for granted. However, a lot of effort and time goes into making sure our sewer system is functioning properly. Furthermore, if disaster strikes, the result can be catastrophic; drinking water can become contaminated, resulting in sickness and a lack of clean drinking water. One crucial aspect of preventing such a disaster is thoroughly inspecting active sewer pipes, to properly estimate the state of the pipe, so that correcting actions can be taken, if needed.

Traditionally, this inspection process has been rather time-consuming. An employee has to maneuver a camera attached to a long cable through the pipes. Next, the video had to be meticulously manually reviewed by an employee, which in turn marks the parts of the video which show pipe abnormalities. Pipe abnormalities can be anything from cracks to dead animals, but are generally some fault or special feature of the pipe. Mindified was asked to develop software that could optimize the inspection of pipes by a Swedish pipe inspection company.

Firstly, to optimize the pipe inspection procedure Mindified developed software where the employees could view the videos from these pipe inspections, pause, and mark where on the video the pipe abnormalities occurred. After the inspection, the employee could generate a pdf of a pipe inspection report, using Mindifieds program, based on the pipe specifications and where on the video the employee marked the abnormalities.

Secondly, Mindified trained a Neural Network to predict where in the pipe inspection video an abnormality occurred. Unfortunately, the customer had limited access to data, however, the web GUI created by Mindified could be used to generate more data, which in turn could improve the performance of the model with further training.

In summary, Mindified created a prototype that could be used for the pipe inspection company to conduct internal testing and possibly optimize its workflow.

For more information about our work on this project, please contact: