Project on the use of indoor inspection data in the drinking water sector: research, survey and interviews show potential for deep learning and autoencoders.
The aim of the project was to determine the utilisation potential of internal inspection data in the drinking water sector. To this end, research was carried out into existing systems for analysing inspection data. In addition, an online survey was conducted, as well as in-depth interviews with water supply companies that agreed to participate in the survey. An exemplary evaluation process for video procedures was also implemented.
The focus of the research was on identifying both commercially available products and R&D applications that are used to analyse video data from sewer inspections in the wastewater sector. There are commercial products, but they are all cloud solutions and therefore require the inspection data to be uploaded to the cloud of the respective provider. What all applications have in common is that they are not only able to recognise damage, but also to classify it.
Transferring such a classification method to the drinking water sector is technically possible, but requires fine-tuning of a deep learning classifier with video data specific to drinking water.
Important prerequisites for this are the creation of a standardised evaluation catalogue and rules for annotating video data. The complexity is greater in the drinking water sector than in the wastewater sector, as pipes can be empty, partially filled or completely filled with water.
More material classes are used in drinking water than in wastewater and water chemical processes lead to an even higher, material-specific variance (due to incrustations, deposits, biofilm). This is exacerbated by reflections in partially and fully filled pipes.
An online survey was conducted among 500 water supply companies, with a response rate of approx. 18 % (79 water supply companies). The majority of the participating water supply companies were medium-sized companies (approx. 50 %) and 25 % each were small and large companies; the survey was analysed according to company size. The companies are primarily interested in leak detection methods and sonic methods, followed by camera methods. Interviews were conducted with a total of 25 water supply companies. In the interviews, the topics from the online survey were explored in greater depth with detailed questions.
