Comparing Hepta’s solutions with helicopter-based inspections for a E.ON Group member in Slovakia

Západoslovenská distribučná, a.s. (ZSDis) is fully owned by Západoslovenská energetika, a.s. (ZSE Group) which is a member of the German energy E.ON Group. E.ON Group itself is one of Europe’s largest operators of energy networks and energy infrastructure.

ZSDis has more than 95 years of history and is the biggest energy distributor in Slovakia. It operates the distribution system, being subject to regulation rules under the relevant laws. As an independent legal entity, it has been active since 1 July 2007 – based on requirements of the European and Slovak laws related to the energy market liberalisation (ZSE Distribúcia, a.s., or ZSE distribúcia, a.s. until 1 January 2013).

ZSDis has been relying on helicopters in their power line inspections in the past, but due to their sub-par performance and quality, has been looking for better alternatives. This led ZSDis to Hepta and its power line inspection platform uBird. ZSDis wanted to see what could be achieved with a specialized power line inspection platform and custom made AI defect detection models. They were looking for solutions, that would raise the quality of inspections, would help to detect more defects and enable quicker inspection processes.

ZSDis and Hepta agreed to put uBird and its defect detection AI models to test, using images that were captured during previous inspections by helicopters. 35.1 km of high voltage line images were chosen to be analysed in uBird and the results would be directly compared to the inspection done by the previous service provider.

At the start of the project, ZSDis provided Hepta with their power line images on which ZSDis specific AI models would be created. Hepta’s specialists reviewed the images, the defects shown in them and requested clarifications, where needed. After reviewing the data, Hepta’s grid analysts started labelling the elements and defects, to start training the AI models on the dataset. Meanwhile, Hepta’s data scientists selected a set of images to benchmark the models against. During the training of AI models, Hepta’s data science team constantly improved their performance by creating a combined pool of data, including ZSDis’ imagery.

As an end result, Hepta created two new, customized AI models for ZSDis, which were able to achieve F2-scores of 84% and 75% respectively. Thus based on the existing dataset Hepta’s platform would be able to provide an immediate value of giving a better overview of defects with the added value of defect detection automation for a comparable expense as the current service.

The analysis of the images gathered previously by helicopter inspection teams proved to be challenging, as those images were taken from far and high, with no crucial details being visible. This is a common problem with helicopter-based inspections as they rely on a specialist taking manual notes of defects during the flights and data capture is not thoroughly focused on. Thus, if the specialist in the helicopter does not detect defects during the flight, it is quite difficult to detect them later on from images.

Due to the low-quality data gathered by the helicopter-based inspection teams, Hepta’s grid analysts were only able to conduct basic RGB image analysis. Although the images were of high definition themselves (with sizes of 60MB or 101MP), they were only taken from one side and from a high altitude. Photos taken only from one side of the asset tell only half a story and allow critical defects to be missed easily. And photos taken from a high altitude make it hard to get a clear overview of the severity of defects, especially due to the fact that most of the small details situate there.

A good top-down image used in power line inspections should be good enough to see separate wire strands. To guarantee the detection of the maximum number of defects, it is necessary to photograph from a closer distance and from different (lower) angles. When collecting data from a helicopter, this data quality requirement significantly increases the cost of flights. Using drones shows major qualitative improvements while offering a competitive price at the cost of new technology implementation efforts.

Although the data captured by the helicopter-based inspection team was sub-par by Hepta’s standards, the analysis process and AI model creation were carried out successfully in the uBird platform. Hepta’s grid analysts inspected 35.1 kilometres of high voltage power lines and 130 steel lattice towers for defects in 6 days. During that same timespan two new, customized AI models were created for ZSDis and tested successfully.

When comparing the analysis work done by Hepta in the uBird platform with the inspections done by the helicopter-based inspection team, it was concluded that not only did Hepta match the results, it surpassed them by a large margin. In total Hepta found 4.8x more defects and 4.5x more unique defects, than the previous inspection team.