Verification, Validation and Processing of Infrastructure Drone Inspection Data

Visual inspection of large infrastructure assets in extreme environments, such as offshore wind turbines, is essential for maintaining performance and ensuring safety. However, it is expensive, labour intensive, and hazardous, with the majority of inspections being carried out manually by inspectors using rope access or boom lift. Currently tens of millions of man hours per annum are spent looking for defects in safety critical infrastructure.

Automatic visual inspections using robots equipped with cameras offers a potential solution, significantly reducing costs, increasing quality assurance and reducing safety concerns. Major progress has been made in capturing images for inspection, notably using drones, and in some cases, these have advanced to the point of being technically viable alternatives to traditional manual inspections.

However, although the capture of images has been automated, the processing of images in order to identify defects remains a semi-automated process, with reliance on visual inspections of image data by trained experts. This means that although the visual data can be collected automatically and quickly, inspections still take a significant amount of time. Hence there is a need for fully automating the data processing component if these systems are to fully realise their potential.

Alongside this there is also a need to garner wider acceptance of automatic processing techniques across the industry. They need to be commercially viable and demonstrably reliable - adoption of new approaches to inspection relies on a full understanding of their limitations and the criteria by which they can be assessed and categorised. This is necessary if such methods are to be acceptable to current and future regulatory requirements. Addressing this issue is therefore of equal importance to technical development.This project aims to address both of these issues. It is a collaboration between Garrad Hassan (DNV-GL), world leaders in risk management and quality assurance, Perceptual Robotics (PR), an SME specialising in visual inspection of wind turbines using drones, and the University of Bristol (UoB), experts in computer vision and AI. An automated processing pipeline will be developed and demonstrated, and incorporated within PRs Dhalian system, a world leading semi-automated drone inspection system for wind turbines. Alongside this, a general framework for assessing, characterising and comparing such systems will be validated and verified, with the aim of generating broader acceptance across the industry and informing future regulation. The project will provide both competitive advantage to PR and contribute to growth of the UK automated inspection industry.