technology

Linkedin

Surface AI

At Quali Drone, we use Desupervised, which is an AGI (Artificial General Intelligence) software company, that has developed a Bayesian AlassS platform, that makes it possible to build, train, deploy and maintain smart costumer-safe Deep Learning algorithms with less data and includes domain knowledge, eliminating many all the challenges with the known machine learning methods. Desupervised’s mission is to develop and implement complex Bayesian Deep Learning algorithms for real-time quality validation and approval of the images taken by the drone. It can also identify, locate and categorize surface defects, combined with precision measurements of top flange for use in proper industry documentation and positing of the drone relative to the inspected structured CAD-drawing for identification of start position of the autonomous drone flight. These are all areas where this type of next generation AI never have been applied before.

The use of AI in inspections of TPs (Transitions Pieces) enhances the process’ efficiency and quality. However, machine learning algorithms today are basically unintelligent and have no common sense to verify if the predictions they make are right or wrong. Instead, the algorithms predict what the answer most likely is, based on the data they are trained with.

If the algorithms are trained on an insufficient basis, the quality of the predictions are poor even though the algorithm will claim that is very sure on its prediction. This makes machine learning unsafe to use in general, and in this case specifically in relation to identifying the abnormalities they should find, leading to the risk of errors in the process they are part of. At the same time, the algorithms require massive amounts of training data to achieve sufficient accuracy in their predictions which takes time to produce, making them both expensive and slow to develop and train.

That is why we at Quali Drone use Desupervised as our AGI software instead of machine learning algorithms.

Autonomues flight routes

Manual inspections of TPs for offshore wind turbines are time-consuming, require expensive machinery and need to comply to high safety standards. The results in the lack of positive documentation for each TP and limited tracking changes during the production process.

By implementing drone technology with cloud connectivity, it will enable autonomous drone flights and the possibility to utilize on-board sensor technology to navigate in a challenging environment. Furthermore, drone flights are currently performed manually which complicates data comparison due to flights not being able to be performed identically. By using sate of the art technology, it is possible to develop and program drone flight routes to implement in drones best suited for the specific job of either error-finding or measuring with artificial intelligence.

To make autonomous drone flight possible, the flight route of the TPs is developed based on CAD-drawings of the structure, information on the camera and lens as well as the surroundings of the TP. The position of the drone for each photo is calculated to ensure that the entire TP will be photographed in an optimal flight pattern. Based on the position of the TP (or other structures), the drone position is the recalculated into real-time GPS coordinates together with information on angle, position and focus of the camera. All this required information is transferred to the drone to ensure the drone being able to perform the flight route. The flight route is programmed specifically for the drone, making it platform independent, which means it can be integrated to the drone best suited for the task across manufacturers.

Photogrammathy

Photogrammetry-based coordinate measurement is utilizing targets (white dot on black background) and coded markers to relate multiple photos into a 3D coordinate system. The targets can also be used as measuring points with very high accuracy. It is also possible to identify contrast features for measurement, but the feature is less accurate. This area of application is well-established and used in various industries.

This project aims to enable high precision measurement of hole structures and/or parts of it without need for manually added targets and coded markers to the structure, as it is done by AI. This can then verify that the structure is built within the tolerances of the project.

Digital Twin

The wind energy industry can be transformed significantly by embracing the advanced digital technologies such as digital twins as well as sensor technologies and drone inspection solutions. The project will contribute to digitalise the design, manufacturing, inspection and installation process of support structures for offshore wind turbines (and other structures).

There exist many definitions of what a digital twin is, but according to Bolton et al. (2018)1 “a digital twin is a dynamic virtual representation of a physical object or system across its lifecycle, using real-time data to enable understanding, learning and reasoning”. A digital twin is dynamic and therefore changing as the physical twin is changing. With the inspection data stored in a digital twin, it is possible to automate and improve quality control and performing it in a cheaper and more efficient way.

Data management

Drones must operate under difficult conditions, as wind, rain and sunshine can challenge the autonomous drone flight as well as the image capture. Heavy gusts destabilize the drone, raindrops complicate image quality, whereas reflections from structures in high sunshine distort images.

Furthermore, performing drone operations at harbours have major consequences for positioning based on GPS due to compass interference, GPS drifting, and multipathing complicates autonomous drone flights. While automating drone flights, and with respect to the above conditions, the software will autonomously perform flight routes around generic structures, angling of the drone camera of point-of-interest and focusing the camera lens in order to provide high-quality images for inspection matters. This also includes development of a mission planning interface, development of precise drone positioning e.g., based on computer vision, data management, and integration to an annotation tool for data processing.