How Mainblades uses machine learning for seamless aircraft damage assessment

How Mainblades uses machine learning for seamless aircraft damage assessment



There was a time when the term artificial intelligence (AI) invoked the images of an apocalyptic uprising. Let’s face it: it’s far from being the scenery depicted in Hollywood movies. In the last few years, AI has gained ground in numerous industries, offering a future full of possibilities. 

The aviation industry is no stranger to the potential of AI (and its subset Machine Learning) in air traffic management, airport security, customer services and damage assessment. Have you crossed paths with Spencer at Schiphol Airport or Pepper at Oakland Airport?

AI-driven drone solution for improved efficiency in aircraft inspections 

There is no secret that drone technology is the latest trend gaining traction in the aviation industry. The idea of using drones to inspect aircraft is becoming sought-after among lessors, airlines, and Maintenance, Repair & Overhaul (MRO) organizations, thanks to the latest work carried out by companies such as Mainblades.  

We strive to disrupt the way aircraft inspections are conducted, turning a tedious and hazardous task into an automated and efficient process. By combining state-of-the art drones, smart algorithms & Machine Learning, we provide an out-of-the-box solution to decrease the time of aircraft inspections by 75%. With our tool, aircraft engineers can collect, report, and share data in real-time at the touch of a button.  

From accessing aircraft logbook, checklists, or other publications, to reviewing data collected during the inspection, do you feel you are drowning in information? More importantly, how do you turn data into actionable insights? This is where Machine Learning comes in.  

Why machine learning matters  

In aircraft inspections delivered by Mainblades, the automated processing and analysis of data is a crucial part. This is possible by what we call our Machine Learning Stack, with data acquisition as its most important component.   

This means that we gather and label all images ever taken during aircraft inspections and their metadata (location, exposure, focal length, or the resolution of the photo, aircraft type, and environmental conditions). Based on these labels, the Machine Learning models will learn to perform object (=damage) detection, predicting its location and type in an image. To ensure correct predictions, this data must contain enough examples of damages (e.g., dents or lightning strikes) and cover all possible situations (e.g., outdoors or bad lighting). 

How Machine Learning and aircraft engineers co-operate to make an assessment

In practice, this means that, after collecting data, the Machine Learning models will make suggestions in the Mainblades Flight App, giving aircraft engineers the chance to make a final assessment. It happens in the background while the drone is flying around the aircraft and taking pictures. As soon as the picture is uploaded to our cloud-based system, aircraft engineers have immediate access to the found damages.  

Besides, by using a state of the art drone for dataset delivery, we can feed high-quality data to our algorithms but also detect  damages of about 2mm size. We are aiming at being able to automatically detect all damages from the Boeing Structural Repair Manual (SRM) – Damages that can occur during service with high true positive rate. In our proof-of-concept study, we achieved 95% true positive rate.  

Why human interference is needed 

As mentioned above, Machine Learning is a subset of AI, which algorithms learn from experience (=data) and are, therefore, highly adaptable. As Machine Learning algorithms still rely on the experience from humans, we use the domain knowledge of our customers in tight collaboration with machine intelligence to develop and continuously improve our own algorithms. Thus, instead of using aircraft engineers’ high-level knowledge and labor for the tedious task of inspecting an aircraft, we use it to improve our model for detecting deviations. It takes less time to label damages than walking around an aircraft looking for them!  

The way forward in Computer Vision 

By exploiting the recent advancements in Computer Vision, we empower our customers with accurate aircraft damage assessment, enhanced safety, and lower costs. Our drone’s ability to identify damages in complex real-world data that humans may struggle with makes it the right tool for visual aircraft inspections and, therefore, well-suited to the aviation industry. It’s an exciting time to be in this industry because there is scope for more – more innovations, more game-changing results! 

learn more about implementing inspection automation

contact us