Mainblades: Accelerating aircraft inspections with an AI-powered drone fleet
About Mainblades
Mainblades is an aircraft drone inspection company based in the Netherlands that uses a fleet of AI-powered drones to detect aircraft structural damage. Founded in 2017, it helps maintenance, repair, and overhaul (MRO) companies and airlines to improve inspection data, shorten turnaround times, and reduce maintenance costs.
Tell us your challenge. We're here to help.
Contact usA lack of graphics processing unit (GPU) availability hampered Mainblades' ability to develop the necessary software for its smart drone fleet. Google Cloud gave Mainblades the freedom to experiment and build the AI and ML models underpinning the fleet.
Google Cloud results
- Accelerate productization of AI-powered drones
- Cut GPU provisioning time from 10 minutes to two minutes
- Capture, store, and process over 200 detailed images per inspection
- Reduce widebody inspection time by up to nine hours
Building an industry-certified drone fleet from scratch
Flying is statistically one of the safest forms of transportation in the world, and the aviation industry puts a lot of work into ensuring that it retains an excellent safety record. A cornerstone of aircraft safety is the regular and thorough inspections and maintenance checks done at specific intervals, called A, B, C, and D checks. Checking a single widebody aircraft can take up to 12 hours, involving three maintenance, repair, and overhaul (MRO) technicians and their heavy equipment.
But when the aircraft is not being used, it's not earning revenue for the airline, and it's not getting travelers to their destinations. The checks themselves cost time and money, and so the pressure is on MRO companies and their technicians to minimize aircraft downtime. That's where extra sets of eyes and more intelligence can help.
"With our automated aircraft drone inspections, we can save up to nine hours of maintenance time and reduce costs. We help airlines fly their aircraft again a lot faster."
—Jochem Verboom, Co-Founder and Chief Technical Officer, MainbladesSpeeding up the aircraft maintenance process
Mainblades, a Netherlands-based provider of AI-powered drones, is on a mission to augment MRO technicians' ability to visually detect aircraft damage so they can ready the aircraft for flight as soon as possible.
"With our automated aircraft drone inspections, we can save up to nine hours of maintenance time and reduce costs. We help airlines fly their aircraft again a lot faster," explains Jochem Verboom, Co-Founder and Chief Technical Officer, Mainblades.
Thanks to the maturity of modern drone technology, the company can use off-the-shelf drones to conduct these automated inspections for every aircraft type in both indoor and outdoor settings. That means the company's engineers can focus more on developing and improving its automated flight and damage detection software. For example, its dent-and-buckle feature uses computer vision to help technicians spot structural damage. The drones then feed this information into a damage assessment report that pinpoints the exact location of structural anomalies. As a result, technicians can spend less time doing visual inspections and focus instead on performing maintenance.
This accelerated timeline is particularly helpful during unscheduled maintenance caused by, for example, lightning or bird strikes that end up becoming a huge operational strain, because unexpected downtime can disrupt tight airport schedules and thin airline margins.
While MRO technicians now have a new tool that they can use to improve their work experience, getting these smart drones to fly and work in the real world has been a group effort in artificial intelligence (AI) and machine learning (ML) experimentation and modeling.
Training drones to be smarter, faster
Today, AI and ML are increasingly important for every company and industry to create real-world impact with data, particularly a company with a high level of automation like Mainblades.
There were three big areas of research that Mainblades was pursuing before it productized its drones.
The first was damage detection through computer vision, where they needed to discover how well they could detect damage and determine the right image resolution for the model. The second was actively learning how to select the right image from a big pool of images. The third area was automated machine learning, where models were trained to match specific customer use cases and improve detection accuracy.
While they had started training these models on another cloud service, the low availability of GPUs on the service hampered the team's ability to meet their ML ambitions. When Mainblades received starter credits from Google Cloud, it allowed them to experiment and iterate toward creating a group of smart support drones.
The bulk of these credits was spent in the third area of the company's early research and development. "These starter credits were crucial for us to spin up virtual machines to train our eyeballs and store our test data, and it gave us a lot of flexibility to do research in that early phase of our operations," says Julian von der Goltz, Machine Learning Architect, Mainblades.
With high GPU availability on hand, their engineers could quickly productize their software and incorporate more cloud-based solutions into their architecture. Mainblades used Cloud SQL as the go-to solution for its database, and von der Goltz explains that the convenience of having it hosted by Google Cloud made it a first choice for the company.
"We can leverage managed services and a lot of support from the Google Cloud team. Even now, if we have an issue, we can just contact the team, and they will find someone who is an expert at Google Cloud for an answer. When we saw how easy it was, we asked ourselves why we didn't rely on their support sooner," adds Verboom.
He credits the support from Google Cloud's AI and infrastructure engineers to help Mainblades unlock the full power of its cloud infrastructure. The full range of solutions that Mainblades now uses helps it to conduct maintenance checks for its global customers, some of which can be happening on the other side of the world.
"When we were looking at other vendors to support AI experimentation, it took at least 10 minutes to provision a GPU. The problem is that we have to pay for that 10 minutes. With GKE, it only takes two minutes, which means we pay less."
—Julian von der Goltz, Machine Learning Architect, MainbladesMaking maintenance easier at any destination
Mainblades uses Cloud Storage to store image data captured by their drones, with each inspection yielding over 200 high-resolution photos. Imaging an Airbus A380, for example, can result in 10 gigabytes of images. But what if that plane is on the other side of the world?
"Google Kubernetes Engine (GKE) helps us scale up and down deployments to support a higher load when we finish an inspection and process the captured images. We have significant data transfer and need to scale out our backend to process these images faster and increase our resources to enable a better user experience," says von der Goltz.
In a real-world example, the Mainblades team successfully transferred images from a hangar in the Philippines to its servers in the Netherlands, processed them and predicted structural damage using only the necessary image resolutions, without having to download the whole data load. "In an environment without a strong internet connection, this is definitely a big win for customers in different areas of the world," says Verboom.
Innovating for global drone operations
As the team looks to further innovate its software, GKE, which handles 90% of the team's compute workload, has also been key in helping them scale up their AI experiments.
"When we were looking at other vendors to support AI experimentation, it took at least 10 minutes to provision a GPU. The problem is that we have to pay for that 10 minutes. With GKE, it only takes two minutes, which means we pay less. Then, if I want to train a new machine learning or AI model, I just need to submit a request for the number of GPUs that I need, and it's just provided. That's a really great benefit," says von der Goltz.
As the company works towards global recognition for its aircraft inspection capabilities, Verboom hopes that Google Cloud will help Mainblades to continue improving its machine learning capabilities. "With Google Cloud, we were able to build a product from scratch that always works and have it certified in the aviation industry, which is no small feat. This gives our clients a high level of trust and confidence in our technology," he says.
"With Google Cloud, we were able to build a product from scratch that always works and have it certified in the aviation industry, which is no small feat. This gives our clients a high level of trust and confidence in our technology."
—Jochem Verboom, Co-Founder and Chief Technical Officer, MainbladesTell us your challenge. We're here to help.
Contact us
About Mainblades
Mainblades is an aircraft drone inspection company based in the Netherlands that uses a fleet of AI-powered drones to detect aircraft structural damage. Founded in 2017, it helps maintenance, repair, and overhaul (MRO) companies and airlines to improve inspection data, shorten turnaround times, and reduce maintenance costs.