AWS Machine Learning Blog
How Nordic Aviation Capital uses Amazon Rekognition to streamline operations and save up to EUR200,000 annually
Nordic Aviation Capital (NAC) is the industry’s leading regional aircraft lessor, serving almost 70 airlines in approximately 45 countries worldwide.
In 2021, NAC turned to AWS to help it use artificial intelligence (AI) to further improve its leasing operations and reduce its reliance on manual labor.
With Amazon Rekognition Custom Labels, NAC built an AI solution that could automatically scan aircraft maintenance records and identify, based on their visual layouts, the specific documents requiring further review. This reduced their reliance on external contractors to do this work, improving speed and saving an estimated EUR200,000 annually in costs.
In this post, we share how NAC uses Amazon Rekognition to streamline their operations.
“Amazon Rekognition Custom Labels has given us superpowers when it comes to improving our aircraft maintenance reviews. We’re both impressed and excited by the opportunities this opens up for our team and the value it can help us create for our customers.”
-Mads Krog-Jensen, Senior Vice President of IT, NAC
Automating the document review process with AI
A key part of NAC’s leasing process is the validation of each of its leased aircraft’s maintenance history to determine the safety and operability of its component parts.
This process requires NAC’s maintenance technicians to validate a variety of key forms, with the collection of documents containing the maintenance history of each of the aircraft’s key parts, known as the maintenance package.
These maintenance packages are extensive and unstructured, often amounting to as many as 10,000 pages, and containing various types and formats of documents that can vary widely based on the age and maintenance history of the aircraft.
The task of finding these specific forms was long and menial, generally performed by external contractors, who could take as long as a week to review each maintenance package and identify any essential forms requiring further review. This created a key process bottleneck that added additional cost and time to NAC’s lending process.
To streamline this process, NAC set out to develop an AI-driven document review workflow that could automate this manual process by scanning entire maintenance packages to accurately identify and return only those documents that required further review by NAC specialists.
Building a custom computer vision solution with Amazon Rekognition
To solve this, NAC’s Director of Software Engineering, Martin Høst Normark, turned to Rekognition Custom Labels, a fully-managed computer vision service that helps developers quickly and easily train and deploy custom computer vision models tailored to any use case.
Rekognition Custom Labels accelerates the development of custom computer vision models by building on the capabilities of Amazon Rekognition and simplifying the key steps of the computer vision development process, such as image labeling, data inspection, and algorithm selection and deployment. Rekognition Custom Labels allows you to build custom computer vision models for image classification and object detection tasks. You can navigate through the image labeling process from within the Rekognotion Custom Labels console or use Amazon SageMaker Ground Truth to allow for image labeling at scale. Rekognition Custom Labels automatically inspects the data, selects the right model framework and algorithm, optimizes the hyperparameters, and trains the model. When you’re satisified with the model accuracy, you can host the trained model with just one click.
NAC chose Amazon Rekognition because it significantly reduced the undifferentiated heavy lifting of training and deploying a custom computer vision model. For example, instead of requiring thousands of labeled training images to get started, as is the case with most custom computer vision models, NAC was able to get started with just a few hundred examples of the types of documents it needed to identify. These images, together with an equal number of negative examples chosen at random, were then loaded into an Amazon Simple Storage Service (Amazon S3) bucket to be used for model training. This also enabled NAC to use Rekognition Custom Label’s automatic labeling service, which could infer the labels of the two types of documents based solely on their S3 folder names.
From there, NAC was able to start training its model in just a few clicks, at which point Rekognition Custom Labels took care of loading and inspecting the training data, selecting the correct machine learning algorithm, training and testing the model, and reporting its performance metrics.
In order for the solution to deliver real business value, NAC identified a minimum performance baseline of 75% recall for its computer vision model, meaning that the solution had to be able to capture at least 75% of all relevant documents in any given maintenance package to warrant being used in production.
Using Rekognition Custom Labels and training on only those initial images, NAC was able to produce an initial model within its first week of development that delivered a recall of 98%, beating its performance baseline by 23 percentage points.
NAC then spent an additional week inspecting the types of pages causing classification errors, and added some additional examples of those challenging examples to its S3 bucket to retrain its model. This step further optimized performance above 99% recall and far exceeded its production performance requirements.
Improving operational efficiency and increasing innovation with AWS
With Rekognition Custom Labels, NAC was able to build, in just two weeks, a production-ready custom computer vision solution that could accurately identify and return relevant documents at higher than 99% accuracy, reducing to a matter of minutes a process that previously took manual reviewers about a week to complete.
This success has enabled NAC to move this solution to production, removing key process bottlenecks in its aircraft maintenance review processes to improve efficiency, reduce reliance on external contractors, and continue to deliver on its 30-year history of technical and commercial innovation in the regional aircraft industry.
Rekognition Custom Labels can help you develop custom computer vision models with ease by simplifying key steps such as image labeling, data inspection, and algorithm selection and deployment.
Learn more about how you can build custom computer vision models tailored to your specific use case by visiting Getting Started with Amazon Rekognition Custom Labels or reviewing the Amazon Rekognition Custom Labels Guide.
About the Author
Daniel Burke is the European lead for AI and ML in the Private Equity group at AWS. Daniel works directly with Private Equity funds and their portfolio companies, helping them accelerate their AI and ML adoption and improve innovation and increase enterprise value.