A recent hackathon at DASA sought to blend data analytics with data security. While quite a challenge, it can be done, explains Tim Madden of BAE Systems.
How can analytics and data models be assessed while maintaining data security?
It’s not straightforward – which is exactly why it was the challenge facing teams from across industry and academia at a recent hackathon hosted by the Defence and Security Accelerator (DASA).
Using a dataset from three fleets of C130s – a four-engine turboprop military transport aircraft – entrants from companies including-Lockheed Martin and Airbus competed against universities and smaller start-ups, as well as a joint team from BAE Systems Air, CORDA and Applied Intelligence.
The dataset, which was hosted on the Oracle cloud platform and stemmed from fleets in the UK, US and France, was varied and impacted by complex access rules. For example, users from the three countries had different security clearances, roles, and employers.
In our winning entry, the team outlined how a future app might deal with the data security issues by separating data access from analytical code.
So, why we did win?
Lessons from the Winner’s Circle
Our approach found favour with the judges after we were able to successfully develop some predictive maintenance analytics housed in a Python web app in a 24 hour development sprint. The custom web pages offered real-time access to analytics, while always ensuring that no users could see any restricted data they didn’t have the access rights to.
Among other analytics, we demonstrated that planes involved in more "hard" landings (spotted in the data by rapid descents and less time spent at low altitude) have much higher rates of failure for undercarriage parts.
We were also able to consider cumulative flying hours per tail number between services, as well as a predictor for number of days a tail number would be out of action dependent on the type of fault. Each analytic was housed in an individual micro service, while the data was retrieved and fed to each plot depending on the current user's permission structure. Users could access all these insights without any risk of breaking data security.
The judges, who were made up of a selection of leaders from the civil service and military, were impressed by the possible expansion on this approach. There is currently a trend in defence platforms for open access to data, where multiple applications from multiple vendors can tie in to a common architecture in order to offer the most benefit.
This proposed solution for the C130 data would offer a common and open access point for both data access and analytics in the form of a RESTful API – an application programme interface that is akin to a receptionist pointing you to the right person in the building to talk to if they believe you are allowed to talk to them. This architecture will allow many other future applications from any other vendor to be built on top of the rich data structure present while access to restricted information is strictly controlled.
The first prize included a pitch slot at the Oracle Openworld 2019 conference, a tour of the Defence Academy for the winning team. Going forwards there is likely to be an opportunity to bid for further funding, via DASA to develop the ideas and solutions touched on in the hackathon. This may lead to a series of projects in the future on data access architecture for complex analytics.
There is also a desire to leave the dataset available on the Oracle cloud until March. Internal projects are ongoing in several areas to continue thinking about how analytics and machine learning can be applied to national level sensitive data, lending insights without revealing secrets.
Watch this space.
About the author
Tim Madden is a data scientist with BAE Systems Applied Intelligence