By Ian Hersey, vice president public sector, Falkonry
Predictive operations have been rapidly adopted in the commercial sector, with successful use case showing significant improvements in industrial operations, saving millions of dollars per year as a result. The technology has now reached a high level of maturity and the defense sector is ready to leverage it, as evidenced by recent deployments.
An example: in August 2019 it was announced that the US Air Force had completed a successful Phase I evaluation of machine-learning technology using Falkonry’s LRS system. During this first phase, critical operational data problems were explored within the Air Force community, which led to STRATCOM’s Joint Warfare Analysis Center (JWAC) decision to go forward into Phase II of the project. This announcement was significant because it means JWAC is now able to analyze massive amounts of operational data and discover patterns at unprecedented scale, which enables real-time inferencing.
Why the Air Force uses machine-learning technology
The Air Force and other military branches have many assets (think aircraft, vehicles, etc.) as well as multiple facilities and data centers that they monitor for security and efficiency. Leveraging predictive operations allows decision-makers to adopt a condition-based maintenance schedule. They can detect, predict and explain conditions preceding equipment failures, which can significantly reduce unexpected periods of downtime in a military environment, just as business owners do in the industrial space.
Predictive operations help the Air Force resolve problems before they become problems. Consider taking data from control systems and mirroring that with the real-world experience of actual faults (such as a valve getting stuck). Another example is an HVAC system in a data center. If this were to go down, the data center couldn’t function and it could cause significant disruption.
The ability to predict and prevent outages like this is critical to keeping operations running smoothly. (Sound familiar?) Furthermore, because our solution provides prediction transparency in the form of confidence values and explanation scores that show the most highly correlated signals, the operator is able to both automate business logic around the prediction and pinpoint which areas of a system are most likely involved in a warning state.
It’s important to note that while the Air Force is now empowered to identify and react to these precursors, they are doing so without requiring data scientists. This has been a significant hurdle for digital-transformation efforts in the past because data scientists are expensive, they are in high demand and it takes additional time to hire and train them in each specific operations area. Those hurdles are why the Air Force chose a system that can be quickly and easily deployed by internal maintenance engineers or facility engineers. By using a system that automates pattern discovery and the explanation of the discovered patterns, existing teams can focus on addressing the identified problems. The reality is, they know the operations better than anyone else; when a pattern or precursor is flagged by the system, they will be the most adept at identifying what it means.
Security: A top priority
As can be expected, one of the biggest concerns of the Air Force in utilizing predictive operations was ensuring the security of the machine-learning system. To provide the highest level of security for government requirements, the Air Force uses an “Air Gap” version of the system, a network-security measure to ensure the user’s network is physically isolated from unsecured networks, such as the public internet or an unsecured local area network. The machine-learning system is integrated within existing Department of Defense analysis platforms and can be securely deployed in both classified and unclassified environments.
Ensuring even more security, the Air Force has adopted a testing system where the first step is to load their data and apply it to one set of problems in an unclassified environment. Next, predictive models are created, where problems are tested and validated, and then they compare one model to another. This ability to test, validate and compare models enables the Air Force to confidently know that any given model is now ready to be deployed for live monitoring. In classified environments, Falkonry instances can also be deployed securely by using specialized classified-cloud services such as Amazon C2S, in addition to on-premises installations.
Predicting the future of predictive operations in the military
As we look to the future, it is clear the military is serious about leveraging the benefits of predictive-operations technology. Through the AFWERX funded technology adaptation process, Falkonry LRS has a “sole-source justification,” meaning the product can now be procured by any government department or national defense unit without having to go through a competitive-acquisition process. This ensures that any armed-forces customer wanting to deploy predictive-operations technology can do so in a much shorter timeframe than was previously possible, and will in turn, obtain quicker results.
Going forward, we expect this technology to proliferate throughout the defense sector and, most importantly, deliver significant operational improvements along the way, just as it has in the wider industrial arena.
