How DoD can get faster insights to support improved readiness

The Defense Department has been spending an increasing amount of its budget on maintenance and support costs to keep its aging fleets of aircraft at a certain l...

August 15, 2023 10:29 am

5 min read

The Defense Department has been spending an increasing amount of its budget on maintenance and support costs to keep its aging fleets of aircraft at a certain level of mission readiness. And yet, according to a November 2022 report from the Government Accountability Office, it’s falling behind. Costs keep rising as the lifecycle of aging weapons platforms gets extended, and budgets struggle to keep up with inflation, ultimately raising potential risks to mission readiness.

Advanced modeling techniques have been cited as critical as far back as the early 1990s. However, the infrastructure and availability of the software, technical expertise, and data availability limited the adoption of the capability. Now with more readily available talent, software, and cloud-based data aggregation, DoD has the opportunity to point these sophisticated forms of data analytics at previously manual processes like fleet management. While industry is moving out with artificial intelligence, machine learning, and large language models, there are still professional techniques like modeling and simulation that yield itself highly effective, accurate, and expedient speed to insight and is a complementary tool within the arsenal of advanced analytic programs within the DoD.

In some areas, DoD is rethinking the way it’s using technology to realize and leverage the value of its data more fully. Rather than collecting and keeping that data stovepiped or stored locally by each individual organization, DoD can use specific analytics tools to increase the speed to insights. Recent improvements in the availability of technology, training, and data systems have made this possible, but it hasn’t achieved wider adoption across the enterprise yet.

“Many weapons platforms have been around for multiple decades. The approach to managing the supply chain, platform maintenance, repairs, overhauls, and upgrades has had to change over the years as technology inevitably advanced. The Department is simply at another evolution of these processes,” said Jon Barcklow, managing director at KPMG. “Think about what we were doing 40 years ago. So now, with the broad adoption of artificial intelligence, machine learning, lightweight modeling, and simulation, agencies can look at what they were doing more real-time, with fewer resources, and find optimization opportunities faster than ever.”

However, with new technology always comes new challenges. Sometimes the problem is too much data to model or overly sophisticated — and costly — AI models and infrastructure for processes where the eighty-twenty rule might apply. For example, Barcklow said an aircraft that is both being flown on active duty and still in production is highly complex. The aircraft has to maintain a certain level of readiness, which means regular maintenance and replacement parts, but the production requires those same parts. This aircraft is also being flown around the world as the fleet constantly moves, and partner nations are assembling their own fleets, even further complicating the supply chain. Additionally, in the absence of perfect data, compartmentalized data infrastructure, or missing digitized processes, most data models will require assumptions and variables.

“In the use-case of a complex weapons platform, this becomes a multi-million variable optimization problem,” Barcklow said.

To handle a problem at this scale, DoD is increasingly turning to artificial intelligence and machine learning. But that can be an expensive proposition, and it has its limitations. For example, Barcklow pointed out that black swan events like what we saw with COVID-19’s impact on the supply chain can cause even the most sophisticated models to break down. By definition, black swan events have very limited analogous examples from which to benchmark. In the case of COVID-19, predicting labor shortages and absenteeism due to infection rates was a challenge in predicting the impact to platform build and repair times. In this type of circumstance, we found that alternative modeling techniques were better suited to develop these types of projections.

“Data has the potential to almost become your enemy,” said Phil Sutton, director at KPMG. “If you rely heavily on and are very focused on just what the data says, you may be led astray and you’re going to have a potentially detrimental forecast. So, we often recommend taking an alternative approach, going back to the basic principles and processes of these systems.”

That alternative approach is rooted in traditional operations research techniques, like simulation. The idea is to use first principles-based techniques to test assumptions and calibrate off of those with varying levels of confidence. Sutton said nobody will get it quite right – who predicted COVID-19, after all? – but those varying levels of confidence can reveal more useful insights than AI or ML, which can only extrapolate on existing data.

The idea is to consider various “what if” scenarios and then find the connections and similarities rather than planning for each of them. That leads to discovering broader sensitivities and potential disruptions. For example, the pandemic disrupted the supply chain, but so could a major geopolitical conflict. Rather than trying to predict which black swan event is coming next, focusing instead on shoring up weaknesses in the supply chain can make DoD more resilient in either scenario.

So, in the case of that multi-million variable aircraft, the idea would be to optimize for just a few broad variables: fleet readiness, cost, risk to the fleet, and risk to production. Those all have to be balanced and optimized against one another. And there are various scales that have to be performed, too. For example, a ground vehicle is its own layer of risk. On a smaller scale, each of that vehicle’s individual parts, subsystems, and components carries its own risk. On a larger scale, DoD has entire fleets of this vehicle. Each of those layers of risk must be taken into consideration.

And this is where AI and ML can come into play: Barcklow said predictive maintenance is one example where those technologies excel.

“Every one of these techniques and technologies has their own place and best fit,” he said. “All of these things have a very specific purpose and an approach, and they’re always best suited towards specific types of problems. And it is very complimentary if you have someone who can help you think about getting all these technologies to work symbiotically.”