Armed with a 25mm M242 Chain gun, TOW anti-tank missile, 600 horsepower diesel engine and 3 crew members, the Bradley Fighting Vehicle is one of America’s best troop carriers. It can also break unexpectedly. Last year, to address this issue, the U.S. Army partnered with Uptake, an Artificial Intelligence / Machine Learning (AI/ML) company, to launch a predictive maintenance program pilot to better address wear in the mission critical vehicle. The aim of the pilot is to improve not only the Bradley maintenance program, but prove to the Department of Defense (DOD) that AI/ML can reduce spend and increase the effectiveness of all of its maintenance programs worldwide.
The DOD has an annual budget of ~$700 Billion, about a half of which is intended to be used on operations and maintenance projects (O&M), its single largest line item.[i][ii] Any scalable learnings and reduction in costs from AI/ML predictive maintenance projects will not only have a profound impact on the rest of the branches that look to the Army for best practices, but the American tax payer as well.
But the Army’s current approach to maintenance is inefficient. The military process for checking vehicle health is called Preventive Maintenance Checks and Services (PMCS). Soldiers are given a handbook used to inspect every visible element of the equipment listed in the handbook on a regular basis. After a long enough period of mileage or time, like a car, the vehicle is then taken into a shop where it is inspected with greater detail. There are two failures to this methodology. First, PMCS overlooks critical elements of a vehicle’s function, such as its internal combustion. Second, the Bradley is designed to be durable under duress, meaning problems can go undiagnosed for a long time with no loss in functionality. At best, the process is expensive and inefficient. At worst, breaks can happen when they are least convenient for troops on the ground. The solutions needed at this point are costly both financially and in human life.
Uptake’s pilot represents a significant improvement in the Army’s approach to maintenance. Rather than relying on time based maintenance checks, the first two-years of the pilot integrates existing vehicle monitoring technology on the Bradley into a cloud-based repository. In that repository, the data is compared to not only other Bradley maintenance patterns, but “billions of hours of operating data” Uptake has taken from other industries with like machines.[iii] For example “[Uptake][has] 230 million hours of data on diesel combustion engines, which is something the Bradley also has” and the Uptake interface is able to provide insight on individual vehicles at the individual component level. If a fault is found, then it is listed along with a fault code, a description, the severity (low, medium, critical), and the first and last occurrence of that particular fault.[iv] When fixes do need to happen, there is significantly less ambiguity about where they are, and often they can be found before there is a large problem. This makes the fix both faster and cheaper to find.
The military will look to use Uptake as a “production” technology if the pilot succeeds. This means providing the technology at a much larger scale in the DOD. In order to do so “The interface largely stays the same from industry to industry, but the machine learning models have to be changed and require a verification period before full deployment”.[v]
There is already high demand for maintenance improvement now, but the long term efficacy of the program is dependent on two factors. First, the Army is still deeply rooted in its traditional model of maintenance, and new maintainers continue to be trained in this model every day. Regardless of whether Uptake is the vendor of choice or not, AI is the future of maintenance, and so no maintenance program can truly succeed without a synchronous training in the integration of AI. Secondly, it is imperative the Army outfit its future vehicles with even better equipment tracking technology. Right now Uptake is relying on the Bradley’s old diagnostics to feed its algorithm, but more tracking nodes will allow for more precision. The opportunity is ripe as the Army recently announced the acquisition of 473 more Bradley’s, as well as the development of a completely new Bradley model.[vi]AI/ML integration will continue to be a key component of any successful military vehicle program moving forward.
2 open strategic questions remain:
- There is still a tremendous amount of tacit knowledge in maintenance- can AI ever reach a point where it absorbs all of the information needed to function autonomously?
- The system is also heavily predicated on operators trusting its insights- what are the best ways to create both world class mechanics and AI interpretation without compromising on either?
[i] Center on Budget and Policy Priorities. Where do our federal tax dollars go? 4 October 2017. Website. 9 November 2018.
[ii] Shane, Leo III. Congress finalizes $717 billion defense budget authorization months ahead of schedule. 11 August 2018. Website. 9 November 2018.
[iii] Leonard, Matt. Defense Systems. 2 July 2018. Website. 9 November 2018.
[vi] Osborn, Kris. Army makes massive Bradley buy – up to 473 vehicles to prep for major power war. 27 June 2018. Website. 9 November 2018.
Title Image & Figure 3:
U.S. Army Maneuver Center of Excellence. eArmor. 9 October 2018. Website. 9 November 2018.
Scheirer, William. BAE Systems Bradley Fighting Vehicle. 1 January 2018. Website. 9 November 2018.
Jordan, Sonja. Army investing in predictive maintenance for Bradleys. 26 September 2018. Website. 9 November 2018.