Artificial Intelligence - The Future of Munitions Readiness

By Col. Ronald "Dave" BrownOctober 21, 2019

Process Map
1 / 3 Show Caption + Hide Caption – (Photo Credit: U.S. Army) VIEW ORIGINAL
Artificial Intelligence Brief
2 / 3 Show Caption + Hide Caption – (Photo Credit: U.S. Army) VIEW ORIGINAL
Artificial Intelligence
3 / 3 Show Caption + Hide Caption – (Photo Credit: U.S. Army) VIEW ORIGINAL

ROCK ISLAND ARSENAL, Ill. - In a time of limited resources, maintaining readiness can be a challenge. When the mission is providing ready, reliable, and lethal munitions to the Joint Warfighter, rising above such challenges can be a matter of life and death. One way to make the best use of available resources is to streamline and upgrade systems that are already in place. With the goal of improving munitions readiness and enterprise synchronization, U.S. Army Joint Munitions Command (JMC) is leading the way by experimenting with the expanded application of Artificial Intelligence (AI) technology to its data systems.

At the direction of Brig. Gen. Michelle Letcher, JMC identified systems that could benefit from the application of such technology and launched two Proof of Concept (POC) tests. These have already revealed the ability to improve the quality of data and save several years' worth of man hours, achieving better accuracy and providing for a more effective use of manpower. "JMC's early efforts are promising and could provide valuable lessons for other organizations looking to streamline and improve their systems through the use of AI technology," said Letcher.

As the JMC team started developing an AI strategy, it first looked at installations where the incorporation of such technology could improve productivity. This search soon expanded to consider systems across the entire command. The team specifically looked for potential test cases in areas that might benefit from the application of two advanced analytics technologies: Machine Learning (ML) and Robotic Process Automation (RPA).

ML refers to systems that can be trained to learn from data to describe, diagnose, predict and remediate operational problems. These algorithms learn from enterprise data, creating classifiers and predictors that model the behavior of complex systems and processes. RPA refers to next-level systems that learn from users. These software robots offload manual repetitive processes from human users to replicate actions and automate tasks. This reduces the need for manual data entry, thereby reducing data entry errors and freeing up users to work on higher value tasks.

JMC understood that delivering meaningful outputs required studying the right issues; this meant choosing the right test cases. The immediate goal was to identify small, manageable projects that could provide early wins, thereby proving out the ML and RPA approaches while providing a roadmap for the future application of AI technology. The team initially considered 19 systems. Based on accessibility, JMC focused on 16 of these and interviewed 40 process owners to determine the feasibility of testing their systems. This revealed 15 potential cases, which JMC prioritized according to alignment, ease and control. The best 6 (3 ML and 3 RPA) were further prioritized based on the time required for implementation and completion. The 2 best possibilities became JMC's pilot POC test cases, one each for evaluating ML and RPA technologies. Ten other systems remain candidates for future test cases.

While JMC has been proactive in considering the application of ML and RPA technology to data systems, the team was aware that other organizations are also developing AI strategies. Government and private entities have been working on this technology for years, recognizing the effective application of AI as a critical element in the future of data analysis. As recently as May 2019, the Senate Artificial Intelligence Caucus recognized this by introducing the Artificial Intelligence Act, which proposes investing $2.2 billion to study AI technology and train the workforce in its use as part of a coordinated national strategy for using AI systems across all levels of society.

In response to the Senate proposal, Carnegie Mellon University issued a statement of support, highlighting the importance of such an act, saying, "AI has the potential to create new opportunities in communities across America and will help address significant challenges in health care, education, energy and the environment and transportation. To get there, we need an 'all of nation' approach." The commitment of elected officials, policy makers and research institutions, as well as communication between government and private entities studying the potential of AI systems, is critical to the development of a comprehensive national AI strategy. Such cooperation would spur inter-organizational sharing of methods, preventing the duplication of efforts.

With this in mind, JMC leaders reached out to other Army organizations already studying AI approaches to data management. In this way, JMC hoped to learn from the experience of others, avoid reinventing the wheel and, where possible, nest with existing systems. JMC first approached the Army Enterprise Systems Integration Portal (AESIP), which provides the Army a single authoritative source for material data supporting all Army constituent systems, old and new. Learning that AESIP was working on using AI technology to pull information from large pools of data, JMC presented its two test cases, becoming the first organization to do so.

AESIP's approach involved the creation and utilization of an Enterprise Data Lake (EDL). Organizational data is often difficult to navigate, as security concerns and bureaucratic structures lead individual users and process holders to store information in areas meant solely for their own use, creating "data silos." EDLs pull information from data silos and consolidate that data in a central location, allowing for better analysis of information. Ideally, allowing MLs and RPAs to pull data from EDLs would maximize the value of data mining.

The AESIP team did not yet have a testable system in place, so it referred JMC to the Army Leader Dashboard (ALD). Recognized by senior leaders as the Army's data integration and analysis platform, the mission of ALD is, "To organize and integrate all Army data, to make it accessible and understandable to inform strategic solutions." In 2017, ALD conducted pilot tests aimed at creating EDLs for AESIP and the General Fund Enterprise Business System (GFEBS). One lesson learned from that process was how difficult it could be to get permission for robotic systems to pull information from data silos, a problem JMC must overcome moving forward.

Before creating an EDL, JMC needed to demonstrate the value of AI technology by applying it to existing data pools. To that end, JMC first completed a test case designed to use ML algorithms to examine condition data for ammunition, including BA 30 and 40mm TNG rounds, with the goal of identifying data errors. According to Steven Taylor, Deputy Chief of Staff for Information Management, "In just three weeks, we demonstrated the ability to query ammunition condition code data to predict problems earlier than we can using traditional analysis." The ML test identified degradation that was not yet on JMC's radar, resulting in the creation of a munitions lot condition watch list. This will ensure that degraded lots are not issued. Traditional methods would not have identified this problem for another two years.

For its first RPA test, JMC looked to validate the technology by applying it to the Munitions History Program (MHP). The MHP, which houses army data for malfunctions, ammunition inspection results and ammunition test results, has an Ammunition Data Card module vulnerable to errors during manual data entry. In particular, errors appear in the form of empty fields for date of next inspection, known as "zero fill" in MHP. This issue is critical to readiness, because ammunition cannot be deployed without a clear inspection record, including the date of the next inspection.

JMC determined that this system was ripe for testing, as it had a backlog of 50,000 "zero fill" incidents, with 1,500 more added each month. This results in the storage of ammunition that is paid for and usable, but which cannot be issued until the errors are corrected. Currently, each error takes 15 minutes to find and correct, workload that requires 2.3 dedicated full time workers. This POC kicked off on June 10, 2019. Using a test data environment, the initial results proved that the RPA system works as expected (see example process map).

According to Marc Dalmasso, Chief of Systems and Sustainment, "Application of this RPA algorithm to the 'no fill' issue will plug the leak by dealing with the 1,500 monthly errors as they arise. Then we can apply the algorithm to the backlog of 50,000 errors in a series of batches, thus eliminating years of work, freeing up members of the workforce for other critical tasks while improving management of the ammunition stockpile."

The next step is for JMC to acquire permission to put the new technology in production. This is a challenge, because getting approval to access data in this way can be difficult. Effective implementation requires an RPA algorithm to run autonomously, which is only possible after the acquisition of a non-person entity (NPE) certificate. System owners who prefer to limit access to their data silos can be reluctant to approve such NPEs. To be successful, JMC will need to get buy-in from senior leaders and work closely with higher commands and 7th SIG CMD to get NPE approval. This is necessary for RPA systems to mine data without needing an attendant to log into multiple interfaces. Until NPE access is authorized, JMC will use "attended" robot software.

Moving forward, the Program Manager for ALD is prepared to help JMC leverage ALD systems for future ML deliverables. As data begins flowing into ALD, JMC will coordinate a "mini-sprint" with ALD to demonstrate what it can do for JMC's AI efforts, including the development of a cloud-based data lake. As test results come in, the effectiveness of ML and RPA will become clear and JMC will learn if AI works best on projects with a smaller scope, as opposed to those using larger data sources. Such feedback will assist in the implementation of future business process reforms, helping systems make the best use of AI technology.

The desired end-state for HQ, JMC is to implement an RPA agent able to perform simple repetitive tasks which can be completed on a scheduled basis 24x7x365, reducing errors and allowing human resources to be allocated to higher priority tasks. This will require mature understanding and use of integration at an enterprise level as well as a willingness to tackle obstacles to integration. To that end, JMC is leading the way toward a smarter approach to data analysis and more effective consumption of outputs. The goal is to improve the JMC Organic Industrial Base and better anticipate Combatant Command requirements. By applying effective AI technology solutions today, while anticipating future needs, JMC can streamline its systems to efficiently provide the Joint Warfighter with lethality that wins.