25 December 2022

Is AI the right fit for predictive military maintenance?

WALTER PINCUS

OPINION — The DoD is continually challenged to provide battle-ready ground combat systems, ships, submarines and aircraft to its warfighters, spending nearly $90 billion each year on weapon systems maintenance, according to a new Government Accountability Office (GAO) report that was sent to Congress on December 8.

Predictive maintenance defined – is a computer-driven technique used to predict the future failure of a component of a weapon or delivery system, so that the Defense Department (DoD) and military services can plan to replace components before they fail.

As the GAO report put it, “Often used in the private sector, predictive maintenance relies on personnel to use condition-monitoring technology and data analytics to schedule maintenance based on evidence of need.”

However, the report said, “The performance of maintenance depends on having a sufficient number of skilled personnel available to perform the work, parts available to use in maintenance, and a sufficient understanding of technology and the technological resources to complete maintenance.”

I picked this rather complex subject to illustrate the problem that the Pentagon – and perhaps we all face – as we transition into a computer, Internet and Artificial Intelligence (AI) dominated world.

I’m writing about the difficulty DoD has had in adopting the most modern computer-driven techniques for maintaining weapons and their delivery systems, but this situation is applicable to other elements within the military and eventually, in most people’s private lives.

According to a DoD Inspector General report released this past June 13, DoD in 2007, issued instructions which required the military services and DoD agencies to gather data about all new and current weapons system components, and fleet conditions –“when technically feasible and beneficial” — to more accurately forecast maintenance requirements, thus enabling preventive maintenance.”

Yes, both the DoD IG and the GAO conducted overlapping studies of predictive maintenance this year. The DoD IG report pointed out that over time, there have been two main categories of DoD maintenance.

Reactive maintenance is when personnel perform maintenance “for items expected to run to failure or those items that fail in an unplanned or unscheduled manner.” Proactive or conditioned-based maintenance took place based on “inspection, assessment, prognostic testing (predicting future failures), diagnostic testing (identifying current failures), servicing, and scheduled replacement or overhaul,” according to the DoD IG report.

Predictive maintenance starts with collected data on the condition of components whose algorithms lead into an AI system that forecasts the need for maintenance. Despite the 2007 instructions, DoD officials did not develop comprehensive policies or strategic plans and did not develop training tailored to the appropriate levels necessary to implement predictive maintenance strategies.

Each of the services carried out their own pilot projects. For example, the GAO reported in 2018, an Army unit maintaining Medium Tactical Vehicles in Hawaii used predictive maintenance techniques to adjust hours for tasks such as oil changes and saved 6,100 hours of labor. The Army exempted the unit from personnel reductions to allow maintenance personnel to address other needs, according to Army officials.

However, the GAO also found, “Officials from all four military services stated that maintenance personnel are sometimes reluctant to complete predictive maintenance due in part to skepticism about the validity of predictions. For example, according to Navy officials, during the Navy’s first demonstration of predictive maintenance aboard a ship, ship’s crew did not take action to address predictive maintenance prompts due to a variety of issues that caused skepticism among end users, such as incorrect algorithms. Officials stated a second demonstration aboard the same ship combined corrections to algorithms with coaching on predictive maintenance that resulted in a higher rate of action as the ship’s crew began to understand the validity of the concept. Army, Marine Corps, Navy, and Air Force officials also said that overburdened personnel are hesitant to conduct maintenance on something that is not broken yet, or to change parameters of preventive maintenance without understanding why.”

While the military services are not close to fully implementing predictive maintenance, they have been replacing parts periodically based on forecasts, using software and health monitoring tools to improve condition-based maintenance (preventive), or improving predictive forecasts with pilot programs, the DoD IG found.

The military services are taking steps to adapt sensors and analytics to current weapon systems as a necessary step to integrate predictive maintenance with preventive maintenance

For example, according to the DoD IG report, the Air Force has two dashboards tracking maintenance data for 16 weapon systems that feed into the enhanced Reliability-Centered Maintenance dashboard to make predictive maintenance forecast decisions and produce monthly reports of how many parts or components personnel removed due to those forecasts.

An Army official stated that the UH-60 Black Hawk platform has been “equipped with software that understands what the maintenance thresholds are and identifies whether a part is starting to see a negative trend so maintainers can replace that component,” according to the DoD IG.

A Naval Air Systems Command official told the DoD IG staff that the F/A-18 Hornet uses a tool that identifies components experiencing degradation and flags the part before failure, allowing maintainers to preventatively replace a part before the part failed. Navy officials also said that Naval Sea Systems Command has a digital modeling concept to identify potential problems with Navy maintainers to reduce the amount of manual inspections and saved hundreds of labor hours.

Marine Corps officials told DoD staff they are working to add data sensors on the Medium Tactical Vehicle replacement and Joint Light Tactical Vehicle and have worked with the contractor to validate predictive maintenance and historical data analysis to use lessons learned from these efforts.

Both the DoD IG and GAO found the military services are developing, or have developed, training to support service-wide implementation of predictive maintenance. The GAO people observed Army personnel getting basic instructions on how to transfer data from ground combat systems and aircraft and how to use predictive maintenance dashboards. The Marine Corps and Navy also showed they had developed training at varying levels and comprehensiveness to their sustainment workforces.

The GAO report found, “The military services have not consistently adopted and tracked implementation of predictive maintenance,” and would be in a better position “to determine how, when, and where to adopt predictive maintenance” if they developed “action plans and milestones for current weapon systems.”

The military services, according to the GAO, also need “outcome-related objectives for predictive maintenance, a process for evaluating progress, and a framework to develop and track milestones of implementation…[Then] the military services will be better positioned to gauge progress and results to inform decision makers about the changes being made to support increased readiness.”

In the end, the GAO said, “We agree that not all weapon systems may be suitable candidates for predictive maintenance, and that deliberate study and analysis will help determine which weapon systems should implement predictive maintenance.”

The lesson we can all take from this is that computers, the Internet and AI may be the future, but not for everything and not without training, testing and analyzing of results to see where they fit in.

No comments: